13559 lines
406 KiB
JSON
13559 lines
406 KiB
JSON
{
|
|
"policy": "beta5-candidate-a-v2-int8-convrot",
|
|
"source": "C:\\comfycli\\models\\diffusion_models\\Minimax-h3_Singularity_ref2va_v1.3_pruned_bf16.safetensors",
|
|
"source_sha256": "c73f5ddbd3b13a0dc523e648bc6fbf38c8d2dfe6b6ede6620df59f1a9608e1f4",
|
|
"output": "C:\\comfycli\\models\\diffusion_models\\Minimax-h3_Singularity_ref2va_v1.3_pruned_w4a8_graft_preserving.safetensors",
|
|
"seed": 1337,
|
|
"device": "cuda",
|
|
"gates": {
|
|
"max_relative_error": 0.13,
|
|
"max_p99_cosine_loss": 0.035,
|
|
"max_cosine_loss": 0.15,
|
|
"max_p99_norm_drift": 0.01,
|
|
"max_norm_drift": 0.03,
|
|
"min_correction": 0.5,
|
|
"max_correction": 2.0,
|
|
"max_extreme_fraction": 0.2,
|
|
"min_code_utilization": 0.75
|
|
},
|
|
"counts_by_format": {
|
|
"bf16": 350,
|
|
"asym_w4a8_int8": 64,
|
|
"int8_tensorwise": 120
|
|
},
|
|
"bytes_by_format": {
|
|
"bf16": 4828202032,
|
|
"asym_w4a8_int8": 4253987008,
|
|
"int8_tensorwise": 10179026792
|
|
},
|
|
"layers": [
|
|
{
|
|
"name": "model.diffusion_model.adaln_basis",
|
|
"shape": [
|
|
8,
|
|
2688
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 43008,
|
|
"output_bytes": 43008,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.26600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.adaln_mean",
|
|
"shape": [
|
|
2688
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 5376,
|
|
"output_bytes": 5376,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.adaln_t_table",
|
|
"shape": [
|
|
1025,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 16400,
|
|
"output_bytes": 16400,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.audio_patch_proj.bias",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.audio_patch_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
32
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 344064,
|
|
"output_bytes": 344064,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.001244803424924612,
|
|
"metrics": {
|
|
"relative_error": 0.07333993911743164,
|
|
"max_absolute_error": 0.10400390625,
|
|
"cosine_loss": {
|
|
"mean": 0.002688098233193159,
|
|
"p95": 0.0027718693017959595,
|
|
"p99": 0.002816513180732727,
|
|
"max": 0.0031262636184692383
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00038123034755699337,
|
|
"p95": 0.000994265079498291,
|
|
"p99": 0.0016849488019943237,
|
|
"max": 0.0038542747497558594
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0035197848919779062,
|
|
"p95": 0.006173640489578247,
|
|
"p99": 0.0074761658906936646,
|
|
"max": 0.01065981388092041
|
|
},
|
|
"correction_min": 0.9912909865379333,
|
|
"correction_max": 1.0107746124267578,
|
|
"codebook": [
|
|
-0.9781808257102966,
|
|
-0.782500684261322,
|
|
-0.6233840584754944,
|
|
-0.4859867990016937,
|
|
-0.3651864528656006,
|
|
-0.25551119446754456,
|
|
-0.15578223764896393,
|
|
-0.06017342209815979,
|
|
0.03794879838824272,
|
|
0.1388709992170334,
|
|
0.24394448101520538,
|
|
0.3559195399284363,
|
|
0.4800650477409363,
|
|
0.6199949383735657,
|
|
0.7826771140098572,
|
|
0.9778964519500732
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9600586891174316,
|
|
"extreme_code_fraction": 0.09571516513824463,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 10.171999999991385
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020304506178945303,
|
|
"max_absolute_error": 0.006529331207275391,
|
|
"cosine_loss": {
|
|
"mean": 1.7581874089955818e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9490718841552734e-05,
|
|
"max": 2.6881694793701172e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043454766273498535,
|
|
"p95": 0.0009828953770920634,
|
|
"p99": 0.003957594744861126,
|
|
"max": 0.006437718868255615
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.625
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012498691212385893,
|
|
"metrics": {
|
|
"relative_error": 0.07311022281646729,
|
|
"max_absolute_error": 0.0972900390625,
|
|
"cosine_loss": {
|
|
"mean": 0.002670143498107791,
|
|
"p95": 0.0027423501014709473,
|
|
"p99": 0.0027718544006347656,
|
|
"max": 0.0028432607650756836
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003935884451493621,
|
|
"p95": 0.001000371528789401,
|
|
"p99": 0.0016133980825543404,
|
|
"max": 0.003739774227142334
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0030887010507285595,
|
|
"p95": 0.005267665721476078,
|
|
"p99": 0.006213247776031494,
|
|
"max": 0.00874251127243042
|
|
},
|
|
"correction_min": 0.9973690509796143,
|
|
"correction_max": 1.008819580078125,
|
|
"codebook": [
|
|
-0.9771839380264282,
|
|
-0.7786573171615601,
|
|
-0.6144593954086304,
|
|
-0.4750567078590393,
|
|
-0.35179296135902405,
|
|
-0.24038851261138916,
|
|
-0.13829182088375092,
|
|
-0.040417879819869995,
|
|
0.05532420054078102,
|
|
0.15195399522781372,
|
|
0.25301477313041687,
|
|
0.36246028542518616,
|
|
0.48404228687286377,
|
|
0.6209474205970764,
|
|
0.7812997102737427,
|
|
0.9773151874542236
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.963273525238037,
|
|
"extreme_code_fraction": 0.09591692686080933,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.6089999999967404
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020764374639838934,
|
|
"max_absolute_error": 0.009232282638549805,
|
|
"cosine_loss": {
|
|
"mean": 1.7853897134045837e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8909573554992676e-05,
|
|
"max": 2.9265880584716797e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004455712914932519,
|
|
"p95": 0.0009689778089523315,
|
|
"p99": 0.004653260111808777,
|
|
"max": 0.006302177906036377
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.0.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012539793970063329,
|
|
"metrics": {
|
|
"relative_error": 0.07340725511312485,
|
|
"max_absolute_error": 0.12109375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026938130613416433,
|
|
"p95": 0.002787947654724121,
|
|
"p99": 0.0028513967990875244,
|
|
"max": 0.003067910671234131
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003579893964342773,
|
|
"p95": 0.0009673237800598145,
|
|
"p99": 0.0016648173332214355,
|
|
"max": 0.0034208297729492188
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003516407683491707,
|
|
"p95": 0.006228134036064148,
|
|
"p99": 0.007754027843475342,
|
|
"max": 0.012950122356414795
|
|
},
|
|
"correction_min": 0.994432806968689,
|
|
"correction_max": 1.0131200551986694,
|
|
"codebook": [
|
|
-0.9775971174240112,
|
|
-0.781457245349884,
|
|
-0.6194100975990295,
|
|
-0.4825979471206665,
|
|
-0.36273646354675293,
|
|
-0.2546025216579437,
|
|
-0.15443125367164612,
|
|
-0.05944608151912689,
|
|
0.03700382634997368,
|
|
0.13625235855579376,
|
|
0.23964624106884003,
|
|
0.3513791859149933,
|
|
0.47511589527130127,
|
|
0.6145520806312561,
|
|
0.7789744734764099,
|
|
0.977308988571167
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.962838649749756,
|
|
"extreme_code_fraction": 0.09600907564163208,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001991287572309375,
|
|
"max_absolute_error": 0.009242057800292969,
|
|
"cosine_loss": {
|
|
"mean": 1.698945879979874e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8894672393798828e-05,
|
|
"max": 2.5510787963867188e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004326818452682346,
|
|
"p95": 0.0009746461873874068,
|
|
"p99": 0.003853871952742338,
|
|
"max": 0.006222963333129883
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5309999999881256
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012473942479118705,
|
|
"metrics": {
|
|
"relative_error": 0.07305016368627548,
|
|
"max_absolute_error": 0.11328125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026657136622816324,
|
|
"p95": 0.0027388930320739746,
|
|
"p99": 0.0027702623046934605,
|
|
"max": 0.002852022647857666
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00041542903636582196,
|
|
"p95": 0.001059889793395996,
|
|
"p99": 0.001667215721681714,
|
|
"max": 0.00352400541305542
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031091796699911356,
|
|
"p95": 0.005354159977287054,
|
|
"p99": 0.006339364685118198,
|
|
"max": 0.008518755435943604
|
|
},
|
|
"correction_min": 0.9972130060195923,
|
|
"correction_max": 1.008591890335083,
|
|
"codebook": [
|
|
-0.9777096509933472,
|
|
-0.7802273035049438,
|
|
-0.6189297437667847,
|
|
-0.4794270694255829,
|
|
-0.35548487305641174,
|
|
-0.24441775679588318,
|
|
-0.14025390148162842,
|
|
-0.041160572320222855,
|
|
0.055459052324295044,
|
|
0.15206541121006012,
|
|
0.25353482365608215,
|
|
0.36307746171951294,
|
|
0.4837256669998169,
|
|
0.6218634247779846,
|
|
0.7841069102287292,
|
|
0.9781795740127563
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9611730575561523,
|
|
"extreme_code_fraction": 0.09559578448534012,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.45300000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002033880213275552,
|
|
"max_absolute_error": 0.010421276092529297,
|
|
"cosine_loss": {
|
|
"mean": 1.7106289078583359e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7777085304260254e-05,
|
|
"max": 2.6345252990722656e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004380510945338756,
|
|
"p95": 0.0009437650442123413,
|
|
"p99": 0.004498600959777832,
|
|
"max": 0.0065421462059021
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3129999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.1.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.001242498285137117,
|
|
"metrics": {
|
|
"relative_error": 0.07308200001716614,
|
|
"max_absolute_error": 0.14703369140625,
|
|
"cosine_loss": {
|
|
"mean": 0.002669881796464324,
|
|
"p95": 0.002740815281867981,
|
|
"p99": 0.002771347761154175,
|
|
"max": 0.00299072265625
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003684363909997046,
|
|
"p95": 0.0009635090827941895,
|
|
"p99": 0.001656860113143921,
|
|
"max": 0.003377854824066162
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0030906510073691607,
|
|
"p95": 0.005269333720207214,
|
|
"p99": 0.00634273886680603,
|
|
"max": 0.00953298807144165
|
|
},
|
|
"correction_min": 0.9981482625007629,
|
|
"correction_max": 1.009624719619751,
|
|
"codebook": [
|
|
-0.9784293174743652,
|
|
-0.7823625802993774,
|
|
-0.6191952228546143,
|
|
-0.47881433367729187,
|
|
-0.35519495606422424,
|
|
-0.24318872392177582,
|
|
-0.13966986536979675,
|
|
-0.04019626975059509,
|
|
0.05644805356860161,
|
|
0.15161067247390747,
|
|
0.25180357694625854,
|
|
0.36020490527153015,
|
|
0.48055893182754517,
|
|
0.6182202696800232,
|
|
0.7790743708610535,
|
|
0.9773420691490173
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961660385131836,
|
|
"extreme_code_fraction": 0.0958864688873291,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.09299999999348074
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002020264510065317,
|
|
"max_absolute_error": 0.01045680046081543,
|
|
"cosine_loss": {
|
|
"mean": 1.752169850988139e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9371509552001953e-05,
|
|
"max": 2.8073787689208984e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043284360435791314,
|
|
"p95": 0.0009753224439918995,
|
|
"p99": 0.0039848266169428825,
|
|
"max": 0.0064375996589660645
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5160000000032596
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001242282916791737,
|
|
"metrics": {
|
|
"relative_error": 0.07312680035829544,
|
|
"max_absolute_error": 0.098876953125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026736108120530844,
|
|
"p95": 0.002746105194091797,
|
|
"p99": 0.0027757883071899414,
|
|
"max": 0.002841770648956299
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004192636697553098,
|
|
"p95": 0.0010715066455304623,
|
|
"p99": 0.0016450383700430393,
|
|
"max": 0.003611266613006592
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031042250338941813,
|
|
"p95": 0.005314244888722897,
|
|
"p99": 0.006241100374609232,
|
|
"max": 0.008564114570617676
|
|
},
|
|
"correction_min": 0.9965611100196838,
|
|
"correction_max": 1.0086381435394287,
|
|
"codebook": [
|
|
-0.9788429141044617,
|
|
-0.7840204834938049,
|
|
-0.6210834980010986,
|
|
-0.47974693775177,
|
|
-0.3559308648109436,
|
|
-0.24446061253547668,
|
|
-0.14096413552761078,
|
|
-0.04169120267033577,
|
|
0.05533875897526741,
|
|
0.15278911590576172,
|
|
0.254069983959198,
|
|
0.36268025636672974,
|
|
0.4825769364833832,
|
|
0.6200460195541382,
|
|
0.7815539836883545,
|
|
0.977949857711792
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9603800773620605,
|
|
"extreme_code_fraction": 0.0953553169965744,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.46900000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019901946652680635,
|
|
"max_absolute_error": 0.008242011070251465,
|
|
"cosine_loss": {
|
|
"mean": 1.6863813243617187e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.753866672515869e-05,
|
|
"max": 2.5033950805664062e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004241076239850372,
|
|
"p95": 0.0009418874979019165,
|
|
"p99": 0.00443682074546814,
|
|
"max": 0.006136715412139893
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.29699999999138527
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.10.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012467429041862488,
|
|
"metrics": {
|
|
"relative_error": 0.07310479134321213,
|
|
"max_absolute_error": 0.119140625,
|
|
"cosine_loss": {
|
|
"mean": 0.0026729917153716087,
|
|
"p95": 0.0027413517236709595,
|
|
"p99": 0.002782955765724182,
|
|
"max": 0.0030121803283691406
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00036935642128810287,
|
|
"p95": 0.0009697675704956055,
|
|
"p99": 0.0016355514526367188,
|
|
"max": 0.0031069517135620117
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032767027150839567,
|
|
"p95": 0.005611643195152283,
|
|
"p99": 0.007039099931716919,
|
|
"max": 0.010638058185577393
|
|
},
|
|
"correction_min": 0.9979181885719299,
|
|
"correction_max": 1.0107524394989014,
|
|
"codebook": [
|
|
-0.9776701927185059,
|
|
-0.7811732888221741,
|
|
-0.618779182434082,
|
|
-0.47832775115966797,
|
|
-0.3552306592464447,
|
|
-0.2431800365447998,
|
|
-0.139456108212471,
|
|
-0.04019813984632492,
|
|
0.05654064193367958,
|
|
0.15183445811271667,
|
|
0.2529996633529663,
|
|
0.3621235489845276,
|
|
0.4828890264034271,
|
|
0.6214597821235657,
|
|
0.7811633348464966,
|
|
0.9778269529342651
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9613091945648193,
|
|
"extreme_code_fraction": 0.09594409912824631,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.10999999998603016
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019921185448765755,
|
|
"max_absolute_error": 0.006544157862663269,
|
|
"cosine_loss": {
|
|
"mean": 1.6996636986732483e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.895427703857422e-05,
|
|
"max": 2.562999725341797e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004220039118081331,
|
|
"p95": 0.0009727298747748137,
|
|
"p99": 0.0036973506212234497,
|
|
"max": 0.006764233112335205
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012495151022449136,
|
|
"metrics": {
|
|
"relative_error": 0.07300650328397751,
|
|
"max_absolute_error": 0.0986328125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026646764017641544,
|
|
"p95": 0.002736985683441162,
|
|
"p99": 0.0027666687965393066,
|
|
"max": 0.002831757068634033
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003851519722957164,
|
|
"p95": 0.0009961663745343685,
|
|
"p99": 0.0016400297172367573,
|
|
"max": 0.0034838318824768066
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003186357906088233,
|
|
"p95": 0.0053767976351082325,
|
|
"p99": 0.006336585618555546,
|
|
"max": 0.009023427963256836
|
|
},
|
|
"correction_min": 0.9973103404045105,
|
|
"correction_max": 1.0091055631637573,
|
|
"codebook": [
|
|
-0.9775742888450623,
|
|
-0.7813267707824707,
|
|
-0.6186107993125916,
|
|
-0.4810049831867218,
|
|
-0.3601757884025574,
|
|
-0.2503643333911896,
|
|
-0.14952151477336884,
|
|
-0.05405949056148529,
|
|
0.04189155623316765,
|
|
0.14027564227581024,
|
|
0.24386917054653168,
|
|
0.3548901379108429,
|
|
0.4781171381473541,
|
|
0.6172211170196533,
|
|
0.7804140448570251,
|
|
0.9777239561080933
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.962014675140381,
|
|
"extreme_code_fraction": 0.09563343226909637,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.5150000000139698
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.00207072077319026,
|
|
"max_absolute_error": 0.00945138931274414,
|
|
"cosine_loss": {
|
|
"mean": 1.7795467783798813e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8864870071411133e-05,
|
|
"max": 2.855062484741211e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004460651252884418,
|
|
"p95": 0.0009689629077911377,
|
|
"p99": 0.004626840353012085,
|
|
"max": 0.006304621696472168
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5309999999881256
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.11.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012501274468377233,
|
|
"metrics": {
|
|
"relative_error": 0.07309050858020782,
|
|
"max_absolute_error": 0.1326904296875,
|
|
"cosine_loss": {
|
|
"mean": 0.002670672256499529,
|
|
"p95": 0.0027390271425247192,
|
|
"p99": 0.002773478627204895,
|
|
"max": 0.0029945969581604004
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003775630611926317,
|
|
"p95": 0.0009788572788238525,
|
|
"p99": 0.0015805363655090332,
|
|
"max": 0.0032261013984680176
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.002938838442787528,
|
|
"p95": 0.005124807357788086,
|
|
"p99": 0.00622183084487915,
|
|
"max": 0.009140431880950928
|
|
},
|
|
"correction_min": 0.9967654347419739,
|
|
"correction_max": 1.009224772453308,
|
|
"codebook": [
|
|
-0.97752845287323,
|
|
-0.781823456287384,
|
|
-0.6193413734436035,
|
|
-0.4795207381248474,
|
|
-0.355741024017334,
|
|
-0.2439635843038559,
|
|
-0.1400844156742096,
|
|
-0.04047439992427826,
|
|
0.05574624985456467,
|
|
0.15061470866203308,
|
|
0.2513759732246399,
|
|
0.3603579103946686,
|
|
0.4818625748157501,
|
|
0.6189157366752625,
|
|
0.7794424891471863,
|
|
0.9774547219276428
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961245059967041,
|
|
"extreme_code_fraction": 0.0958656519651413,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0159999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020116129890084267,
|
|
"max_absolute_error": 0.006899237632751465,
|
|
"cosine_loss": {
|
|
"mean": 1.733407543724752e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.931015867739916e-05,
|
|
"max": 2.6702880859375e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042788495193235576,
|
|
"p95": 0.0009734926279634237,
|
|
"p99": 0.0038922156672924757,
|
|
"max": 0.006502926349639893
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5310000000172295
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001250534551218152,
|
|
"metrics": {
|
|
"relative_error": 0.07307212054729462,
|
|
"max_absolute_error": 0.08929443359375,
|
|
"cosine_loss": {
|
|
"mean": 0.002669368637725711,
|
|
"p95": 0.002741098403930664,
|
|
"p99": 0.0027704585809260607,
|
|
"max": 0.002871990203857422
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003981050685979426,
|
|
"p95": 0.001021694391965866,
|
|
"p99": 0.0016148437280207872,
|
|
"max": 0.003493189811706543
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032833670265972614,
|
|
"p95": 0.005451646633446217,
|
|
"p99": 0.006373908370733261,
|
|
"max": 0.008414208889007568
|
|
},
|
|
"correction_min": 0.9960657954216003,
|
|
"correction_max": 1.0084855556488037,
|
|
"codebook": [
|
|
-0.9785934090614319,
|
|
-0.7837721705436707,
|
|
-0.6218938231468201,
|
|
-0.48331597447395325,
|
|
-0.36056530475616455,
|
|
-0.2513148784637451,
|
|
-0.15021704137325287,
|
|
-0.05433487892150879,
|
|
0.041156645864248276,
|
|
0.14042259752750397,
|
|
0.24376802146434784,
|
|
0.3540858030319214,
|
|
0.4770125448703766,
|
|
0.6161900758743286,
|
|
0.7800747156143188,
|
|
0.977480411529541
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9618277549743652,
|
|
"extreme_code_fraction": 0.09556522220373154,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.48399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002008908661082387,
|
|
"max_absolute_error": 0.009792208671569824,
|
|
"cosine_loss": {
|
|
"mean": 1.6997635157167679e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7911195755004883e-05,
|
|
"max": 2.574920654296875e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042294684681110084,
|
|
"p95": 0.0009382367134094238,
|
|
"p99": 0.004433467984199524,
|
|
"max": 0.0063187479972839355
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.35999999998603016
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.12.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.001251488458365202,
|
|
"metrics": {
|
|
"relative_error": 0.07316353172063828,
|
|
"max_absolute_error": 0.1395263671875,
|
|
"cosine_loss": {
|
|
"mean": 0.002675688127055764,
|
|
"p95": 0.0027439743280410767,
|
|
"p99": 0.00278647243976593,
|
|
"max": 0.003142833709716797
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003680155787151307,
|
|
"p95": 0.0009369701147079468,
|
|
"p99": 0.001551896333694458,
|
|
"max": 0.0028165578842163086
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031716220546513796,
|
|
"p95": 0.005440220236778259,
|
|
"p99": 0.006627902388572693,
|
|
"max": 0.010525882244110107
|
|
},
|
|
"correction_min": 0.9977091550827026,
|
|
"correction_max": 1.010637879371643,
|
|
"codebook": [
|
|
-0.9774544835090637,
|
|
-0.7802976965904236,
|
|
-0.6196171641349792,
|
|
-0.4821310341358185,
|
|
-0.36147281527519226,
|
|
-0.25308236479759216,
|
|
-0.15187254548072815,
|
|
-0.05669642984867096,
|
|
0.039405643939971924,
|
|
0.13696736097335815,
|
|
0.2391573190689087,
|
|
0.3501948416233063,
|
|
0.47257527709007263,
|
|
0.6121187806129456,
|
|
0.7769340872764587,
|
|
0.9771581292152405
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.964081048965454,
|
|
"extreme_code_fraction": 0.09631023555994034,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.07899999999790452
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020099845714867115,
|
|
"max_absolute_error": 0.008899569511413574,
|
|
"cosine_loss": {
|
|
"mean": 1.7381805719196564e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9490718841552734e-05,
|
|
"max": 2.7060508728027344e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042651526746340096,
|
|
"p95": 0.0009618997573852539,
|
|
"p99": 0.003906559199094772,
|
|
"max": 0.007296621799468994
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.6719999999913853
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012470604851841927,
|
|
"metrics": {
|
|
"relative_error": 0.07303886860609055,
|
|
"max_absolute_error": 0.18798828125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026669842191040516,
|
|
"p95": 0.0027381449472159147,
|
|
"p99": 0.002767758909612894,
|
|
"max": 0.0028380155563354492
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003901764575857669,
|
|
"p95": 0.0010146766435354948,
|
|
"p99": 0.0016471605049446225,
|
|
"max": 0.003560960292816162
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003306081984192133,
|
|
"p95": 0.005485137924551964,
|
|
"p99": 0.006460738368332386,
|
|
"max": 0.00932377576828003
|
|
},
|
|
"correction_min": 0.9971664547920227,
|
|
"correction_max": 1.0094115734100342,
|
|
"codebook": [
|
|
-0.9781873226165771,
|
|
-0.7809696197509766,
|
|
-0.6175658702850342,
|
|
-0.476555734872818,
|
|
-0.353920578956604,
|
|
-0.24286001920700073,
|
|
-0.14049503207206726,
|
|
-0.04136274382472038,
|
|
0.05487225204706192,
|
|
0.15118180215358734,
|
|
0.2525712251663208,
|
|
0.36205747723579407,
|
|
0.483178973197937,
|
|
0.6204369068145752,
|
|
0.7808820605278015,
|
|
0.9779331684112549
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9621734619140625,
|
|
"extreme_code_fraction": 0.09576839208602905,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.5160000000032596
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001992687350139022,
|
|
"max_absolute_error": 0.0524749755859375,
|
|
"cosine_loss": {
|
|
"mean": 1.686215000518132e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7762184143066406e-05,
|
|
"max": 2.473592758178711e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004168023297097534,
|
|
"p95": 0.0009166896343231201,
|
|
"p99": 0.004493847489356995,
|
|
"max": 0.006108999252319336
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3120000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.13.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0159999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012473929673433304,
|
|
"metrics": {
|
|
"relative_error": 0.07310152053833008,
|
|
"max_absolute_error": 0.1279296875,
|
|
"cosine_loss": {
|
|
"mean": 0.002672035014256835,
|
|
"p95": 0.0027447640895843506,
|
|
"p99": 0.002779737114906311,
|
|
"max": 0.0031083226203918457
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00037799763958901167,
|
|
"p95": 0.0009820014238357544,
|
|
"p99": 0.001643359661102295,
|
|
"max": 0.00329667329788208
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032753022387623787,
|
|
"p95": 0.005480825901031494,
|
|
"p99": 0.006889224052429199,
|
|
"max": 0.010572433471679688
|
|
},
|
|
"correction_min": 0.9971933960914612,
|
|
"correction_max": 1.0106854438781738,
|
|
"codebook": [
|
|
-0.9779297113418579,
|
|
-0.778724193572998,
|
|
-0.6150991916656494,
|
|
-0.47555792331695557,
|
|
-0.35308051109313965,
|
|
-0.24183154106140137,
|
|
-0.13882197439670563,
|
|
-0.039306655526161194,
|
|
0.0560217909514904,
|
|
0.15119558572769165,
|
|
0.2516520023345947,
|
|
0.36042702198028564,
|
|
0.4812542498111725,
|
|
0.6207666397094727,
|
|
0.7822993993759155,
|
|
0.9784083962440491
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9628043174743652,
|
|
"extreme_code_fraction": 0.09603386372327805,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019883604254573584,
|
|
"max_absolute_error": 0.007346987724304199,
|
|
"cosine_loss": {
|
|
"mean": 1.6983083241939312e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8714112229645252e-05,
|
|
"max": 2.682209014892578e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004204170254524797,
|
|
"p95": 0.0009537218138575554,
|
|
"p99": 0.0038563450798392296,
|
|
"max": 0.006922364234924316
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5620000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012508513173088431,
|
|
"metrics": {
|
|
"relative_error": 0.07303076237440109,
|
|
"max_absolute_error": 0.10546875,
|
|
"cosine_loss": {
|
|
"mean": 0.0026664086617529392,
|
|
"p95": 0.0027385354042053223,
|
|
"p99": 0.002769249025732279,
|
|
"max": 0.002866208553314209
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003814221709035337,
|
|
"p95": 0.0009932785760611296,
|
|
"p99": 0.0016132615273818374,
|
|
"max": 0.0035634636878967285
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032437562476843596,
|
|
"p95": 0.005418673157691956,
|
|
"p99": 0.006387337110936642,
|
|
"max": 0.008859515190124512
|
|
},
|
|
"correction_min": 0.9977355003356934,
|
|
"correction_max": 1.0089386701583862,
|
|
"codebook": [
|
|
-0.9778354167938232,
|
|
-0.7787066102027893,
|
|
-0.6157930493354797,
|
|
-0.476812481880188,
|
|
-0.3544365465641022,
|
|
-0.24255920946598053,
|
|
-0.13910767436027527,
|
|
-0.04105423018336296,
|
|
0.05446354299783707,
|
|
0.15029285848140717,
|
|
0.25135621428489685,
|
|
0.36047711968421936,
|
|
0.48072370886802673,
|
|
0.6184316277503967,
|
|
0.779494047164917,
|
|
0.9773350954055786
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9626681804656982,
|
|
"extreme_code_fraction": 0.09601429104804993,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.4379999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020204465836286545,
|
|
"max_absolute_error": 0.007839679718017578,
|
|
"cosine_loss": {
|
|
"mean": 1.7048636209437973e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8134713172912598e-05,
|
|
"max": 2.5212764739990234e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004240304697304964,
|
|
"p95": 0.0009344965219497681,
|
|
"p99": 0.004492074251174927,
|
|
"max": 0.006403982639312744
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.14.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012470567598938942,
|
|
"metrics": {
|
|
"relative_error": 0.07316920906305313,
|
|
"max_absolute_error": 0.147705078125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026776136364787817,
|
|
"p95": 0.0027504712343215942,
|
|
"p99": 0.0028213560581207275,
|
|
"max": 0.003237009048461914
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00037381151923909783,
|
|
"p95": 0.0009397715330123901,
|
|
"p99": 0.0016309022903442383,
|
|
"max": 0.0035163164138793945
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032574280630797148,
|
|
"p95": 0.0055668652057647705,
|
|
"p99": 0.006941437721252441,
|
|
"max": 0.014011144638061523
|
|
},
|
|
"correction_min": 0.9861825108528137,
|
|
"correction_max": 1.0096632242202759,
|
|
"codebook": [
|
|
-0.9773112535476685,
|
|
-0.7793248295783997,
|
|
-0.6170827746391296,
|
|
-0.47794193029403687,
|
|
-0.35434049367904663,
|
|
-0.24220328032970428,
|
|
-0.13868241012096405,
|
|
-0.03915394842624664,
|
|
0.056868020445108414,
|
|
0.15331031382083893,
|
|
0.2542584240436554,
|
|
0.3626466989517212,
|
|
0.4835317134857178,
|
|
0.6241016387939453,
|
|
0.7855135798454285,
|
|
0.9781840443611145
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9613234996795654,
|
|
"extreme_code_fraction": 0.09553208947181702,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.10899999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002001227578148246,
|
|
"max_absolute_error": 0.006946474313735962,
|
|
"cosine_loss": {
|
|
"mean": 1.7133424989879131e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.901388168334961e-05,
|
|
"max": 2.866983413696289e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004236939421389252,
|
|
"p95": 0.0009576315060257912,
|
|
"p99": 0.0038896193727850914,
|
|
"max": 0.006603598594665527
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5469999999913853
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012460905127227306,
|
|
"metrics": {
|
|
"relative_error": 0.07304420322179794,
|
|
"max_absolute_error": 0.1185302734375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026674356777220964,
|
|
"p95": 0.0027406811714172363,
|
|
"p99": 0.0027719312347471714,
|
|
"max": 0.0028562545776367188
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003940349561162293,
|
|
"p95": 0.0010043676011264324,
|
|
"p99": 0.001610094914212823,
|
|
"max": 0.003499448299407959
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003006667597219348,
|
|
"p95": 0.005190989002585411,
|
|
"p99": 0.006229549180716276,
|
|
"max": 0.009110093116760254
|
|
},
|
|
"correction_min": 0.9974376559257507,
|
|
"correction_max": 1.0091938972473145,
|
|
"codebook": [
|
|
-0.9774013161659241,
|
|
-0.779917299747467,
|
|
-0.6195953488349915,
|
|
-0.48110535740852356,
|
|
-0.36020711064338684,
|
|
-0.25047388672828674,
|
|
-0.14957427978515625,
|
|
-0.05394934490323067,
|
|
0.042049482464790344,
|
|
0.14225314557552338,
|
|
0.24588792026042938,
|
|
0.3571067452430725,
|
|
0.47919800877571106,
|
|
0.6201030015945435,
|
|
0.7839480042457581,
|
|
0.9782183766365051
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9606971740722656,
|
|
"extreme_code_fraction": 0.09558678418397903,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.5160000000032596
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002016552723944187,
|
|
"max_absolute_error": 0.00784921646118164,
|
|
"cosine_loss": {
|
|
"mean": 1.717070631457318e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8894672393798828e-05,
|
|
"max": 2.5510787963867188e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042395349009893835,
|
|
"p95": 0.0009297728538513184,
|
|
"p99": 0.004399627447128296,
|
|
"max": 0.006489932537078857
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.29699999999138527
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.15.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012501849560067058,
|
|
"metrics": {
|
|
"relative_error": 0.07313832640647888,
|
|
"max_absolute_error": 0.083984375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026745230425149202,
|
|
"p95": 0.002744913101196289,
|
|
"p99": 0.002779260277748108,
|
|
"max": 0.003042161464691162
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00038003685767762363,
|
|
"p95": 0.000990256667137146,
|
|
"p99": 0.0016626715660095215,
|
|
"max": 0.003084540367126465
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0030626768711954355,
|
|
"p95": 0.005208134651184082,
|
|
"p99": 0.006397351622581482,
|
|
"max": 0.009221017360687256
|
|
},
|
|
"correction_min": 0.9977101683616638,
|
|
"correction_max": 1.009306788444519,
|
|
"codebook": [
|
|
-0.9782308340072632,
|
|
-0.7817301750183105,
|
|
-0.6194829344749451,
|
|
-0.4807693064212799,
|
|
-0.3604210913181305,
|
|
-0.25175201892852783,
|
|
-0.15156155824661255,
|
|
-0.05620460957288742,
|
|
0.039736490696668625,
|
|
0.13845272362232208,
|
|
0.24236690998077393,
|
|
0.3543306887149811,
|
|
0.47805920243263245,
|
|
0.6188528537750244,
|
|
0.7828078866004944,
|
|
0.9780296087265015
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9615774154663086,
|
|
"extreme_code_fraction": 0.09558224678039551,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.10999999998603016
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019987639971077442,
|
|
"max_absolute_error": 0.006994485855102539,
|
|
"cosine_loss": {
|
|
"mean": 1.7135864709416637e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.901388168334961e-05,
|
|
"max": 2.5987625122070312e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042383073014207184,
|
|
"p95": 0.0009635001188144088,
|
|
"p99": 0.0037483801133930683,
|
|
"max": 0.006652414798736572
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5310000000172295
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012462000595405698,
|
|
"metrics": {
|
|
"relative_error": 0.0731026902794838,
|
|
"max_absolute_error": 0.1015625,
|
|
"cosine_loss": {
|
|
"mean": 0.0026716208085417747,
|
|
"p95": 0.002743065357208252,
|
|
"p99": 0.002774059772491455,
|
|
"max": 0.00284731388092041
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00039459249819628894,
|
|
"p95": 0.0010203865822404623,
|
|
"p99": 0.001628237427212298,
|
|
"max": 0.0034230947494506836
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031069444958120584,
|
|
"p95": 0.005323556251823902,
|
|
"p99": 0.006274878978729248,
|
|
"max": 0.009039819240570068
|
|
},
|
|
"correction_min": 0.9972710609436035,
|
|
"correction_max": 1.0091222524642944,
|
|
"codebook": [
|
|
-0.9788122773170471,
|
|
-0.7834769487380981,
|
|
-0.6198369264602661,
|
|
-0.4782615900039673,
|
|
-0.35483551025390625,
|
|
-0.24387358129024506,
|
|
-0.1409580558538437,
|
|
-0.042012955993413925,
|
|
0.05453219637274742,
|
|
0.15002252161502838,
|
|
0.2507072687149048,
|
|
0.3597245216369629,
|
|
0.48037007451057434,
|
|
0.6193817853927612,
|
|
0.7801131010055542,
|
|
0.9774166345596313
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9614005088806152,
|
|
"extreme_code_fraction": 0.09559067338705063,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.48399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020260901656001806,
|
|
"max_absolute_error": 0.007052063941955566,
|
|
"cosine_loss": {
|
|
"mean": 1.7179354472318664e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.722574234008789e-05,
|
|
"max": 2.6047229766845703e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004316517151892185,
|
|
"p95": 0.0009427070617675781,
|
|
"p99": 0.00473436713218689,
|
|
"max": 0.006128787994384766
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3289999999979045
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.16.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012499531731009483,
|
|
"metrics": {
|
|
"relative_error": 0.07309065014123917,
|
|
"max_absolute_error": 0.1484375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026705730706453323,
|
|
"p95": 0.0027374178171157837,
|
|
"p99": 0.0027818530797958374,
|
|
"max": 0.003048539161682129
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00038499306538142264,
|
|
"p95": 0.0010050386190414429,
|
|
"p99": 0.0016624927520751953,
|
|
"max": 0.0033288002014160156
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0030650950502604246,
|
|
"p95": 0.005257695913314819,
|
|
"p99": 0.006419703364372253,
|
|
"max": 0.008618772029876709
|
|
},
|
|
"correction_min": 0.99528568983078,
|
|
"correction_max": 1.0086936950683594,
|
|
"codebook": [
|
|
-0.9776725172996521,
|
|
-0.783072829246521,
|
|
-0.620984673500061,
|
|
-0.48263436555862427,
|
|
-0.3611128628253937,
|
|
-0.25233861804008484,
|
|
-0.15135052800178528,
|
|
-0.05555819347500801,
|
|
0.04052455723285675,
|
|
0.13981547951698303,
|
|
0.24309740960597992,
|
|
0.35460084676742554,
|
|
0.4775941073894501,
|
|
0.6177645325660706,
|
|
0.7814970016479492,
|
|
0.9779009222984314
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9615092277526855,
|
|
"extreme_code_fraction": 0.09555917978286743,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.0779999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019927294924855232,
|
|
"max_absolute_error": 0.008461475372314453,
|
|
"cosine_loss": {
|
|
"mean": 1.7144983530670288e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8894672393798828e-05,
|
|
"max": 2.6464462280273438e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004216799861751497,
|
|
"p95": 0.0009601443307474256,
|
|
"p99": 0.0038513303734362125,
|
|
"max": 0.005940377712249756
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.48399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012507587671279907,
|
|
"metrics": {
|
|
"relative_error": 0.07305126637220383,
|
|
"max_absolute_error": 0.09228515625,
|
|
"cosine_loss": {
|
|
"mean": 0.002667315537109971,
|
|
"p95": 0.002740144729614258,
|
|
"p99": 0.0027707815170288086,
|
|
"max": 0.0028555989265441895
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00038501303060911596,
|
|
"p95": 0.0010023117065429688,
|
|
"p99": 0.0016150182345882058,
|
|
"max": 0.003234267234802246
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0030764879193156958,
|
|
"p95": 0.0053377654403448105,
|
|
"p99": 0.006354060024023056,
|
|
"max": 0.008716940879821777
|
|
},
|
|
"correction_min": 0.9974031448364258,
|
|
"correction_max": 1.008793592453003,
|
|
"codebook": [
|
|
-0.9769930839538574,
|
|
-0.7808471918106079,
|
|
-0.619684636592865,
|
|
-0.48189234733581543,
|
|
-0.36052975058555603,
|
|
-0.25111502408981323,
|
|
-0.15028952062129974,
|
|
-0.0540250726044178,
|
|
0.041254572570323944,
|
|
0.1402992308139801,
|
|
0.24423393607139587,
|
|
0.35468557476997375,
|
|
0.47738608717918396,
|
|
0.6152099967002869,
|
|
0.7785996198654175,
|
|
0.9769843220710754
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9620776176452637,
|
|
"extreme_code_fraction": 0.09597839415073395,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.45300000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001998881809413433,
|
|
"max_absolute_error": 0.01208353042602539,
|
|
"cosine_loss": {
|
|
"mean": 1.7059057881851913e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8134713172912598e-05,
|
|
"max": 2.5331974029541016e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004256315005477518,
|
|
"p95": 0.0009316354990005493,
|
|
"p99": 0.004417300224304199,
|
|
"max": 0.006493747234344482
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3289999999979045
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.17.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012503421166911721,
|
|
"metrics": {
|
|
"relative_error": 0.07306862622499466,
|
|
"max_absolute_error": 0.119140625,
|
|
"cosine_loss": {
|
|
"mean": 0.0026705306954681873,
|
|
"p95": 0.0027378350496292114,
|
|
"p99": 0.00277651846408844,
|
|
"max": 0.0031507015228271484
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00036703021032735705,
|
|
"p95": 0.0009689778089523315,
|
|
"p99": 0.0015984177589416504,
|
|
"max": 0.003071308135986328
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032374546863138676,
|
|
"p95": 0.005413532257080078,
|
|
"p99": 0.006616577506065369,
|
|
"max": 0.01045602560043335
|
|
},
|
|
"correction_min": 0.9959708452224731,
|
|
"correction_max": 1.0105664730072021,
|
|
"codebook": [
|
|
-0.9771445393562317,
|
|
-0.7788976430892944,
|
|
-0.6172286868095398,
|
|
-0.4785648584365845,
|
|
-0.3550886809825897,
|
|
-0.24221321940422058,
|
|
-0.13902994990348816,
|
|
-0.040557801723480225,
|
|
0.05557725206017494,
|
|
0.15182317793369293,
|
|
0.25308385491371155,
|
|
0.3624213635921478,
|
|
0.4841189384460449,
|
|
0.6219184994697571,
|
|
0.784030556678772,
|
|
0.9782681465148926
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961500883102417,
|
|
"extreme_code_fraction": 0.09569644927978516,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.0779999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020025367848575115,
|
|
"max_absolute_error": 0.007061302661895752,
|
|
"cosine_loss": {
|
|
"mean": 1.7311623423665878e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9729137420654297e-05,
|
|
"max": 2.586841583251953e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004288815543986857,
|
|
"p95": 0.000973820686340332,
|
|
"p99": 0.0037694773636758327,
|
|
"max": 0.006679713726043701
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5470000000204891
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012427933979779482,
|
|
"metrics": {
|
|
"relative_error": 0.0730639398097992,
|
|
"max_absolute_error": 0.10308837890625,
|
|
"cosine_loss": {
|
|
"mean": 0.002668644767254591,
|
|
"p95": 0.002740946365520358,
|
|
"p99": 0.0027703815139830112,
|
|
"max": 0.0028285980224609375
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000397516560042277,
|
|
"p95": 0.0010126172564923763,
|
|
"p99": 0.0016231287736445665,
|
|
"max": 0.003497600555419922
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003108740784227848,
|
|
"p95": 0.005336371250450611,
|
|
"p99": 0.006300187669694424,
|
|
"max": 0.008803784847259521
|
|
},
|
|
"correction_min": 0.9967655539512634,
|
|
"correction_max": 1.0088820457458496,
|
|
"codebook": [
|
|
-0.9775850772857666,
|
|
-0.7817254066467285,
|
|
-0.617661714553833,
|
|
-0.4781286418437958,
|
|
-0.3544659912586212,
|
|
-0.24410703778266907,
|
|
-0.1404627561569214,
|
|
-0.04147262126207352,
|
|
0.05550379678606987,
|
|
0.1519991010427475,
|
|
0.25282198190689087,
|
|
0.36182042956352234,
|
|
0.48360759019851685,
|
|
0.6216828227043152,
|
|
0.7816290855407715,
|
|
0.9781396389007568
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9612231254577637,
|
|
"extreme_code_fraction": 0.0954444631934166,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.5309999999881256
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002011829288676381,
|
|
"max_absolute_error": 0.007781863212585449,
|
|
"cosine_loss": {
|
|
"mean": 1.7025353145072586e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7762184143066406e-05,
|
|
"max": 2.5570392608642578e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042800596565939486,
|
|
"p95": 0.0009342879056930542,
|
|
"p99": 0.004579260945320129,
|
|
"max": 0.006739020347595215
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.4539999999979045
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.18.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012485758634284139,
|
|
"metrics": {
|
|
"relative_error": 0.07306292653083801,
|
|
"max_absolute_error": 0.108642578125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026723158080130816,
|
|
"p95": 0.0027530938386917114,
|
|
"p99": 0.0028320997953414917,
|
|
"max": 0.0031780600547790527
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000352613307768479,
|
|
"p95": 0.0009527355432510376,
|
|
"p99": 0.0016046017408370972,
|
|
"max": 0.0035677552223205566
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003177021397277713,
|
|
"p95": 0.005685746669769287,
|
|
"p99": 0.0075121670961380005,
|
|
"max": 0.012910544872283936
|
|
},
|
|
"correction_min": 0.9906858205795288,
|
|
"correction_max": 1.0130794048309326,
|
|
"codebook": [
|
|
-0.9769662022590637,
|
|
-0.7800645232200623,
|
|
-0.617825448513031,
|
|
-0.4787161648273468,
|
|
-0.3582826554775238,
|
|
-0.25007393956184387,
|
|
-0.15070584416389465,
|
|
-0.05632083863019943,
|
|
0.03995516151189804,
|
|
0.1397925168275833,
|
|
0.24363696575164795,
|
|
0.35521507263183594,
|
|
0.47798383235931396,
|
|
0.6186531782150269,
|
|
0.7804432511329651,
|
|
0.9780014753341675
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9622342586517334,
|
|
"extreme_code_fraction": 0.09600837528705597,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.09299999999348074
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002001724438741803,
|
|
"max_absolute_error": 0.008861303329467773,
|
|
"cosine_loss": {
|
|
"mean": 1.7232461004823563e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9131344743072987e-05,
|
|
"max": 2.676248550415039e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004258172702975571,
|
|
"p95": 0.0009651121217757463,
|
|
"p99": 0.003900570096448064,
|
|
"max": 0.006104588508605957
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5780000000086147
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012461126316338778,
|
|
"metrics": {
|
|
"relative_error": 0.07306897640228271,
|
|
"max_absolute_error": 0.107421875,
|
|
"cosine_loss": {
|
|
"mean": 0.00266926153562963,
|
|
"p95": 0.002741515636444092,
|
|
"p99": 0.0027714543975889683,
|
|
"max": 0.002853214740753174
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003879388968925923,
|
|
"p95": 0.0010165658313781023,
|
|
"p99": 0.001656864071264863,
|
|
"max": 0.003631114959716797
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0030925541650503874,
|
|
"p95": 0.005401255562901497,
|
|
"p99": 0.0064064147882163525,
|
|
"max": 0.009013950824737549
|
|
},
|
|
"correction_min": 0.9969428181648254,
|
|
"correction_max": 1.0090959072113037,
|
|
"codebook": [
|
|
-0.9766501188278198,
|
|
-0.7766307592391968,
|
|
-0.6148095726966858,
|
|
-0.4780255854129791,
|
|
-0.3555803894996643,
|
|
-0.24515295028686523,
|
|
-0.14102588593959808,
|
|
-0.04187783598899841,
|
|
0.05478145182132721,
|
|
0.15131346881389618,
|
|
0.25194934010505676,
|
|
0.36097419261932373,
|
|
0.48253145813941956,
|
|
0.6207032203674316,
|
|
0.7806132435798645,
|
|
0.9779626727104187
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961913585662842,
|
|
"extreme_code_fraction": 0.09616309404373169,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.4539999999979045
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019881611224263906,
|
|
"max_absolute_error": 0.007498323917388916,
|
|
"cosine_loss": {
|
|
"mean": 1.6876119843800552e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.806020736694336e-05,
|
|
"max": 2.6941299438476562e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004188230959698558,
|
|
"p95": 0.0009246915578842163,
|
|
"p99": 0.0043170154094696045,
|
|
"max": 0.006418883800506592
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3120000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.19.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012417329708114266,
|
|
"metrics": {
|
|
"relative_error": 0.07335453480482101,
|
|
"max_absolute_error": 0.12103271484375,
|
|
"cosine_loss": {
|
|
"mean": 0.00268702395260334,
|
|
"p95": 0.0027737170457839966,
|
|
"p99": 0.002825304865837097,
|
|
"max": 0.003297269344329834
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003589142288547009,
|
|
"p95": 0.0009537488222122192,
|
|
"p99": 0.0016285628080368042,
|
|
"max": 0.003449559211730957
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003612634725868702,
|
|
"p95": 0.006310075521469116,
|
|
"p99": 0.007793202996253967,
|
|
"max": 0.011671245098114014
|
|
},
|
|
"correction_min": 0.9976886510848999,
|
|
"correction_max": 1.0118091106414795,
|
|
"codebook": [
|
|
-0.9781890511512756,
|
|
-0.7811108827590942,
|
|
-0.6216534376144409,
|
|
-0.4840618073940277,
|
|
-0.36329561471939087,
|
|
-0.25457146763801575,
|
|
-0.15423960983753204,
|
|
-0.05910244956612587,
|
|
0.038383495062589645,
|
|
0.13873717188835144,
|
|
0.24406872689723969,
|
|
0.35590463876724243,
|
|
0.47906258702278137,
|
|
0.6186375617980957,
|
|
0.7820883393287659,
|
|
0.9780892729759216
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9609053134918213,
|
|
"extreme_code_fraction": 0.0959828644990921,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.09399999998277053
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019923276267945766,
|
|
"max_absolute_error": 0.009412050247192383,
|
|
"cosine_loss": {
|
|
"mean": 1.6983083241939312e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9073486328125e-05,
|
|
"max": 2.7358531951904297e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004333192191552371,
|
|
"p95": 0.0009881258010864258,
|
|
"p99": 0.0039663459174335,
|
|
"max": 0.00716012716293335
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5620000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012460768921300769,
|
|
"metrics": {
|
|
"relative_error": 0.07307019829750061,
|
|
"max_absolute_error": 0.083984375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026690128725022078,
|
|
"p95": 0.002741575241088867,
|
|
"p99": 0.002772008301690221,
|
|
"max": 0.0028429031372070312
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00039721428765915334,
|
|
"p95": 0.0010078817140311003,
|
|
"p99": 0.0016266889870166779,
|
|
"max": 0.003782808780670166
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003442144952714443,
|
|
"p95": 0.005600818432867527,
|
|
"p99": 0.006483420729637146,
|
|
"max": 0.009023725986480713
|
|
},
|
|
"correction_min": 0.9977447986602783,
|
|
"correction_max": 1.009105920791626,
|
|
"codebook": [
|
|
-0.9790468811988831,
|
|
-0.7838441133499146,
|
|
-0.6235384941101074,
|
|
-0.4846635162830353,
|
|
-0.3624899089336395,
|
|
-0.25315892696380615,
|
|
-0.15189972519874573,
|
|
-0.055098842829465866,
|
|
0.04065001755952835,
|
|
0.1378309726715088,
|
|
0.24034000933170319,
|
|
0.3521324098110199,
|
|
0.47537943720817566,
|
|
0.615478515625,
|
|
0.7778687477111816,
|
|
0.9771729111671448
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.963102340698242,
|
|
"extreme_code_fraction": 0.09584935009479523,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.4379999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002043333603069186,
|
|
"max_absolute_error": 0.007771492004394531,
|
|
"cosine_loss": {
|
|
"mean": 1.7313064972768188e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.780688762664795e-05,
|
|
"max": 2.586841583251953e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004358365258667618,
|
|
"p95": 0.0009390711784362793,
|
|
"p99": 0.0046952515840530396,
|
|
"max": 0.0066184401512146
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3120000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.2.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001993573736399412,
|
|
"max_absolute_error": 0.011240720748901367,
|
|
"cosine_loss": {
|
|
"mean": 1.6188826066354522e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.4722347259521484e-05,
|
|
"max": 2.5510787963867188e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000470608938485384,
|
|
"p95": 0.0010488331317901611,
|
|
"p99": 0.005024775862693787,
|
|
"max": 0.007374465465545654
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.1720000000204891
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020009588915854692,
|
|
"max_absolute_error": 0.009727954864501953,
|
|
"cosine_loss": {
|
|
"mean": 1.7098029729822883e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9012135453522205e-05,
|
|
"max": 2.6285648345947266e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004234626248944551,
|
|
"p95": 0.000962716294452548,
|
|
"p99": 0.00383986858651042,
|
|
"max": 0.00682598352432251
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5309999999881256
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001247445004992187,
|
|
"metrics": {
|
|
"relative_error": 0.07307162135839462,
|
|
"max_absolute_error": 0.095703125,
|
|
"cosine_loss": {
|
|
"mean": 0.002669470151886344,
|
|
"p95": 0.002741873264312744,
|
|
"p99": 0.002772765699774027,
|
|
"max": 0.002853095531463623
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00041518043144606054,
|
|
"p95": 0.0010445117950439453,
|
|
"p99": 0.0016341307200491428,
|
|
"max": 0.0032913684844970703
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031671086326241493,
|
|
"p95": 0.005361670162528753,
|
|
"p99": 0.006290687248110771,
|
|
"max": 0.009485065937042236
|
|
},
|
|
"correction_min": 0.9978384971618652,
|
|
"correction_max": 1.0095758438110352,
|
|
"codebook": [
|
|
-0.9789506196975708,
|
|
-0.7844958901405334,
|
|
-0.6224448084831238,
|
|
-0.48222219944000244,
|
|
-0.3570234179496765,
|
|
-0.24454346299171448,
|
|
-0.14050468802452087,
|
|
-0.04143466055393219,
|
|
0.054315198212862015,
|
|
0.14928169548511505,
|
|
0.2509315609931946,
|
|
0.3600258231163025,
|
|
0.48128944635391235,
|
|
0.6198111772537231,
|
|
0.7811448574066162,
|
|
0.9778244495391846
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.960285186767578,
|
|
"extreme_code_fraction": 0.09541946649551392,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.48499999998603016
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002025807276368141,
|
|
"max_absolute_error": 0.009476065635681152,
|
|
"cosine_loss": {
|
|
"mean": 1.7199643025378464e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8149614334106445e-05,
|
|
"max": 2.7120113372802734e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042740683420561254,
|
|
"p95": 0.0009321868419647217,
|
|
"p99": 0.004452049732208252,
|
|
"max": 0.006491243839263916
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.20.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002010381082072854,
|
|
"max_absolute_error": 0.006331026554107666,
|
|
"cosine_loss": {
|
|
"mean": 1.6760036487539764e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7762184143066406e-05,
|
|
"max": 2.485513687133789e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004578141961246729,
|
|
"p95": 0.00099848210811615,
|
|
"p99": 0.004613116383552551,
|
|
"max": 0.006993532180786133
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.15600000001722947
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020029814913868904,
|
|
"max_absolute_error": 0.007250964641571045,
|
|
"cosine_loss": {
|
|
"mean": 1.7222594124177704e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.919269561767578e-05,
|
|
"max": 2.8192996978759766e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042828600271604955,
|
|
"p95": 0.0009914098773151636,
|
|
"p99": 0.0038266312330961227,
|
|
"max": 0.0070694684982299805
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5629999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012502626050263643,
|
|
"metrics": {
|
|
"relative_error": 0.0730261504650116,
|
|
"max_absolute_error": 0.115234375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026662438176572323,
|
|
"p95": 0.0027382969856262207,
|
|
"p99": 0.0027689337730407715,
|
|
"max": 0.0028535127639770508
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00039593837573193014,
|
|
"p95": 0.001024460420012474,
|
|
"p99": 0.0016203731065616012,
|
|
"max": 0.0038710832595825195
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003040649462491274,
|
|
"p95": 0.005217015743255615,
|
|
"p99": 0.006183601450175047,
|
|
"max": 0.00865703821182251
|
|
},
|
|
"correction_min": 0.997248649597168,
|
|
"correction_max": 1.0087326765060425,
|
|
"codebook": [
|
|
-0.9771376252174377,
|
|
-0.779828667640686,
|
|
-0.6178810000419617,
|
|
-0.4804900288581848,
|
|
-0.3595130145549774,
|
|
-0.25013846158981323,
|
|
-0.14887571334838867,
|
|
-0.05392884835600853,
|
|
0.041207585483789444,
|
|
0.14040899276733398,
|
|
0.2443542778491974,
|
|
0.3560110032558441,
|
|
0.479487806558609,
|
|
0.619018018245697,
|
|
0.7813010811805725,
|
|
0.9780855774879456
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961355209350586,
|
|
"extreme_code_fraction": 0.09564834833145142,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.4529999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020252070389688015,
|
|
"max_absolute_error": 0.0068929195404052734,
|
|
"cosine_loss": {
|
|
"mean": 1.7345882952213287e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7896294593811035e-05,
|
|
"max": 2.6702880859375e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004329505318310112,
|
|
"p95": 0.0009413957595825195,
|
|
"p99": 0.004616990685462952,
|
|
"max": 0.006472468376159668
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3120000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.21.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020183364395052195,
|
|
"max_absolute_error": 0.007724940776824951,
|
|
"cosine_loss": {
|
|
"mean": 1.6804718825369491e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.74790620803833e-05,
|
|
"max": 2.5153160095214844e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004553064936771989,
|
|
"p95": 0.0010284781455993652,
|
|
"p99": 0.004538744688034058,
|
|
"max": 0.006571769714355469
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.20300000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019941439386457205,
|
|
"max_absolute_error": 0.0063790082931518555,
|
|
"cosine_loss": {
|
|
"mean": 1.7167352552860393e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9431114196777344e-05,
|
|
"max": 2.6345252990722656e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004237725806888193,
|
|
"p95": 0.0009546427754685283,
|
|
"p99": 0.003759596962481737,
|
|
"max": 0.0065645575523376465
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.7349999999860302
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012484043836593628,
|
|
"metrics": {
|
|
"relative_error": 0.07304798066616058,
|
|
"max_absolute_error": 0.0977783203125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026679050642997026,
|
|
"p95": 0.002739429473876953,
|
|
"p99": 0.0027694106101989746,
|
|
"max": 0.0028437376022338867
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00040070791146717966,
|
|
"p95": 0.0010253458749502897,
|
|
"p99": 0.001655392232351005,
|
|
"max": 0.0035730600357055664
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003179574850946665,
|
|
"p95": 0.005345242563635111,
|
|
"p99": 0.006268493831157684,
|
|
"max": 0.00945425033569336
|
|
},
|
|
"correction_min": 0.9971667528152466,
|
|
"correction_max": 1.0095444917678833,
|
|
"codebook": [
|
|
-0.9782240986824036,
|
|
-0.7831404805183411,
|
|
-0.6192691326141357,
|
|
-0.47905799746513367,
|
|
-0.3552889823913574,
|
|
-0.2436036914587021,
|
|
-0.14057311415672302,
|
|
-0.041776735335588455,
|
|
0.0546707846224308,
|
|
0.15074673295021057,
|
|
0.2518273591995239,
|
|
0.3604937493801117,
|
|
0.48187920451164246,
|
|
0.619606614112854,
|
|
0.7806544303894043,
|
|
0.9773144721984863
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9613404273986816,
|
|
"extreme_code_fraction": 0.09557387977838516,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.5780000000086147
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002102671889588237,
|
|
"max_absolute_error": 0.007472038269042969,
|
|
"cosine_loss": {
|
|
"mean": 1.8298050008525024e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9729137420654297e-05,
|
|
"max": 2.676248550415039e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00045291538117453456,
|
|
"p95": 0.0009874850511550903,
|
|
"p99": 0.004779160022735596,
|
|
"max": 0.007166743278503418
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.34400000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.22.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020432595629245043,
|
|
"max_absolute_error": 0.011760175228118896,
|
|
"cosine_loss": {
|
|
"mean": 1.6905057691474212e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7404556274414062e-05,
|
|
"max": 2.5391578674316406e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00047360273310914636,
|
|
"p95": 0.0010646581649780273,
|
|
"p99": 0.004756167531013489,
|
|
"max": 0.007371842861175537
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.139999999984866
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002014057943597436,
|
|
"max_absolute_error": 0.010355949401855469,
|
|
"cosine_loss": {
|
|
"mean": 1.729757059365511e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9252300262451172e-05,
|
|
"max": 2.5987625122070312e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043009090586565435,
|
|
"p95": 0.0009843169245868921,
|
|
"p99": 0.003918290603905916,
|
|
"max": 0.006603360176086426
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012529321247711778,
|
|
"metrics": {
|
|
"relative_error": 0.0731460228562355,
|
|
"max_absolute_error": 0.1015625,
|
|
"cosine_loss": {
|
|
"mean": 0.002674962393939495,
|
|
"p95": 0.0027467876207083464,
|
|
"p99": 0.002777236048132181,
|
|
"max": 0.002861320972442627
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003926205972675234,
|
|
"p95": 0.0010289100464433432,
|
|
"p99": 0.0016138989012688398,
|
|
"max": 0.0036466121673583984
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003438070882111788,
|
|
"p95": 0.005655318032950163,
|
|
"p99": 0.006603876128792763,
|
|
"max": 0.010676205158233643
|
|
},
|
|
"correction_min": 0.9975972175598145,
|
|
"correction_max": 1.0107914209365845,
|
|
"codebook": [
|
|
-0.9763455986976624,
|
|
-0.7757295370101929,
|
|
-0.6128580570220947,
|
|
-0.474846750497818,
|
|
-0.35319674015045166,
|
|
-0.2431473731994629,
|
|
-0.14002324640750885,
|
|
-0.041238896548748016,
|
|
0.05544481426477432,
|
|
0.15242242813110352,
|
|
0.25336143374443054,
|
|
0.36293548345565796,
|
|
0.4838194251060486,
|
|
0.6215898990631104,
|
|
0.7812080383300781,
|
|
0.9773171544075012
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.962594509124756,
|
|
"extreme_code_fraction": 0.09632913023233414,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.5320000000065193
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020231001544743776,
|
|
"max_absolute_error": 0.0071572065353393555,
|
|
"cosine_loss": {
|
|
"mean": 1.7122920326073654e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.87605619430542e-05,
|
|
"max": 2.5570392608642578e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042616043356247246,
|
|
"p95": 0.0009460747241973877,
|
|
"p99": 0.004407867789268494,
|
|
"max": 0.006392478942871094
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.34299999999348074
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.23.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002033951925113797,
|
|
"max_absolute_error": 0.007194876670837402,
|
|
"cosine_loss": {
|
|
"mean": 1.6852171711434494e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.755356788635254e-05,
|
|
"max": 2.6464462280273438e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004737508133985102,
|
|
"p95": 0.0010604411363601685,
|
|
"p99": 0.004773303866386414,
|
|
"max": 0.00746232271194458
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.15700000000651926
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020055249333381653,
|
|
"max_absolute_error": 0.009220719337463379,
|
|
"cosine_loss": {
|
|
"mean": 1.7276062180826557e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.936976332217455e-05,
|
|
"max": 2.6464462280273438e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004258749249856919,
|
|
"p95": 0.0009649095591157675,
|
|
"p99": 0.003835939336568117,
|
|
"max": 0.006241500377655029
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5460000000020955
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001244729501195252,
|
|
"metrics": {
|
|
"relative_error": 0.07302900403738022,
|
|
"max_absolute_error": 0.1015625,
|
|
"cosine_loss": {
|
|
"mean": 0.0026664799079298973,
|
|
"p95": 0.0027384161949157715,
|
|
"p99": 0.002766907215118408,
|
|
"max": 0.0028293728828430176
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00039584198384545743,
|
|
"p95": 0.0010173588525503874,
|
|
"p99": 0.0016018152236938477,
|
|
"max": 0.003432154655456543
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003285657148808241,
|
|
"p95": 0.005464660935103893,
|
|
"p99": 0.006396831478923559,
|
|
"max": 0.008360683917999268
|
|
},
|
|
"correction_min": 0.9976272583007812,
|
|
"correction_max": 1.0084311962127686,
|
|
"codebook": [
|
|
-0.9781870245933533,
|
|
-0.7813246846199036,
|
|
-0.6171754002571106,
|
|
-0.4766973555088043,
|
|
-0.35328835248947144,
|
|
-0.24206753075122833,
|
|
-0.13942237198352814,
|
|
-0.041098520159721375,
|
|
0.054546840488910675,
|
|
0.15140147507190704,
|
|
0.25194406509399414,
|
|
0.3604848086833954,
|
|
0.48080846667289734,
|
|
0.6179195642471313,
|
|
0.778866171836853,
|
|
0.9774799346923828
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9630789756774902,
|
|
"extreme_code_fraction": 0.09569866210222244,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.48499999998603016
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002012686338275671,
|
|
"max_absolute_error": 0.009005308151245117,
|
|
"cosine_loss": {
|
|
"mean": 1.6904281210372574e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7508864402770996e-05,
|
|
"max": 2.47955322265625e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004191319749224931,
|
|
"p95": 0.0009107142686843872,
|
|
"p99": 0.004593804478645325,
|
|
"max": 0.006107926368713379
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.34400000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.24.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002029566327109933,
|
|
"max_absolute_error": 0.006404876708984375,
|
|
"cosine_loss": {
|
|
"mean": 1.7011826685120468e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.71661376953125e-05,
|
|
"max": 2.5272369384765625e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00046829532948322594,
|
|
"p95": 0.0010110139846801758,
|
|
"p99": 0.004683181643486023,
|
|
"max": 0.006733715534210205
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.1409999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020115019287914038,
|
|
"max_absolute_error": 0.006864666938781738,
|
|
"cosine_loss": {
|
|
"mean": 1.7383717931807041e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9012135453522205e-05,
|
|
"max": 2.676248550415039e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043007900239899755,
|
|
"p95": 0.000979047967121005,
|
|
"p99": 0.003981146961450577,
|
|
"max": 0.0060999393463134766
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5620000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012500284938141704,
|
|
"metrics": {
|
|
"relative_error": 0.07304198294878006,
|
|
"max_absolute_error": 0.121826171875,
|
|
"cosine_loss": {
|
|
"mean": 0.0026673779357224703,
|
|
"p95": 0.002738773822784424,
|
|
"p99": 0.0027679302729666233,
|
|
"max": 0.002834916114807129
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003804749867413193,
|
|
"p95": 0.0009882061276584864,
|
|
"p99": 0.0016247176099568605,
|
|
"max": 0.003682374954223633
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031780300196260214,
|
|
"p95": 0.005515006370842457,
|
|
"p99": 0.0065300352871418,
|
|
"max": 0.010756075382232666
|
|
},
|
|
"correction_min": 0.9963030815124512,
|
|
"correction_max": 1.0108729600906372,
|
|
"codebook": [
|
|
-0.9769720435142517,
|
|
-0.7762575149536133,
|
|
-0.6153621077537537,
|
|
-0.4790832996368408,
|
|
-0.3592900335788727,
|
|
-0.2509567439556122,
|
|
-0.15073372423648834,
|
|
-0.054918162524700165,
|
|
0.04191010445356369,
|
|
0.14143937826156616,
|
|
0.2449364811182022,
|
|
0.35567593574523926,
|
|
0.477987676858902,
|
|
0.6171525120735168,
|
|
0.7789350748062134,
|
|
0.9767820835113525
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.963003635406494,
|
|
"extreme_code_fraction": 0.09627622365951538,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.5
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020194407552480698,
|
|
"max_absolute_error": 0.0076448917388916016,
|
|
"cosine_loss": {
|
|
"mean": 1.7254968724955688e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.735985279083252e-05,
|
|
"max": 2.7835369110107422e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004294300451874733,
|
|
"p95": 0.0009266287088394165,
|
|
"p99": 0.004655823111534119,
|
|
"max": 0.006606996059417725
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.34400000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.25.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020535048097372055,
|
|
"max_absolute_error": 0.008224934339523315,
|
|
"cosine_loss": {
|
|
"mean": 1.6889424614419113e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.6987323760986328e-05,
|
|
"max": 2.6702880859375e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004886323004029691,
|
|
"p95": 0.0010824054479599,
|
|
"p99": 0.005139157176017761,
|
|
"max": 0.007185876369476318
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.14100000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020206132903695107,
|
|
"max_absolute_error": 0.009591341018676758,
|
|
"cosine_loss": {
|
|
"mean": 1.7557205183038604e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9786995835602283e-05,
|
|
"max": 2.765655517578125e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043253370677120984,
|
|
"p95": 0.0009720056550577283,
|
|
"p99": 0.003840502817183733,
|
|
"max": 0.006171464920043945
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5620000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012506833299994469,
|
|
"metrics": {
|
|
"relative_error": 0.0731029361486435,
|
|
"max_absolute_error": 0.1263427734375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026718100998550653,
|
|
"p95": 0.0027437806129455566,
|
|
"p99": 0.002773123327642679,
|
|
"max": 0.0028377771377563477
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004104463441763073,
|
|
"p95": 0.001028591301292181,
|
|
"p99": 0.0016283837612718344,
|
|
"max": 0.0032432079315185547
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.002898795995861292,
|
|
"p95": 0.005080803297460079,
|
|
"p99": 0.006070846691727638,
|
|
"max": 0.00849759578704834
|
|
},
|
|
"correction_min": 0.9976011514663696,
|
|
"correction_max": 1.0085704326629639,
|
|
"codebook": [
|
|
-0.9774266481399536,
|
|
-0.7812771201133728,
|
|
-0.620167076587677,
|
|
-0.48324206471443176,
|
|
-0.362003356218338,
|
|
-0.2522214651107788,
|
|
-0.15119211375713348,
|
|
-0.05508837848901749,
|
|
0.04174823313951492,
|
|
0.14111612737178802,
|
|
0.24450068175792694,
|
|
0.3557751178741455,
|
|
0.47951415181159973,
|
|
0.6202238202095032,
|
|
0.7824305891990662,
|
|
0.9782564640045166
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9604110717773438,
|
|
"extreme_code_fraction": 0.09542886912822723,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.5159999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002010982483625412,
|
|
"max_absolute_error": 0.009250402450561523,
|
|
"cosine_loss": {
|
|
"mean": 1.710684387035144e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8134713172912598e-05,
|
|
"max": 2.5331974029541016e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042544101597741246,
|
|
"p95": 0.0009279847145080566,
|
|
"p99": 0.004552692174911499,
|
|
"max": 0.006285488605499268
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.26.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002048676833510399,
|
|
"max_absolute_error": 0.010514438152313232,
|
|
"cosine_loss": {
|
|
"mean": 1.6873901813596603e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.735985279083252e-05,
|
|
"max": 2.4020671844482422e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004836437292397022,
|
|
"p95": 0.0010831505060195923,
|
|
"p99": 0.004761844873428345,
|
|
"max": 0.006913185119628906
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020284049678593874,
|
|
"max_absolute_error": 0.011728644371032715,
|
|
"cosine_loss": {
|
|
"mean": 1.7606460005481495e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9490718841552734e-05,
|
|
"max": 2.6285648345947266e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043626510887406766,
|
|
"p95": 0.0009838638361543417,
|
|
"p99": 0.004000476561486721,
|
|
"max": 0.005999147891998291
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.6710000000020955
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012449173955246806,
|
|
"metrics": {
|
|
"relative_error": 0.07311682403087616,
|
|
"max_absolute_error": 0.1142578125,
|
|
"cosine_loss": {
|
|
"mean": 0.00267297332175076,
|
|
"p95": 0.0027461647987365723,
|
|
"p99": 0.0027776360511779785,
|
|
"max": 0.0028561949729919434
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000402721663704142,
|
|
"p95": 0.001028439262881875,
|
|
"p99": 0.0016139247454702854,
|
|
"max": 0.0033186674118041992
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0030299543868750334,
|
|
"p95": 0.005178895778954029,
|
|
"p99": 0.006120704114437103,
|
|
"max": 0.008606672286987305
|
|
},
|
|
"correction_min": 0.9974709749221802,
|
|
"correction_max": 1.0086814165115356,
|
|
"codebook": [
|
|
-0.9778342247009277,
|
|
-0.7823407649993896,
|
|
-0.6209824681282043,
|
|
-0.4836476743221283,
|
|
-0.36186227202415466,
|
|
-0.2516743838787079,
|
|
-0.15101103484630585,
|
|
-0.054845597594976425,
|
|
0.041924554854631424,
|
|
0.14139854907989502,
|
|
0.2450602948665619,
|
|
0.35625383257865906,
|
|
0.4791353642940521,
|
|
0.6206207871437073,
|
|
0.7836724519729614,
|
|
0.9785200953483582
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9604501724243164,
|
|
"extreme_code_fraction": 0.09535805135965347,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.6099999999860302
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020174458622932434,
|
|
"max_absolute_error": 0.013645172119140625,
|
|
"cosine_loss": {
|
|
"mean": 1.737459911055339e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.901388168334961e-05,
|
|
"max": 2.5987625122070312e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004230924532748759,
|
|
"p95": 0.0009134262800216675,
|
|
"p99": 0.004379048943519592,
|
|
"max": 0.0061419010162353516
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.27.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020408094860613346,
|
|
"max_absolute_error": 0.011786788702011108,
|
|
"cosine_loss": {
|
|
"mean": 1.6917364291657577e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7344951629638672e-05,
|
|
"max": 2.6524066925048828e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004710892972070724,
|
|
"p95": 0.0010690391063690186,
|
|
"p99": 0.004683971405029297,
|
|
"max": 0.007322132587432861
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.14100000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019979202188551426,
|
|
"max_absolute_error": 0.008053779602050781,
|
|
"cosine_loss": {
|
|
"mean": 1.7285375406572712e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9252300262451172e-05,
|
|
"max": 2.7239322662353516e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042413806659169495,
|
|
"p95": 0.000964128877967596,
|
|
"p99": 0.003768572583794594,
|
|
"max": 0.0063745975494384766
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.48399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012463305611163378,
|
|
"metrics": {
|
|
"relative_error": 0.07305961847305298,
|
|
"max_absolute_error": 0.1201171875,
|
|
"cosine_loss": {
|
|
"mean": 0.002669163281098008,
|
|
"p95": 0.0027411580085754395,
|
|
"p99": 0.002770202700048685,
|
|
"max": 0.002839982509613037
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00038881757063791156,
|
|
"p95": 0.0009897048585116863,
|
|
"p99": 0.0015888756606727839,
|
|
"max": 0.0037201642990112305
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0033767251297831535,
|
|
"p95": 0.005545386113226414,
|
|
"p99": 0.006471415050327778,
|
|
"max": 0.008712232112884521
|
|
},
|
|
"correction_min": 0.99753338098526,
|
|
"correction_max": 1.008788824081421,
|
|
"codebook": [
|
|
-0.9779796004295349,
|
|
-0.778030514717102,
|
|
-0.6136416792869568,
|
|
-0.474779337644577,
|
|
-0.35199350118637085,
|
|
-0.24111084640026093,
|
|
-0.1393001228570938,
|
|
-0.04144776985049248,
|
|
0.05470388010144234,
|
|
0.15134268999099731,
|
|
0.25211963057518005,
|
|
0.36152026057243347,
|
|
0.4836532473564148,
|
|
0.621981680393219,
|
|
0.7839962244033813,
|
|
0.9789667725563049
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9632811546325684,
|
|
"extreme_code_fraction": 0.09577586501836777,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.48399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020208628848195076,
|
|
"max_absolute_error": 0.012722492218017578,
|
|
"cosine_loss": {
|
|
"mean": 1.7256631963391555e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8224120140075684e-05,
|
|
"max": 2.6047229766845703e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000425056554377079,
|
|
"p95": 0.0009205341339111328,
|
|
"p99": 0.004567995667457581,
|
|
"max": 0.006639957427978516
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.2970000000204891
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.28.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0159999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001998183550313115,
|
|
"max_absolute_error": 0.007251381874084473,
|
|
"cosine_loss": {
|
|
"mean": 1.652232754167926e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.6689300537109375e-05,
|
|
"max": 2.6047229766845703e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004582441470120102,
|
|
"p95": 0.0010364800691604614,
|
|
"p99": 0.004545480012893677,
|
|
"max": 0.007384240627288818
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020148695912212133,
|
|
"max_absolute_error": 0.009276866912841797,
|
|
"cosine_loss": {
|
|
"mean": 1.7465763448853977e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9667786546051502e-05,
|
|
"max": 2.777576446533203e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043070511310361326,
|
|
"p95": 0.000976225477643311,
|
|
"p99": 0.0038602545391768217,
|
|
"max": 0.006132364273071289
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.5
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001249536988325417,
|
|
"metrics": {
|
|
"relative_error": 0.07301115244626999,
|
|
"max_absolute_error": 0.12109375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026655022520571947,
|
|
"p95": 0.0027376413345336914,
|
|
"p99": 0.0027672648429870605,
|
|
"max": 0.0028440356254577637
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003746664442587644,
|
|
"p95": 0.0009453832171857357,
|
|
"p99": 0.0015365503495559096,
|
|
"max": 0.0036973953247070312
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003261665580794215,
|
|
"p95": 0.005475410260260105,
|
|
"p99": 0.0064895316027104855,
|
|
"max": 0.008906245231628418
|
|
},
|
|
"correction_min": 0.9967798590660095,
|
|
"correction_max": 1.008986234664917,
|
|
"codebook": [
|
|
-0.9780310988426208,
|
|
-0.7808122038841248,
|
|
-0.6172031164169312,
|
|
-0.4773983061313629,
|
|
-0.3546149730682373,
|
|
-0.24336963891983032,
|
|
-0.13978011906147003,
|
|
-0.04109622538089752,
|
|
0.05490574613213539,
|
|
0.15157757699489594,
|
|
0.25312066078186035,
|
|
0.36196696758270264,
|
|
0.4816761910915375,
|
|
0.620995044708252,
|
|
0.7844831347465515,
|
|
0.9784205555915833
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961660861968994,
|
|
"extreme_code_fraction": 0.09543697535991669,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.4220000000204891
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019840742461383343,
|
|
"max_absolute_error": 0.010926365852355957,
|
|
"cosine_loss": {
|
|
"mean": 1.6872683090696228e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.767277717590332e-05,
|
|
"max": 2.562999725341797e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00041459102067165077,
|
|
"p95": 0.0009192526340484619,
|
|
"p99": 0.00438903272151947,
|
|
"max": 0.005964457988739014
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.29.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.001248509855940938,
|
|
"metrics": {
|
|
"relative_error": 0.07333096861839294,
|
|
"max_absolute_error": 0.089630126953125,
|
|
"cosine_loss": {
|
|
"mean": 0.002686302177608013,
|
|
"p95": 0.0027664899826049805,
|
|
"p99": 0.002816617488861084,
|
|
"max": 0.0029743313789367676
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003749376046471298,
|
|
"p95": 0.000999167561531067,
|
|
"p99": 0.0016755610704421997,
|
|
"max": 0.0032570958137512207
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0035119063686579466,
|
|
"p95": 0.006020054221153259,
|
|
"p99": 0.007387101650238037,
|
|
"max": 0.010042071342468262
|
|
},
|
|
"correction_min": 0.9966763257980347,
|
|
"correction_max": 1.0101439952850342,
|
|
"codebook": [
|
|
-0.9780982136726379,
|
|
-0.7814736366271973,
|
|
-0.6209588646888733,
|
|
-0.48313865065574646,
|
|
-0.3621237277984619,
|
|
-0.2535964548587799,
|
|
-0.1536179482936859,
|
|
-0.05881449580192566,
|
|
0.03777993097901344,
|
|
0.13778099417686462,
|
|
0.23988988995552063,
|
|
0.35115888714790344,
|
|
0.47338876128196716,
|
|
0.6130291819572449,
|
|
0.7781274318695068,
|
|
0.9775399565696716
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9631967544555664,
|
|
"extreme_code_fraction": 0.0961267352104187,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.0779999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001988836796954274,
|
|
"max_absolute_error": 0.010752201080322266,
|
|
"cosine_loss": {
|
|
"mean": 1.6953979411482578e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.919269561767578e-05,
|
|
"max": 2.8073787689208984e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042519040289334953,
|
|
"p95": 0.000974160386249423,
|
|
"p99": 0.0038834395818412304,
|
|
"max": 0.0072321295738220215
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21900000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001241085585206747,
|
|
"metrics": {
|
|
"relative_error": 0.07308437675237656,
|
|
"max_absolute_error": 0.0859375,
|
|
"cosine_loss": {
|
|
"mean": 0.002670402405783534,
|
|
"p95": 0.002743363380432129,
|
|
"p99": 0.002773880958557129,
|
|
"max": 0.0028484463691711426
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004131786117795855,
|
|
"p95": 0.0010472415015101433,
|
|
"p99": 0.0016216355143114924,
|
|
"max": 0.0034682750701904297
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0029563321731984615,
|
|
"p95": 0.005126308649778366,
|
|
"p99": 0.006054830737411976,
|
|
"max": 0.008166074752807617
|
|
},
|
|
"correction_min": 0.9971889853477478,
|
|
"correction_max": 1.0082333087921143,
|
|
"codebook": [
|
|
-0.9782000184059143,
|
|
-0.7822687029838562,
|
|
-0.6207121014595032,
|
|
-0.48226460814476013,
|
|
-0.3602254092693329,
|
|
-0.25131309032440186,
|
|
-0.1504942774772644,
|
|
-0.054728880524635315,
|
|
0.04204811155796051,
|
|
0.14084869623184204,
|
|
0.24497868120670319,
|
|
0.35626333951950073,
|
|
0.47989675402641296,
|
|
0.6199299693107605,
|
|
0.7829342484474182,
|
|
0.9781232476234436
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9603891372680664,
|
|
"extreme_code_fraction": 0.09537434577941895,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.26500000001396984
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020288960076868534,
|
|
"max_absolute_error": 0.007882535457611084,
|
|
"cosine_loss": {
|
|
"mean": 1.7331581148027908e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.806020736694336e-05,
|
|
"max": 2.6166439056396484e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004273711529094726,
|
|
"p95": 0.0009351521730422974,
|
|
"p99": 0.004440858960151672,
|
|
"max": 0.006464600563049316
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.3.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0159999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.00201789871789515,
|
|
"max_absolute_error": 0.010447025299072266,
|
|
"cosine_loss": {
|
|
"mean": 1.6608253190497635e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.704692840576172e-05,
|
|
"max": 2.580881118774414e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004742526216432452,
|
|
"p95": 0.0010795295238494873,
|
|
"p99": 0.004728153347969055,
|
|
"max": 0.007334887981414795
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.062000000005355105
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002019731793552637,
|
|
"max_absolute_error": 0.00908041000366211,
|
|
"cosine_loss": {
|
|
"mean": 1.748430690895475e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.942936796694994e-05,
|
|
"max": 2.676248550415039e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004298316198401153,
|
|
"p95": 0.0009663432138040662,
|
|
"p99": 0.00391407823190093,
|
|
"max": 0.006651520729064941
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.20300000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001253885100595653,
|
|
"metrics": {
|
|
"relative_error": 0.07305403053760529,
|
|
"max_absolute_error": 0.138671875,
|
|
"cosine_loss": {
|
|
"mean": 0.0026685898192226887,
|
|
"p95": 0.0027410387992858887,
|
|
"p99": 0.0027700066566467285,
|
|
"max": 0.002850949764251709
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00037697487277910113,
|
|
"p95": 0.000967556145042181,
|
|
"p99": 0.0015912891831248999,
|
|
"max": 0.003481745719909668
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031519639305770397,
|
|
"p95": 0.005262711085379124,
|
|
"p99": 0.006189619190990925,
|
|
"max": 0.008190512657165527
|
|
},
|
|
"correction_min": 0.9978846907615662,
|
|
"correction_max": 1.0082582235336304,
|
|
"codebook": [
|
|
-0.9769112467765808,
|
|
-0.7797420024871826,
|
|
-0.6185195446014404,
|
|
-0.4787740409374237,
|
|
-0.35444211959838867,
|
|
-0.24289147555828094,
|
|
-0.13915520906448364,
|
|
-0.040499910712242126,
|
|
0.055088989436626434,
|
|
0.15143312513828278,
|
|
0.2524990737438202,
|
|
0.3623327612876892,
|
|
0.483857661485672,
|
|
0.6214639544487,
|
|
0.7834796905517578,
|
|
0.9782072901725769
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9613595008850098,
|
|
"extreme_code_fraction": 0.09562772512435913,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.25
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020088020246475935,
|
|
"max_absolute_error": 0.009713888168334961,
|
|
"cosine_loss": {
|
|
"mean": 1.7232461004823563e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.843273639678955e-05,
|
|
"max": 2.658367156982422e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004257551918271929,
|
|
"p95": 0.0009309649467468262,
|
|
"p99": 0.004365578293800354,
|
|
"max": 0.006164371967315674
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.14100000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.30.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020321584306657314,
|
|
"max_absolute_error": 0.007509708404541016,
|
|
"cosine_loss": {
|
|
"mean": 1.7180794884552597e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8715858459472656e-05,
|
|
"max": 2.5391578674316406e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004627829766832292,
|
|
"p95": 0.0010295957326889038,
|
|
"p99": 0.004564300179481506,
|
|
"max": 0.007531642913818359
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020144071895629168,
|
|
"max_absolute_error": 0.009605884552001953,
|
|
"cosine_loss": {
|
|
"mean": 1.7481978602518211e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9727391190826893e-05,
|
|
"max": 2.676248550415039e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004298739368095994,
|
|
"p95": 0.0009850859642028809,
|
|
"p99": 0.0038463652599602938,
|
|
"max": 0.0067237019538879395
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21899999998277053
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012406249297782779,
|
|
"metrics": {
|
|
"relative_error": 0.07306524366140366,
|
|
"max_absolute_error": 0.1182861328125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026692962273955345,
|
|
"p95": 0.002742019249126315,
|
|
"p99": 0.002772008301690221,
|
|
"max": 0.002895653247833252
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004078919591847807,
|
|
"p95": 0.001030566869303584,
|
|
"p99": 0.0016216853400692344,
|
|
"max": 0.003520488739013672
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003073148662224412,
|
|
"p95": 0.005230301525443792,
|
|
"p99": 0.006183797027915716,
|
|
"max": 0.008941948413848877
|
|
},
|
|
"correction_min": 0.9974748492240906,
|
|
"correction_max": 1.00902259349823,
|
|
"codebook": [
|
|
-0.9782317280769348,
|
|
-0.7830264568328857,
|
|
-0.6196824312210083,
|
|
-0.4792953133583069,
|
|
-0.35550495982170105,
|
|
-0.24390457570552826,
|
|
-0.14040027558803558,
|
|
-0.04181209206581116,
|
|
0.05486529320478439,
|
|
0.1506640613079071,
|
|
0.25184962153434753,
|
|
0.3611058294773102,
|
|
0.48199817538261414,
|
|
0.6204051971435547,
|
|
0.782315194606781,
|
|
0.9782853722572327
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9608535766601562,
|
|
"extreme_code_fraction": 0.09537451714277267,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.25
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001989028649404645,
|
|
"max_absolute_error": 0.010844707489013672,
|
|
"cosine_loss": {
|
|
"mean": 1.6910601061681518e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.806020736694336e-05,
|
|
"max": 2.5033950805664062e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004131786117795855,
|
|
"p95": 0.0009268820285797119,
|
|
"p99": 0.004348725080490112,
|
|
"max": 0.006067991256713867
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.31.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002009335905313492,
|
|
"max_absolute_error": 0.010454148054122925,
|
|
"cosine_loss": {
|
|
"mean": 1.6560690028200042e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.640617847442627e-05,
|
|
"max": 2.6106834411621094e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00046990354894660413,
|
|
"p95": 0.0010421723127365112,
|
|
"p99": 0.004964083433151245,
|
|
"max": 0.006506800651550293
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.062000000005355105
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020046725403517485,
|
|
"max_absolute_error": 0.010899543762207031,
|
|
"cosine_loss": {
|
|
"mean": 1.7432280401408207e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9550323486328125e-05,
|
|
"max": 2.682209014892578e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004290579818189144,
|
|
"p95": 0.0009646057151257992,
|
|
"p99": 0.00392147246748209,
|
|
"max": 0.006552934646606445
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21899999998277053
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.00124881521333009,
|
|
"metrics": {
|
|
"relative_error": 0.07299942523241043,
|
|
"max_absolute_error": 0.208984375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026644407771527767,
|
|
"p95": 0.0027359724044799805,
|
|
"p99": 0.0027666438836604357,
|
|
"max": 0.002832353115081787
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00037328654434531927,
|
|
"p95": 0.0009703843388706446,
|
|
"p99": 0.0016140136867761612,
|
|
"max": 0.00342559814453125
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032489870209246874,
|
|
"p95": 0.005412610247731209,
|
|
"p99": 0.006319914944469929,
|
|
"max": 0.009790599346160889
|
|
},
|
|
"correction_min": 0.9958752393722534,
|
|
"correction_max": 1.009887456893921,
|
|
"codebook": [
|
|
-0.977198600769043,
|
|
-0.7779897451400757,
|
|
-0.6152374744415283,
|
|
-0.4759490489959717,
|
|
-0.3532382547855377,
|
|
-0.2429375797510147,
|
|
-0.13967777788639069,
|
|
-0.04127717390656471,
|
|
0.055094070732593536,
|
|
0.1513012796640396,
|
|
0.2518676519393921,
|
|
0.3604922592639923,
|
|
0.481608510017395,
|
|
0.6188541650772095,
|
|
0.7809634208679199,
|
|
0.9777427911758423
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.962972640991211,
|
|
"extreme_code_fraction": 0.09604103118181229,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.26500000001396984
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002005381044000387,
|
|
"max_absolute_error": 0.0178682804107666,
|
|
"cosine_loss": {
|
|
"mean": 1.6931887785176514e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.730024814605713e-05,
|
|
"max": 2.4497509002685547e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042626369395293295,
|
|
"p95": 0.0009289383888244629,
|
|
"p99": 0.0044654905796051025,
|
|
"max": 0.006574809551239014
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.14100000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.32.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019615874625742435,
|
|
"max_absolute_error": 0.007788658142089844,
|
|
"cosine_loss": {
|
|
"mean": 1.6345044286936172e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.728534698486328e-05,
|
|
"max": 2.7477741241455078e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004392964474391192,
|
|
"p95": 0.0009876489639282227,
|
|
"p99": 0.004130616784095764,
|
|
"max": 0.0065577030181884766
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002002844586968422,
|
|
"max_absolute_error": 0.007723212242126465,
|
|
"cosine_loss": {
|
|
"mean": 1.7319911194135784e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8835067749023438e-05,
|
|
"max": 2.7298927307128906e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042964552994817495,
|
|
"p95": 0.0009829280897974968,
|
|
"p99": 0.003966827876865864,
|
|
"max": 0.006191670894622803
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.2029999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001250106724910438,
|
|
"metrics": {
|
|
"relative_error": 0.07310451567173004,
|
|
"max_absolute_error": 0.1982421875,
|
|
"cosine_loss": {
|
|
"mean": 0.0026724105700850487,
|
|
"p95": 0.002744615077972412,
|
|
"p99": 0.0027745538391172886,
|
|
"max": 0.0028365254402160645
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00038200838025659323,
|
|
"p95": 0.0009780344553291798,
|
|
"p99": 0.0016137708444148302,
|
|
"max": 0.0036393404006958008
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032103604171425104,
|
|
"p95": 0.005329722072929144,
|
|
"p99": 0.006213197950273752,
|
|
"max": 0.009481489658355713
|
|
},
|
|
"correction_min": 0.9978947043418884,
|
|
"correction_max": 1.0095722675323486,
|
|
"codebook": [
|
|
-0.9768491387367249,
|
|
-0.7758722305297852,
|
|
-0.6136537790298462,
|
|
-0.4760285019874573,
|
|
-0.35416653752326965,
|
|
-0.24455974996089935,
|
|
-0.14151625335216522,
|
|
-0.041733283549547195,
|
|
0.055084604769945145,
|
|
0.1520918905735016,
|
|
0.253745973110199,
|
|
0.3631852865219116,
|
|
0.4842439591884613,
|
|
0.6221279501914978,
|
|
0.7837051153182983,
|
|
0.9779179096221924
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.962277412414551,
|
|
"extreme_code_fraction": 0.09596065431833267,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.26600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019841091707348824,
|
|
"max_absolute_error": 0.010486602783203125,
|
|
"cosine_loss": {
|
|
"mean": 1.6824563999762177e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7657876014709473e-05,
|
|
"max": 2.3603439331054688e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000409879517974332,
|
|
"p95": 0.0008910000324249268,
|
|
"p99": 0.004261046648025513,
|
|
"max": 0.006205737590789795
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.15600000001722947
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.33.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0159999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002017530845478177,
|
|
"max_absolute_error": 0.00927269458770752,
|
|
"cosine_loss": {
|
|
"mean": 1.6838533838381409e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.6927719116210938e-05,
|
|
"max": 2.5153160095214844e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004625408328138292,
|
|
"p95": 0.0010350346565246582,
|
|
"p99": 0.004589900374412537,
|
|
"max": 0.00713956356048584
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.062000000005355105
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001993941841647029,
|
|
"max_absolute_error": 0.015691518783569336,
|
|
"cosine_loss": {
|
|
"mean": 1.7262146911889431e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.919269561767578e-05,
|
|
"max": 2.5451183319091797e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004280403954908252,
|
|
"p95": 0.000974839786067605,
|
|
"p99": 0.0038635157980024815,
|
|
"max": 0.00628209114074707
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21900000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001247832435183227,
|
|
"metrics": {
|
|
"relative_error": 0.07299865037202835,
|
|
"max_absolute_error": 0.206298828125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026646938640624285,
|
|
"p95": 0.002736833645030856,
|
|
"p99": 0.0027672399301081896,
|
|
"max": 0.002850472927093506
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000369483808754012,
|
|
"p95": 0.0009517013095319271,
|
|
"p99": 0.001593582215718925,
|
|
"max": 0.0033713579177856445
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003005464095622301,
|
|
"p95": 0.005180459935218096,
|
|
"p99": 0.0061152600683271885,
|
|
"max": 0.010274827480316162
|
|
},
|
|
"correction_min": 0.996484100818634,
|
|
"correction_max": 1.0103814601898193,
|
|
"codebook": [
|
|
-0.9771851897239685,
|
|
-0.778667688369751,
|
|
-0.6175464391708374,
|
|
-0.4780702590942383,
|
|
-0.3571279048919678,
|
|
-0.24908684194087982,
|
|
-0.149146169424057,
|
|
-0.05436915159225464,
|
|
0.04252328351140022,
|
|
0.14264236390590668,
|
|
0.24694983661174774,
|
|
0.3584011495113373,
|
|
0.4806990623474121,
|
|
0.6199353933334351,
|
|
0.7824805378913879,
|
|
0.9775183200836182
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961390256881714,
|
|
"extreme_code_fraction": 0.0958012193441391,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.264999999984866
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020407449919730425,
|
|
"max_absolute_error": 0.009943962097167969,
|
|
"cosine_loss": {
|
|
"mean": 1.7592240055819275e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.849234104156494e-05,
|
|
"max": 2.4616718292236328e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004290846991352737,
|
|
"p95": 0.0009341537952423096,
|
|
"p99": 0.004502341151237488,
|
|
"max": 0.0067768096923828125
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.14100000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.34.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001984329428523779,
|
|
"max_absolute_error": 0.009370803833007812,
|
|
"cosine_loss": {
|
|
"mean": 1.6725111890991684e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7583370208740234e-05,
|
|
"max": 2.6047229766845703e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00045389586011879146,
|
|
"p95": 0.0010006725788116455,
|
|
"p99": 0.0043983012437820435,
|
|
"max": 0.006352424621582031
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.00200832262635231,
|
|
"max_absolute_error": 0.009335756301879883,
|
|
"cosine_loss": {
|
|
"mean": 1.7453650116294739e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.931015867739916e-05,
|
|
"max": 2.8252601623535156e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043161900248378515,
|
|
"p95": 0.000977241899818182,
|
|
"p99": 0.0038881460204720497,
|
|
"max": 0.006905555725097656
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21899999998277053
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012483905302360654,
|
|
"metrics": {
|
|
"relative_error": 0.07307074218988419,
|
|
"max_absolute_error": 0.1690673828125,
|
|
"cosine_loss": {
|
|
"mean": 0.002670415909960866,
|
|
"p95": 0.0027428865432739258,
|
|
"p99": 0.0027725272811949253,
|
|
"max": 0.0028429627418518066
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004106044361833483,
|
|
"p95": 0.0010442733764648438,
|
|
"p99": 0.001684037153609097,
|
|
"max": 0.0034829378128051758
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003184275468811393,
|
|
"p95": 0.00538279116153717,
|
|
"p99": 0.006343028973788023,
|
|
"max": 0.010256767272949219
|
|
},
|
|
"correction_min": 0.9967244267463684,
|
|
"correction_max": 1.0103631019592285,
|
|
"codebook": [
|
|
-0.9786864519119263,
|
|
-0.782408595085144,
|
|
-0.6203655004501343,
|
|
-0.4818713366985321,
|
|
-0.3611769676208496,
|
|
-0.2523319721221924,
|
|
-0.15149684250354767,
|
|
-0.05502648651599884,
|
|
0.040913935750722885,
|
|
0.1398293673992157,
|
|
0.24446256458759308,
|
|
0.3570542335510254,
|
|
0.48107171058654785,
|
|
0.6223275065422058,
|
|
0.7845889329910278,
|
|
0.9782666563987732
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.96038818359375,
|
|
"extreme_code_fraction": 0.09525423496961594,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.26500000001396984
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019508793484419584,
|
|
"max_absolute_error": 0.011107921600341797,
|
|
"cosine_loss": {
|
|
"mean": 1.6408351939389831e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.683831214904785e-05,
|
|
"max": 2.6047229766845703e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004044708330184221,
|
|
"p95": 0.0009187012910842896,
|
|
"p99": 0.00415271520614624,
|
|
"max": 0.0063364505767822266
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.35.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.00196879543364048,
|
|
"max_absolute_error": 0.010997295379638672,
|
|
"cosine_loss": {
|
|
"mean": 1.6238052467087982e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.582503318786621e-05,
|
|
"max": 2.6404857635498047e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004469958948902786,
|
|
"p95": 0.0010022521018981934,
|
|
"p99": 0.004504352807998657,
|
|
"max": 0.006867766380310059
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.0629999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001986699877306819,
|
|
"max_absolute_error": 0.012770652770996094,
|
|
"cosine_loss": {
|
|
"mean": 1.7033918311426532e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.895427703857422e-05,
|
|
"max": 2.562999725341797e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004249975027050823,
|
|
"p95": 0.0009715467458590865,
|
|
"p99": 0.003826211905106902,
|
|
"max": 0.006167948246002197
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.20300000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012450780486688018,
|
|
"metrics": {
|
|
"relative_error": 0.07303226739168167,
|
|
"max_absolute_error": 0.17578125,
|
|
"cosine_loss": {
|
|
"mean": 0.002667494583874941,
|
|
"p95": 0.002740144729614258,
|
|
"p99": 0.00277033937163651,
|
|
"max": 0.0028433799743652344
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003950742247980088,
|
|
"p95": 0.0010308856144547462,
|
|
"p99": 0.001713178469799459,
|
|
"max": 0.0035518407821655273
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003246934385970235,
|
|
"p95": 0.005517232231795788,
|
|
"p99": 0.006492559798061848,
|
|
"max": 0.009613215923309326
|
|
},
|
|
"correction_min": 0.9961419701576233,
|
|
"correction_max": 1.0097064971923828,
|
|
"codebook": [
|
|
-0.9777414202690125,
|
|
-0.7794392704963684,
|
|
-0.6161606311798096,
|
|
-0.4770963490009308,
|
|
-0.3552229404449463,
|
|
-0.24383842945098877,
|
|
-0.14032109081745148,
|
|
-0.04146174341440201,
|
|
0.054458022117614746,
|
|
0.15024656057357788,
|
|
0.250999391078949,
|
|
0.36059755086898804,
|
|
0.48197147250175476,
|
|
0.6210299730300903,
|
|
0.7842826843261719,
|
|
0.9789801836013794
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9619481563568115,
|
|
"extreme_code_fraction": 0.0954987034201622,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.28099999998812564
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020511415787041187,
|
|
"max_absolute_error": 0.010866761207580566,
|
|
"cosine_loss": {
|
|
"mean": 1.76319326783414e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7777085304260254e-05,
|
|
"max": 2.5391578674316406e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004430787521414459,
|
|
"p95": 0.0009497255086898804,
|
|
"p99": 0.004555940628051758,
|
|
"max": 0.0061356425285339355
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.36.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019426231738179922,
|
|
"max_absolute_error": 0.00911569595336914,
|
|
"cosine_loss": {
|
|
"mean": 1.5869514982114197e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.5094876289367676e-05,
|
|
"max": 2.4437904357910156e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00045506961760111153,
|
|
"p95": 0.0010185390710830688,
|
|
"p99": 0.0045824795961380005,
|
|
"max": 0.00682908296585083
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.062000000005355105
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020016981288790703,
|
|
"max_absolute_error": 0.009991168975830078,
|
|
"cosine_loss": {
|
|
"mean": 1.7495698330094456e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9667786546051502e-05,
|
|
"max": 2.7000904083251953e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043095179717056453,
|
|
"p95": 0.0009715198539197445,
|
|
"p99": 0.003895312547683716,
|
|
"max": 0.006824493408203125
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21899999998277053
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012453810777515173,
|
|
"metrics": {
|
|
"relative_error": 0.07295717298984528,
|
|
"max_absolute_error": 0.1787109375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026618733536452055,
|
|
"p95": 0.0027326345443725586,
|
|
"p99": 0.0027641654014587402,
|
|
"max": 0.002842247486114502
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00038376296288333833,
|
|
"p95": 0.000995331909507513,
|
|
"p99": 0.0016237390227615833,
|
|
"max": 0.003565371036529541
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031919502653181553,
|
|
"p95": 0.00542376609519124,
|
|
"p99": 0.006423306651413441,
|
|
"max": 0.009538769721984863
|
|
},
|
|
"correction_min": 0.9972690343856812,
|
|
"correction_max": 1.0096306800842285,
|
|
"codebook": [
|
|
-0.9776356220245361,
|
|
-0.7812257409095764,
|
|
-0.619156539440155,
|
|
-0.4800339937210083,
|
|
-0.3559003472328186,
|
|
-0.2436470091342926,
|
|
-0.1394958794116974,
|
|
-0.04109684377908707,
|
|
0.05479571595788002,
|
|
0.15031175315380096,
|
|
0.25119173526763916,
|
|
0.36016228795051575,
|
|
0.480923593044281,
|
|
0.6182668805122375,
|
|
0.779725193977356,
|
|
0.9778404831886292
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961549758911133,
|
|
"extreme_code_fraction": 0.09583096206188202,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.26500000001396984
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019514954183250666,
|
|
"max_absolute_error": 0.01793384552001953,
|
|
"cosine_loss": {
|
|
"mean": 1.653252866162802e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7121434211730957e-05,
|
|
"max": 2.47955322265625e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000405794067773968,
|
|
"p95": 0.0008865445852279663,
|
|
"p99": 0.004166126251220703,
|
|
"max": 0.006094992160797119
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.14100000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.37.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020385810639709234,
|
|
"max_absolute_error": 0.011425495147705078,
|
|
"cosine_loss": {
|
|
"mean": 1.6345709354936844e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.4856457710266113e-05,
|
|
"max": 2.771615982055664e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004910981515422463,
|
|
"p95": 0.0010852813720703125,
|
|
"p99": 0.005380481481552124,
|
|
"max": 0.007123768329620361
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002012653509154916,
|
|
"max_absolute_error": 0.01897907257080078,
|
|
"cosine_loss": {
|
|
"mean": 1.7492621964265709e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9431114196777344e-05,
|
|
"max": 2.771615982055664e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004309646610636264,
|
|
"p95": 0.0009828447364270687,
|
|
"p99": 0.0038724893238395452,
|
|
"max": 0.006483197212219238
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21899999998277053
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012444627936929464,
|
|
"metrics": {
|
|
"relative_error": 0.07299815863370895,
|
|
"max_absolute_error": 0.125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026645187754184008,
|
|
"p95": 0.0027367472648620605,
|
|
"p99": 0.0027663458604365587,
|
|
"max": 0.0028344392776489258
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003807014727499336,
|
|
"p95": 0.0009929235093295574,
|
|
"p99": 0.001638387911953032,
|
|
"max": 0.003461599349975586
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003204318229109049,
|
|
"p95": 0.0054106321185827255,
|
|
"p99": 0.006342733278870583,
|
|
"max": 0.008667290210723877
|
|
},
|
|
"correction_min": 0.9969305396080017,
|
|
"correction_max": 1.0087430477142334,
|
|
"codebook": [
|
|
-0.97697514295578,
|
|
-0.7805021405220032,
|
|
-0.6179852485656738,
|
|
-0.48084595799446106,
|
|
-0.3602040112018585,
|
|
-0.251314640045166,
|
|
-0.15024788677692413,
|
|
-0.054757580161094666,
|
|
0.04155717045068741,
|
|
0.1410692185163498,
|
|
0.2451775223016739,
|
|
0.35687699913978577,
|
|
0.47887203097343445,
|
|
0.6168984770774841,
|
|
0.7793774604797363,
|
|
0.9778100848197937
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9620449542999268,
|
|
"extreme_code_fraction": 0.09587403386831284,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.25
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020102267153561115,
|
|
"max_absolute_error": 0.014111757278442383,
|
|
"cosine_loss": {
|
|
"mean": 1.717746954454924e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.837313175201416e-05,
|
|
"max": 2.47955322265625e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004146473074797541,
|
|
"p95": 0.0009118914604187012,
|
|
"p99": 0.004362508654594421,
|
|
"max": 0.006416380405426025
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.14100000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.38.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019736825488507748,
|
|
"max_absolute_error": 0.017963886260986328,
|
|
"cosine_loss": {
|
|
"mean": 1.6264107216557022e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.6883015632629395e-05,
|
|
"max": 3.1888484954833984e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004421180346980691,
|
|
"p95": 0.0009904950857162476,
|
|
"p99": 0.004382804036140442,
|
|
"max": 0.00667649507522583
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.062000000005355105
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019942456856369972,
|
|
"max_absolute_error": 0.014349937438964844,
|
|
"cosine_loss": {
|
|
"mean": 1.7220487507074722e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9131344743072987e-05,
|
|
"max": 2.8967857360839844e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004279866407159716,
|
|
"p95": 0.0009689985308796167,
|
|
"p99": 0.0038762050680816174,
|
|
"max": 0.006326198577880859
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.2029999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001243667327798903,
|
|
"metrics": {
|
|
"relative_error": 0.07307980954647064,
|
|
"max_absolute_error": 0.14501953125,
|
|
"cosine_loss": {
|
|
"mean": 0.002670579357072711,
|
|
"p95": 0.0027436285745352507,
|
|
"p99": 0.0027746905107051134,
|
|
"max": 0.0028513669967651367
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003629142011050135,
|
|
"p95": 0.0009535308927297592,
|
|
"p99": 0.0016326147597283125,
|
|
"max": 0.003648042678833008
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.002952531212940812,
|
|
"p95": 0.0051865605637431145,
|
|
"p99": 0.0062077948823571205,
|
|
"max": 0.008977413177490234
|
|
},
|
|
"correction_min": 0.9958015084266663,
|
|
"correction_max": 1.0090587139129639,
|
|
"codebook": [
|
|
-0.9795480966567993,
|
|
-0.7862575650215149,
|
|
-0.6232652068138123,
|
|
-0.48442503809928894,
|
|
-0.3623507022857666,
|
|
-0.25260624289512634,
|
|
-0.15194781124591827,
|
|
-0.05497950688004494,
|
|
0.04178842529654503,
|
|
0.14108815789222717,
|
|
0.24469447135925293,
|
|
0.35549256205558777,
|
|
0.47779273986816406,
|
|
0.6166594624519348,
|
|
0.7789222002029419,
|
|
0.9772070646286011
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9614598751068115,
|
|
"extreme_code_fraction": 0.09521685540676117,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.26500000001396984
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019707416649907827,
|
|
"max_absolute_error": 0.009639739990234375,
|
|
"cosine_loss": {
|
|
"mean": 1.6669787328282837e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7777085304260254e-05,
|
|
"max": 2.5391578674316406e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000413017172832042,
|
|
"p95": 0.0009298175573348999,
|
|
"p99": 0.004203811287879944,
|
|
"max": 0.006374835968017578
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.39.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012464046012610197,
|
|
"metrics": {
|
|
"relative_error": 0.07315001636743546,
|
|
"max_absolute_error": 0.15611648559570312,
|
|
"cosine_loss": {
|
|
"mean": 0.0026744073256850243,
|
|
"p95": 0.0027467161417007446,
|
|
"p99": 0.0027779191732406616,
|
|
"max": 0.002866804599761963
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003710661258082837,
|
|
"p95": 0.0009708702564239502,
|
|
"p99": 0.0016870051622390747,
|
|
"max": 0.0032552480697631836
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003279553260654211,
|
|
"p95": 0.005474358797073364,
|
|
"p99": 0.006536856293678284,
|
|
"max": 0.010082483291625977
|
|
},
|
|
"correction_min": 0.9945154190063477,
|
|
"correction_max": 1.0101852416992188,
|
|
"codebook": [
|
|
-0.9781378507614136,
|
|
-0.7839423418045044,
|
|
-0.6227880716323853,
|
|
-0.4839661419391632,
|
|
-0.3628697395324707,
|
|
-0.25372743606567383,
|
|
-0.15326912701129913,
|
|
-0.056723788380622864,
|
|
0.04007958248257637,
|
|
0.13886047899723053,
|
|
0.2424309253692627,
|
|
0.3543403148651123,
|
|
0.4759240746498108,
|
|
0.6146320104598999,
|
|
0.7776326537132263,
|
|
0.976752758026123
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9618642330169678,
|
|
"extreme_code_fraction": 0.09594316780567169,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.09399999998277053
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020035908091813326,
|
|
"max_absolute_error": 0.011018753051757812,
|
|
"cosine_loss": {
|
|
"mean": 1.7169791135529522e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8952530808746815e-05,
|
|
"max": 2.771615982055664e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004248654586263001,
|
|
"p95": 0.0009627637919038534,
|
|
"p99": 0.0039007626473903656,
|
|
"max": 0.006585419178009033
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21900000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012436609249562025,
|
|
"metrics": {
|
|
"relative_error": 0.07303894311189651,
|
|
"max_absolute_error": 0.08447265625,
|
|
"cosine_loss": {
|
|
"mean": 0.0026672666426748037,
|
|
"p95": 0.0027396350633352995,
|
|
"p99": 0.0027697086334228516,
|
|
"max": 0.0028595924377441406
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003851433575619012,
|
|
"p95": 0.0009902059100568295,
|
|
"p99": 0.0016250471817329526,
|
|
"max": 0.0033261775970458984
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003166851820424199,
|
|
"p95": 0.00544940959662199,
|
|
"p99": 0.006430481560528278,
|
|
"max": 0.009563028812408447
|
|
},
|
|
"correction_min": 0.9967681169509888,
|
|
"correction_max": 1.0096553564071655,
|
|
"codebook": [
|
|
-0.9776038527488708,
|
|
-0.7807790040969849,
|
|
-0.6184500455856323,
|
|
-0.4815872013568878,
|
|
-0.3610677123069763,
|
|
-0.25330308079719543,
|
|
-0.15202292799949646,
|
|
-0.055170901119709015,
|
|
0.04223569482564926,
|
|
0.14201021194458008,
|
|
0.24476800858974457,
|
|
0.3552801012992859,
|
|
0.47805124521255493,
|
|
0.6169023513793945,
|
|
0.7791463136672974,
|
|
0.9768280982971191
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9618101119995117,
|
|
"extreme_code_fraction": 0.09586209803819656,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.26500000001396984
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001990499207749963,
|
|
"max_absolute_error": 0.007176041603088379,
|
|
"cosine_loss": {
|
|
"mean": 1.6906387827475555e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7121434211730957e-05,
|
|
"max": 2.2709369659423828e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004155316564720124,
|
|
"p95": 0.0009182989597320557,
|
|
"p99": 0.004466459155082703,
|
|
"max": 0.006058394908905029
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.1409999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.4.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001968046184629202,
|
|
"max_absolute_error": 0.008983254432678223,
|
|
"cosine_loss": {
|
|
"mean": 1.6130728681673645e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.615285873413086e-05,
|
|
"max": 2.2649765014648438e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004489065904635936,
|
|
"p95": 0.0009943395853042603,
|
|
"p99": 0.004447266459465027,
|
|
"max": 0.007030129432678223
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019669239409267902,
|
|
"max_absolute_error": 0.009577155113220215,
|
|
"cosine_loss": {
|
|
"mean": 1.689524538051046e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8892926163971424e-05,
|
|
"max": 2.6285648345947266e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042107675108127296,
|
|
"p95": 0.0009646236430853605,
|
|
"p99": 0.003821098245680332,
|
|
"max": 0.006686985492706299
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21900000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012415909441187978,
|
|
"metrics": {
|
|
"relative_error": 0.07304056733846664,
|
|
"max_absolute_error": 0.1318359375,
|
|
"cosine_loss": {
|
|
"mean": 0.0026676610577851534,
|
|
"p95": 0.00274050235748291,
|
|
"p99": 0.0027710546273738146,
|
|
"max": 0.0028508901596069336
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000398894859245047,
|
|
"p95": 0.0010315326508134604,
|
|
"p99": 0.0016814785776659846,
|
|
"max": 0.0035026073455810547
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031234989874064922,
|
|
"p95": 0.005360206589102745,
|
|
"p99": 0.006327176466584206,
|
|
"max": 0.008907318115234375
|
|
},
|
|
"correction_min": 0.9969831109046936,
|
|
"correction_max": 1.0089874267578125,
|
|
"codebook": [
|
|
-0.9783331155776978,
|
|
-0.7806264162063599,
|
|
-0.6207178235054016,
|
|
-0.48334774374961853,
|
|
-0.3614010512828827,
|
|
-0.25271713733673096,
|
|
-0.1512456238269806,
|
|
-0.054828036576509476,
|
|
0.04184975475072861,
|
|
0.14142847061157227,
|
|
0.24499298632144928,
|
|
0.35716015100479126,
|
|
0.4802519977092743,
|
|
0.6189593076705933,
|
|
0.7807095050811768,
|
|
0.9776467084884644
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9605650901794434,
|
|
"extreme_code_fraction": 0.09581725299358368,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.28099999998812564
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019305940950289369,
|
|
"max_absolute_error": 0.013215065002441406,
|
|
"cosine_loss": {
|
|
"mean": 1.6163658074219711e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.621246337890625e-05,
|
|
"max": 2.562999725341797e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004013652796857059,
|
|
"p95": 0.0009004175662994385,
|
|
"p99": 0.004192844033241272,
|
|
"max": 0.006117403507232666
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.40.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.00196996727026999,
|
|
"max_absolute_error": 0.016460895538330078,
|
|
"cosine_loss": {
|
|
"mean": 1.6127403341670288e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.5333294868469238e-05,
|
|
"max": 2.3543834686279297e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00044614242506213486,
|
|
"p95": 0.001005902886390686,
|
|
"p99": 0.004525139927864075,
|
|
"max": 0.006640613079071045
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.062000000005355105
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001993794459849596,
|
|
"max_absolute_error": 0.010920286178588867,
|
|
"cosine_loss": {
|
|
"mean": 1.715060989226913e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9071740098297596e-05,
|
|
"max": 2.5987625122070312e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004268398624844849,
|
|
"p95": 0.0009680237853899598,
|
|
"p99": 0.0038722525350749493,
|
|
"max": 0.006326198577880859
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 2.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012516810093075037,
|
|
"metrics": {
|
|
"relative_error": 0.07305211573839188,
|
|
"max_absolute_error": 0.1728515625,
|
|
"cosine_loss": {
|
|
"mean": 0.002668574219569564,
|
|
"p95": 0.002741217613220215,
|
|
"p99": 0.002773063722997904,
|
|
"max": 0.002848803997039795
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003884465841110796,
|
|
"p95": 0.0009865639731287956,
|
|
"p99": 0.001655498519539833,
|
|
"max": 0.003571808338165283
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031577718909829855,
|
|
"p95": 0.005357325077056885,
|
|
"p99": 0.0063515305519104,
|
|
"max": 0.008494555950164795
|
|
},
|
|
"correction_min": 0.9979194402694702,
|
|
"correction_max": 1.0085673332214355,
|
|
"codebook": [
|
|
-0.9780268669128418,
|
|
-0.7814483642578125,
|
|
-0.6186333894729614,
|
|
-0.47907555103302,
|
|
-0.3559096157550812,
|
|
-0.24348606169223785,
|
|
-0.1405736207962036,
|
|
-0.04131673276424408,
|
|
0.05465918779373169,
|
|
0.15000052750110626,
|
|
0.25074538588523865,
|
|
0.35963737964630127,
|
|
0.4811890125274658,
|
|
0.620879590511322,
|
|
0.7847096920013428,
|
|
0.9782000184059143
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.96142578125,
|
|
"extreme_code_fraction": 0.09528116881847382,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 1.1559999999881256
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019922407809644938,
|
|
"max_absolute_error": 0.010066986083984375,
|
|
"cosine_loss": {
|
|
"mean": 1.7121590190072311e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8775463104248047e-05,
|
|
"max": 2.5928020477294922e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004139709344599396,
|
|
"p95": 0.0009106248617172241,
|
|
"p99": 0.004168331623077393,
|
|
"max": 0.006630837917327881
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.7820000000065193
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.41.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001973547041416168,
|
|
"max_absolute_error": 0.011379122734069824,
|
|
"cosine_loss": {
|
|
"mean": 1.6136494878082885e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.5214085578918457e-05,
|
|
"max": 2.6047229766845703e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00045583717292174697,
|
|
"p95": 0.0010171383619308472,
|
|
"p99": 0.004529222846031189,
|
|
"max": 0.006952047348022461
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.2960000000020955
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020045265555381775,
|
|
"max_absolute_error": 0.01162409782409668,
|
|
"cosine_loss": {
|
|
"mean": 1.7298680177191272e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9371509552001953e-05,
|
|
"max": 2.86102294921875e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042751897126436234,
|
|
"p95": 0.0009714185725897551,
|
|
"p99": 0.0037836660631000996,
|
|
"max": 0.006380796432495117
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 1.4070000000065193
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001244814833626151,
|
|
"metrics": {
|
|
"relative_error": 0.07305724918842316,
|
|
"max_absolute_error": 0.2197265625,
|
|
"cosine_loss": {
|
|
"mean": 0.0026689597871154547,
|
|
"p95": 0.0027417540550231934,
|
|
"p99": 0.0027729272842407227,
|
|
"max": 0.002838611602783203
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00041171134216710925,
|
|
"p95": 0.0010456321761012077,
|
|
"p99": 0.0016811414388939738,
|
|
"max": 0.0034797191619873047
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003002099459990859,
|
|
"p95": 0.005162449553608894,
|
|
"p99": 0.006070065777748823,
|
|
"max": 0.008185744285583496
|
|
},
|
|
"correction_min": 0.9974595308303833,
|
|
"correction_max": 1.0082533359527588,
|
|
"codebook": [
|
|
-0.9778673648834229,
|
|
-0.7831031680107117,
|
|
-0.6213921308517456,
|
|
-0.4829126000404358,
|
|
-0.3608231842517853,
|
|
-0.2521478235721588,
|
|
-0.15156252682209015,
|
|
-0.055112335830926895,
|
|
0.041730910539627075,
|
|
0.14115463197231293,
|
|
0.24605007469654083,
|
|
0.3577843904495239,
|
|
0.4797259569168091,
|
|
0.6193499565124512,
|
|
0.7817577123641968,
|
|
0.978272557258606
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9604456424713135,
|
|
"extreme_code_fraction": 0.09542959928512573,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 1.139999999984866
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020350466948002577,
|
|
"max_absolute_error": 0.01613473892211914,
|
|
"cosine_loss": {
|
|
"mean": 1.743702000567282e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8268823623657227e-05,
|
|
"max": 2.777576446533203e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004350693488959223,
|
|
"p95": 0.0009554922580718994,
|
|
"p99": 0.00465141236782074,
|
|
"max": 0.006248354911804199
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 2.062999999994645
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.42.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020307572558522224,
|
|
"max_absolute_error": 0.012030959129333496,
|
|
"cosine_loss": {
|
|
"mean": 1.646722466830397e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.583993434906006e-05,
|
|
"max": 2.47955322265625e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00048525078454986215,
|
|
"p95": 0.0010824501514434814,
|
|
"p99": 0.005167186260223389,
|
|
"max": 0.00676959753036499
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3129999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001990760676562786,
|
|
"max_absolute_error": 0.016553401947021484,
|
|
"cosine_loss": {
|
|
"mean": 1.7200059119204525e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9252300262451172e-05,
|
|
"max": 2.7418136596679688e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004221780982334167,
|
|
"p95": 0.0009693561587482691,
|
|
"p99": 0.0037284879945218563,
|
|
"max": 0.0060669779777526855
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 1.375
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012469219509512186,
|
|
"metrics": {
|
|
"relative_error": 0.07298513501882553,
|
|
"max_absolute_error": 0.23828125,
|
|
"cosine_loss": {
|
|
"mean": 0.0026639753486961126,
|
|
"p95": 0.002736985683441162,
|
|
"p99": 0.0027675628662109375,
|
|
"max": 0.002884387969970703
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003967649827245623,
|
|
"p95": 0.0010186433792114258,
|
|
"p99": 0.0016512568108737469,
|
|
"max": 0.0037592053413391113
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003054032800719142,
|
|
"p95": 0.0053132278844714165,
|
|
"p99": 0.006276192609220743,
|
|
"max": 0.00874471664428711
|
|
},
|
|
"correction_min": 0.9970909357070923,
|
|
"correction_max": 1.0088218450546265,
|
|
"codebook": [
|
|
-0.9777928590774536,
|
|
-0.7812478542327881,
|
|
-0.6201498508453369,
|
|
-0.4800931513309479,
|
|
-0.3562115430831909,
|
|
-0.24360288679599762,
|
|
-0.1401202231645584,
|
|
-0.04069075733423233,
|
|
0.05483170971274376,
|
|
0.1500106006860733,
|
|
0.25119444727897644,
|
|
0.3611961007118225,
|
|
0.4814349114894867,
|
|
0.6190650463104248,
|
|
0.7807448506355286,
|
|
0.9771194458007812
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961092472076416,
|
|
"extreme_code_fraction": 0.09583647549152374,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 1.0940000000118744
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002029067138209939,
|
|
"max_absolute_error": 0.013540267944335938,
|
|
"cosine_loss": {
|
|
"mean": 1.7284349951296463e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.831352710723877e-05,
|
|
"max": 2.5272369384765625e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042803751421160996,
|
|
"p95": 0.0009276270866394043,
|
|
"p99": 0.004535168409347534,
|
|
"max": 0.006413817405700684
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.7969999999913853
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.43.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 77070336,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.9379999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 231211008,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 3.4059999999881256
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 308281344,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 3.8589999999967404
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 154140672,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 3.8440000000118744
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.44.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 77070336,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.7339999999967404
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 231211008,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.23499999998603016
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 308281344,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.3120000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 154140672,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.15700000000651926
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.45.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 77070336,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 231211008,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.23399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 308281344,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.3129999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 154140672,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.15599999998812564
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.46.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 77070336,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 231211008,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.23399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 308281344,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.3129999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 154140672,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.15600000001722947
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.47.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 77070336,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0779999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 231211008,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.28100000001722947
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 308281344,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 154140672,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.1409999999741558
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.48.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.015000000013969839
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 77070336,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0629999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 231211008,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.23399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 308281344,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 154140672,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.17199999999138527
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.49.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012500437442213297,
|
|
"metrics": {
|
|
"relative_error": 0.07318248599767685,
|
|
"max_absolute_error": 0.13671875,
|
|
"cosine_loss": {
|
|
"mean": 0.0026760390028357506,
|
|
"p95": 0.00275285542011261,
|
|
"p99": 0.00280722975730896,
|
|
"max": 0.003040611743927002
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003667028504423797,
|
|
"p95": 0.000975877046585083,
|
|
"p99": 0.0016880631446838379,
|
|
"max": 0.0035460591316223145
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0033163779880851507,
|
|
"p95": 0.005777254700660706,
|
|
"p99": 0.007030755281448364,
|
|
"max": 0.011080801486968994
|
|
},
|
|
"correction_min": 0.994564950466156,
|
|
"correction_max": 1.0112049579620361,
|
|
"codebook": [
|
|
-0.9772642254829407,
|
|
-0.7790042161941528,
|
|
-0.6151407957077026,
|
|
-0.4779094457626343,
|
|
-0.3578839600086212,
|
|
-0.25041818618774414,
|
|
-0.1515365093946457,
|
|
-0.056646838784217834,
|
|
0.038937605917453766,
|
|
0.1386009156703949,
|
|
0.24302421510219574,
|
|
0.35498446226119995,
|
|
0.4788782596588135,
|
|
0.6168863773345947,
|
|
0.7798320055007935,
|
|
0.9773619174957275
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.963097333908081,
|
|
"extreme_code_fraction": 0.09626616537570953,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.28200000000651926
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020201411098241806,
|
|
"max_absolute_error": 0.006856203079223633,
|
|
"cosine_loss": {
|
|
"mean": 1.738787545946252e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9727391190826893e-05,
|
|
"max": 2.6464462280273438e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00043354759691283107,
|
|
"p95": 0.0009784519206732512,
|
|
"p99": 0.004009914118796587,
|
|
"max": 0.006324887275695801
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21799999999348074
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001248552929610014,
|
|
"metrics": {
|
|
"relative_error": 0.07299058139324188,
|
|
"max_absolute_error": 0.08203125,
|
|
"cosine_loss": {
|
|
"mean": 0.00266367313452065,
|
|
"p95": 0.002734959125518799,
|
|
"p99": 0.0027641654014587402,
|
|
"max": 0.0028316378593444824
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003723015252035111,
|
|
"p95": 0.0009634641464799643,
|
|
"p99": 0.0015920097939670086,
|
|
"max": 0.0034276247024536133
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0031614082399755716,
|
|
"p95": 0.005406289827078581,
|
|
"p99": 0.006389055401086807,
|
|
"max": 0.009225726127624512
|
|
},
|
|
"correction_min": 0.9976897835731506,
|
|
"correction_max": 1.0093116760253906,
|
|
"codebook": [
|
|
-0.9772158265113831,
|
|
-0.781013011932373,
|
|
-0.61967533826828,
|
|
-0.48108962178230286,
|
|
-0.35896146297454834,
|
|
-0.24980200827121735,
|
|
-0.14943787455558777,
|
|
-0.054330650717020035,
|
|
0.041299186646938324,
|
|
0.14018811285495758,
|
|
0.24372105300426483,
|
|
0.35527628660202026,
|
|
0.4775860011577606,
|
|
0.6180062890052795,
|
|
0.7794657945632935,
|
|
0.9775897860527039
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.962005615234375,
|
|
"extreme_code_fraction": 0.0958636924624443,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.26600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.002019473584368825,
|
|
"max_absolute_error": 0.007541656494140625,
|
|
"cosine_loss": {
|
|
"mean": 1.7197980923810974e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7642974853515625e-05,
|
|
"max": 2.7000904083251953e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004260815039742738,
|
|
"p95": 0.0009221136569976807,
|
|
"p99": 0.004576936364173889,
|
|
"max": 0.006372332572937012
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.5.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012452160008251667,
|
|
"metrics": {
|
|
"relative_error": 0.07310004532337189,
|
|
"max_absolute_error": 0.112060546875,
|
|
"cosine_loss": {
|
|
"mean": 0.0026715362910181284,
|
|
"p95": 0.0027427226305007935,
|
|
"p99": 0.002779453992843628,
|
|
"max": 0.0028954744338989258
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003718055668286979,
|
|
"p95": 0.0009635686874389648,
|
|
"p99": 0.0016308426856994629,
|
|
"max": 0.003155231475830078
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032127131707966328,
|
|
"p95": 0.00548398494720459,
|
|
"p99": 0.006658419966697693,
|
|
"max": 0.0090981125831604
|
|
},
|
|
"correction_min": 0.9976754188537598,
|
|
"correction_max": 1.0091816186904907,
|
|
"codebook": [
|
|
-0.9782091379165649,
|
|
-0.7829001545906067,
|
|
-0.621691107749939,
|
|
-0.4815986752510071,
|
|
-0.3574022054672241,
|
|
-0.24404406547546387,
|
|
-0.140096977353096,
|
|
-0.04003969579935074,
|
|
0.05609834939241409,
|
|
0.150473952293396,
|
|
0.2505653202533722,
|
|
0.3598858416080475,
|
|
0.480586439371109,
|
|
0.61762535572052,
|
|
0.7817372679710388,
|
|
0.9783883690834045
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9611289501190186,
|
|
"extreme_code_fraction": 0.09569370746612549,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019913017749786377,
|
|
"max_absolute_error": 0.008243203163146973,
|
|
"cosine_loss": {
|
|
"mean": 1.7179769429276348e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.913309097290039e-05,
|
|
"max": 2.6285648345947266e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042294609011150897,
|
|
"p95": 0.0009611845016479492,
|
|
"p99": 0.00378808262757957,
|
|
"max": 0.005908012390136719
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.23399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.0012480929726734757,
|
|
"metrics": {
|
|
"relative_error": 0.07305481284856796,
|
|
"max_absolute_error": 0.0888671875,
|
|
"cosine_loss": {
|
|
"mean": 0.002668400527909398,
|
|
"p95": 0.0027399063110351562,
|
|
"p99": 0.0027707391418516636,
|
|
"max": 0.002831697463989258
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003935018612537533,
|
|
"p95": 0.0009935169946402311,
|
|
"p99": 0.0016146226553246379,
|
|
"max": 0.003575146198272705
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003306427737697959,
|
|
"p95": 0.005466335918754339,
|
|
"p99": 0.0063713183626532555,
|
|
"max": 0.009096622467041016
|
|
},
|
|
"correction_min": 0.9980355501174927,
|
|
"correction_max": 1.0091800689697266,
|
|
"codebook": [
|
|
-0.9771572351455688,
|
|
-0.7776321768760681,
|
|
-0.6154751777648926,
|
|
-0.4765172004699707,
|
|
-0.3537605106830597,
|
|
-0.24290956556797028,
|
|
-0.13975869119167328,
|
|
-0.041085101664066315,
|
|
0.054962411522865295,
|
|
0.15146400034427643,
|
|
0.25201189517974854,
|
|
0.36184316873550415,
|
|
0.48231276869773865,
|
|
0.6209070086479187,
|
|
0.7841023802757263,
|
|
0.9786476492881775
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.962562322616577,
|
|
"extreme_code_fraction": 0.09569928050041199,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0019790558144450188,
|
|
"max_absolute_error": 0.007255375385284424,
|
|
"cosine_loss": {
|
|
"mean": 1.6753274394432083e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.7777085304260254e-05,
|
|
"max": 2.6881694793701172e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000409579137340188,
|
|
"p95": 0.000888630747795105,
|
|
"p99": 0.004351958632469177,
|
|
"max": 0.006424665451049805
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.1879999999946449
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.6.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012474153190851212,
|
|
"metrics": {
|
|
"relative_error": 0.07318229973316193,
|
|
"max_absolute_error": 0.122314453125,
|
|
"cosine_loss": {
|
|
"mean": 0.002677238779142499,
|
|
"p95": 0.0027510076761245728,
|
|
"p99": 0.0027839988470077515,
|
|
"max": 0.0030983686447143555
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003608409024309367,
|
|
"p95": 0.0009364038705825806,
|
|
"p99": 0.0015747696161270142,
|
|
"max": 0.0031912922859191895
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0032540347892791033,
|
|
"p95": 0.005646601319313049,
|
|
"p99": 0.006917282938957214,
|
|
"max": 0.010605156421661377
|
|
},
|
|
"correction_min": 0.9982367753982544,
|
|
"correction_max": 1.010718822479248,
|
|
"codebook": [
|
|
-0.97788006067276,
|
|
-0.7803740501403809,
|
|
-0.6186602115631104,
|
|
-0.47834649682044983,
|
|
-0.3542231023311615,
|
|
-0.24162131547927856,
|
|
-0.13798685371875763,
|
|
-0.0387776754796505,
|
|
0.056990690529346466,
|
|
0.15299220383167267,
|
|
0.2530665695667267,
|
|
0.3611892759799957,
|
|
0.48181062936782837,
|
|
0.6207167506217957,
|
|
0.7823190689086914,
|
|
0.9777308106422424
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.961667537689209,
|
|
"extreme_code_fraction": 0.09594261646270752,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001996236387640238,
|
|
"max_absolute_error": 0.007451057434082031,
|
|
"cosine_loss": {
|
|
"mean": 1.7390536868333584e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9252300262451172e-05,
|
|
"max": 2.7120113372802734e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042997696436941624,
|
|
"p95": 0.0009640661301091313,
|
|
"p99": 0.003934645093977451,
|
|
"max": 0.006398200988769531
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.35899999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001243651146069169,
|
|
"metrics": {
|
|
"relative_error": 0.07309281826019287,
|
|
"max_absolute_error": 0.0888671875,
|
|
"cosine_loss": {
|
|
"mean": 0.002671149093657732,
|
|
"p95": 0.0027427077293395996,
|
|
"p99": 0.0027732003945857286,
|
|
"max": 0.00284421443939209
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004187453887425363,
|
|
"p95": 0.0010675727389752865,
|
|
"p99": 0.00165452528744936,
|
|
"max": 0.0035424232482910156
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0029334083665162325,
|
|
"p95": 0.005060070659965277,
|
|
"p99": 0.005979868583381176,
|
|
"max": 0.010420680046081543
|
|
},
|
|
"correction_min": 0.9972631931304932,
|
|
"correction_max": 1.0105304718017578,
|
|
"codebook": [
|
|
-0.9776100516319275,
|
|
-0.781499445438385,
|
|
-0.6194900274276733,
|
|
-0.47921231389045715,
|
|
-0.35625752806663513,
|
|
-0.2446018010377884,
|
|
-0.14087313413619995,
|
|
-0.04180346429347992,
|
|
0.055522941052913666,
|
|
0.1517220139503479,
|
|
0.25260162353515625,
|
|
0.36084097623825073,
|
|
0.4836371839046478,
|
|
0.6214587092399597,
|
|
0.7827306985855103,
|
|
0.9781309962272644
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9603233337402344,
|
|
"extreme_code_fraction": 0.09543422609567642,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.40599999998812564
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020017337519675493,
|
|
"max_absolute_error": 0.008139729499816895,
|
|
"cosine_loss": {
|
|
"mean": 1.7027791727741715e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.749396324157715e-05,
|
|
"max": 2.658367156982422e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042318186024203897,
|
|
"p95": 0.0009182095527648926,
|
|
"p99": 0.004456460475921631,
|
|
"max": 0.006814122200012207
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.20300000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.01600000000325963
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.7.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012448488269001245,
|
|
"metrics": {
|
|
"relative_error": 0.07313834875822067,
|
|
"max_absolute_error": 0.08984375,
|
|
"cosine_loss": {
|
|
"mean": 0.00267433631233871,
|
|
"p95": 0.0027458667755126953,
|
|
"p99": 0.0027815252542495728,
|
|
"max": 0.002895653247833252
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00037671448080800474,
|
|
"p95": 0.0009728521108627319,
|
|
"p99": 0.001581341028213501,
|
|
"max": 0.003104567527770996
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.00305312592536211,
|
|
"p95": 0.005303993821144104,
|
|
"p99": 0.006512805819511414,
|
|
"max": 0.00919872522354126
|
|
},
|
|
"correction_min": 0.997974693775177,
|
|
"correction_max": 1.0092841386795044,
|
|
"codebook": [
|
|
-0.9782582521438599,
|
|
-0.7821155190467834,
|
|
-0.6198243498802185,
|
|
-0.48232218623161316,
|
|
-0.3617279827594757,
|
|
-0.253024160861969,
|
|
-0.15272098779678345,
|
|
-0.056674934923648834,
|
|
0.039800822734832764,
|
|
0.1398584395647049,
|
|
0.24473048746585846,
|
|
0.3564830720424652,
|
|
0.4810809791088104,
|
|
0.6207761168479919,
|
|
0.7822837233543396,
|
|
0.9779335856437683
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.960641384124756,
|
|
"extreme_code_fraction": 0.09561644494533539,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020006883423775434,
|
|
"max_absolute_error": 0.007881760597229004,
|
|
"cosine_loss": {
|
|
"mean": 1.737473780849541e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.9371509552001953e-05,
|
|
"max": 2.6285648345947266e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004310527292545885,
|
|
"p95": 0.0009826152818277478,
|
|
"p99": 0.003850929904729128,
|
|
"max": 0.006331384181976318
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.32800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001250812434591353,
|
|
"metrics": {
|
|
"relative_error": 0.07303152233362198,
|
|
"max_absolute_error": 0.08935546875,
|
|
"cosine_loss": {
|
|
"mean": 0.0026666002813726664,
|
|
"p95": 0.002738502575084567,
|
|
"p99": 0.00276869535446167,
|
|
"max": 0.0028434395790100098
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0003851038636639714,
|
|
"p95": 0.000992265297099948,
|
|
"p99": 0.0016075416933745146,
|
|
"max": 0.0037217140197753906
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.0030037995893508196,
|
|
"p95": 0.005261578597128391,
|
|
"p99": 0.006254888139665127,
|
|
"max": 0.009166121482849121
|
|
},
|
|
"correction_min": 0.9967193603515625,
|
|
"correction_max": 1.0092509984970093,
|
|
"codebook": [
|
|
-0.9778463840484619,
|
|
-0.7813081741333008,
|
|
-0.61932772397995,
|
|
-0.4815598428249359,
|
|
-0.3592739701271057,
|
|
-0.25016891956329346,
|
|
-0.14995525777339935,
|
|
-0.05443304777145386,
|
|
0.04133214056491852,
|
|
0.14062170684337616,
|
|
0.24485275149345398,
|
|
0.35577526688575745,
|
|
0.47856011986732483,
|
|
0.6179800033569336,
|
|
0.7797194719314575,
|
|
0.9772695899009705
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.9613466262817383,
|
|
"extreme_code_fraction": 0.09568049758672714,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.42199999999138527
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020122863352298737,
|
|
"max_absolute_error": 0.008110642433166504,
|
|
"cosine_loss": {
|
|
"mean": 1.730175654302002e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.862645149230957e-05,
|
|
"max": 2.6106834411621094e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00042323049274273217,
|
|
"p95": 0.0009491145610809326,
|
|
"p99": 0.004304111003875732,
|
|
"max": 0.006005764007568359
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.20300000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.8.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.adaln_proj.linear.bias",
|
|
"shape": [
|
|
96768
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 193536,
|
|
"output_bytes": 193536,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.adaln_proj.linear.weight",
|
|
"shape": [
|
|
96768,
|
|
8
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1548288,
|
|
"output_bytes": 1548288,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 21697667,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 293888,
|
|
"normalized_fit_mse": 0.0012476989068090916,
|
|
"metrics": {
|
|
"relative_error": 0.07312382757663727,
|
|
"max_absolute_error": 0.1025390625,
|
|
"cosine_loss": {
|
|
"mean": 0.002674717688933015,
|
|
"p95": 0.0027508139610290527,
|
|
"p99": 0.002796068787574768,
|
|
"max": 0.0029882192611694336
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00036233413266018033,
|
|
"p95": 0.0009482502937316895,
|
|
"p99": 0.0015804320573806763,
|
|
"max": 0.0033846497535705566
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.00337620940990746,
|
|
"p95": 0.005731850862503052,
|
|
"p99": 0.007070198655128479,
|
|
"max": 0.011585891246795654
|
|
},
|
|
"correction_min": 0.9970964193344116,
|
|
"correction_max": 1.0117217302322388,
|
|
"codebook": [
|
|
-0.9771981835365295,
|
|
-0.7777514457702637,
|
|
-0.6154248118400574,
|
|
-0.47603675723075867,
|
|
-0.35274672508239746,
|
|
-0.24217519164085388,
|
|
-0.1382102072238922,
|
|
-0.03870409354567528,
|
|
0.05598434433341026,
|
|
0.15161946415901184,
|
|
0.25182580947875977,
|
|
0.3605615794658661,
|
|
0.4822239577770233,
|
|
0.6206473112106323,
|
|
0.7824888825416565,
|
|
0.978449285030365
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.962596893310547,
|
|
"extreme_code_fraction": 0.0961371436715126,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.125
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.001993322977796197,
|
|
"max_absolute_error": 0.006951808929443359,
|
|
"cosine_loss": {
|
|
"mean": 1.7084641967812786e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.913309097290039e-05,
|
|
"max": 2.5928020477294922e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004286865587346256,
|
|
"p95": 0.0009663105010986328,
|
|
"p99": 0.0037762336432933807,
|
|
"max": 0.006458282470703125
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.34399999998277053
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "asym_w4a8_int8",
|
|
"selected_format": "asym_w4a8_int8",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 86818947,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"fit_iterations": 25,
|
|
"fit_sample_count": 295680,
|
|
"normalized_fit_mse": 0.001244992483407259,
|
|
"metrics": {
|
|
"relative_error": 0.07305676490068436,
|
|
"max_absolute_error": 0.095703125,
|
|
"cosine_loss": {
|
|
"mean": 0.002668433589860797,
|
|
"p95": 0.0027410387992858887,
|
|
"p99": 0.0027714469470083714,
|
|
"max": 0.002857983112335205
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.000404836522648111,
|
|
"p95": 0.0010353205725550652,
|
|
"p99": 0.0016376473940908909,
|
|
"max": 0.0036153793334960938
|
|
}
|
|
},
|
|
"qkv_metrics": null,
|
|
"row_norm_before_correction": {
|
|
"mean": 0.003118226071819663,
|
|
"p95": 0.005363797768950462,
|
|
"p99": 0.006361367646604776,
|
|
"max": 0.008808553218841553
|
|
},
|
|
"correction_min": 0.996654748916626,
|
|
"correction_max": 1.0088868141174316,
|
|
"codebook": [
|
|
-0.9776243567466736,
|
|
-0.7809601426124573,
|
|
-0.6193476319313049,
|
|
-0.4795742928981781,
|
|
-0.35619643330574036,
|
|
-0.24399328231811523,
|
|
-0.14091169834136963,
|
|
-0.04185844212770462,
|
|
0.0550784133374691,
|
|
0.15116961300373077,
|
|
0.2527432441711426,
|
|
0.36188364028930664,
|
|
0.48242810368537903,
|
|
0.6194941997528076,
|
|
0.7809097170829773,
|
|
0.9775297045707703
|
|
],
|
|
"code_utilization": 1.0,
|
|
"code_entropy_bits": 3.960763454437256,
|
|
"extreme_code_fraction": 0.09578825533390045,
|
|
"maximum_occupied_code": 15
|
|
},
|
|
"duration_seconds": 0.46900000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.0020263034384697676,
|
|
"max_absolute_error": 0.007393240928649902,
|
|
"cosine_loss": {
|
|
"mean": 1.746994826135051e-06,
|
|
"p95": 1.4901161193847656e-06,
|
|
"p99": 1.8462538719177246e-05,
|
|
"max": 2.5987625122070312e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0004347178037278354,
|
|
"p95": 0.0009466558694839478,
|
|
"p99": 0.004420801997184753,
|
|
"max": 0.006360530853271484
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.2029999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.blocks.9.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.condition_proj.bias",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.condition_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
5120
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 55050240,
|
|
"output_bytes": 55050240,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.09400000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.final_layer.adaln_proj.linear.bias",
|
|
"shape": [
|
|
10752
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 21504,
|
|
"output_bytes": 21504,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.final_layer.adaln_proj.linear.weight",
|
|
"shape": [
|
|
10752,
|
|
2688
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 57802752,
|
|
"output_bytes": 57802752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.07800000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.final_layer.audio_out.bias",
|
|
"shape": [
|
|
32
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 64,
|
|
"output_bytes": 64,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.014999999984866008
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.final_layer.audio_out.weight",
|
|
"shape": [
|
|
32,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 344064,
|
|
"output_bytes": 344064,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.final_layer.norm.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.final_layer.video_out.bias",
|
|
"shape": [
|
|
96
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 192,
|
|
"output_bytes": 192,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.final_layer.video_out.weight",
|
|
"shape": [
|
|
96,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1032192,
|
|
"output_bytes": 1032192,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.rope.inv_freq",
|
|
"shape": [
|
|
16
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 32,
|
|
"output_bytes": 32,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.0.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.0.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.010350251570343971,
|
|
"max_absolute_error": 0.0059805866330862045,
|
|
"cosine_loss": {
|
|
"mean": 5.3372856200439855e-05,
|
|
"p95": 7.253885269165039e-05,
|
|
"p99": 8.630752563476562e-05,
|
|
"max": 0.00012618303298950195
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00010547354759182781,
|
|
"p95": 0.0002624094486236572,
|
|
"p99": 0.0003477334976196289,
|
|
"max": 0.0005298852920532227
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.11000000001513399
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.0.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.0.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.008820964023470879,
|
|
"max_absolute_error": 0.006594359874725342,
|
|
"cosine_loss": {
|
|
"mean": 3.890880543622188e-05,
|
|
"p95": 5.0476170144975185e-05,
|
|
"p99": 5.9066456742584705e-05,
|
|
"max": 8.040666580200195e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0001002402714220807,
|
|
"p95": 0.0002504587173461914,
|
|
"p99": 0.00032876525074243546,
|
|
"max": 0.0006040334701538086
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3279999999795109
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.0.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 154255427,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.008826748467981815,
|
|
"max_absolute_error": 0.011630639433860779,
|
|
"cosine_loss": {
|
|
"mean": 3.891330197802745e-05,
|
|
"p95": 5.054473876953125e-05,
|
|
"p99": 5.884713027626276e-05,
|
|
"max": 8.440017700195312e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00010049519914900884,
|
|
"p95": 0.0002473592758178711,
|
|
"p99": 0.0003258336801081896,
|
|
"max": 0.0006692409515380859
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.4370000000053551
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.0.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.009366999380290508,
|
|
"max_absolute_error": 0.008116163313388824,
|
|
"cosine_loss": {
|
|
"mean": 4.383830673759803e-05,
|
|
"p95": 5.53131103515625e-05,
|
|
"p99": 6.35981559753418e-05,
|
|
"max": 8.851289749145508e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 7.212423952296376e-05,
|
|
"p95": 0.00017762184143066406,
|
|
"p99": 0.00023180246353149414,
|
|
"max": 0.00039207935333251953
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.21900000001187436
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.0.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.0.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.1.attn.k_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.1.attn.out_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
7168
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 77070336,
|
|
"output_bytes": 38556739,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.010118324309587479,
|
|
"max_absolute_error": 0.007411450147628784,
|
|
"cosine_loss": {
|
|
"mean": 5.098741894471459e-05,
|
|
"p95": 6.867945194244385e-05,
|
|
"p99": 8.177757263183594e-05,
|
|
"max": 0.00013756752014160156
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.0001026169047690928,
|
|
"p95": 0.00025093555450439453,
|
|
"p99": 0.00033801794052124023,
|
|
"max": 0.0005092620849609375
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.10899999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.1.attn.q_norm.weight",
|
|
"shape": [
|
|
128
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 256,
|
|
"output_bytes": 256,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.1.attn.qkv_proj.weight",
|
|
"shape": [
|
|
21504,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 231211008,
|
|
"output_bytes": 115691587,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.008821888826787472,
|
|
"max_absolute_error": 0.008736267685890198,
|
|
"cosine_loss": {
|
|
"mean": 3.8911093724891543e-05,
|
|
"p95": 5.060434341430664e-05,
|
|
"p99": 5.829334259033203e-05,
|
|
"max": 9.626150131225586e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00010037351603386924,
|
|
"p95": 0.00024819374084472656,
|
|
"p99": 0.0003319961251690984,
|
|
"max": 0.0005831718444824219
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.3289999999979045
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.1.mlp.fc1.weight",
|
|
"shape": [
|
|
28672,
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 308281344,
|
|
"output_bytes": 154255427,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.00881592184305191,
|
|
"max_absolute_error": 0.010055914521217346,
|
|
"cosine_loss": {
|
|
"mean": 3.8818699977127835e-05,
|
|
"p95": 5.048513412475586e-05,
|
|
"p99": 5.811452865600586e-05,
|
|
"max": 8.803606033325195e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 0.00010119312355527654,
|
|
"p95": 0.00025022029876708984,
|
|
"p99": 0.0003261913079768419,
|
|
"max": 0.0005408525466918945
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.45300000000861473
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.1.mlp.fc2.weight",
|
|
"shape": [
|
|
5376,
|
|
14336
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "int8_tensorwise",
|
|
"selected_format": "int8_tensorwise",
|
|
"original_bytes": 154140672,
|
|
"output_bytes": 77091907,
|
|
"promotion_reason": null,
|
|
"diagnostics": {
|
|
"int8": {
|
|
"metrics": {
|
|
"relative_error": 0.009371442720293999,
|
|
"max_absolute_error": 0.005852743983268738,
|
|
"cosine_loss": {
|
|
"mean": 4.390416506794281e-05,
|
|
"p95": 5.5730342864990234e-05,
|
|
"p99": 6.461143493652344e-05,
|
|
"max": 8.755922317504883e-05
|
|
},
|
|
"norm_drift": {
|
|
"mean": 7.03453115420416e-05,
|
|
"p95": 0.00017049908638000488,
|
|
"p99": 0.0002288520336151123,
|
|
"max": 0.00037169456481933594
|
|
}
|
|
},
|
|
"convrot": true,
|
|
"convrot_groupsize": 256
|
|
}
|
|
},
|
|
"duration_seconds": 0.23399999999674037
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.1.norm1.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.blocks.1.norm2.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.token_refiner.final_norm.weight",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.video_patch_proj.bias",
|
|
"shape": [
|
|
5376
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 10752,
|
|
"output_bytes": 10752,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
},
|
|
{
|
|
"name": "model.diffusion_model.video_patch_proj.weight",
|
|
"shape": [
|
|
5376,
|
|
96
|
|
],
|
|
"source_dtype": "torch.bfloat16",
|
|
"planned_format": "bf16",
|
|
"selected_format": "bf16",
|
|
"original_bytes": 1032192,
|
|
"output_bytes": 1032192,
|
|
"promotion_reason": null,
|
|
"diagnostics": {},
|
|
"duration_seconds": 0.0
|
|
}
|
|
],
|
|
"output_sha256": "2242d907fa3bcbbe3bc01890a989a5468966801aae5922df8ae021aaafb11632"
|
|
}
|