from pathlib import Path import sys import uuid import pytest import torch sys.path.insert(0, str(Path(__file__).resolve().parents[2])) sys.modules["decode_cache_test_fixtures"] = sys.modules[__name__] class NativeVideoModel(torch.nn.Module): __module__ = "comfy.ldm.minimax.vae" def __init__(self): super().__init__() self.register_parameter("weight", torch.nn.Parameter(torch.ones(1), requires_grad=False)) self.register_buffer("offset", torch.zeros(1)) self.tile_size = 256 self.eval() def decode(self, tensor): return tensor * self.weight + self.offset class NativeAudioModel(NativeVideoModel): __module__ = "comfy.ldm.minimax.audio_vae" class Patcher: def __init__(self): self.patches_uuid = uuid.uuid4() self.object_patches = {} class FakeVAE: """Controlled delegate fixture, NOT a real MiniMax model/GPU execution.""" __module__ = "comfy.sd" def __init__(self, stream="video"): self.first_stage_model = NativeVideoModel() if stream == "video" else NativeAudioModel() self.patcher = Patcher() self.vae_dtype = torch.float32 self.output_device = torch.device("cpu") self.device = torch.device("cpu") self.audio_sample_rate = 32000 self.calls = 0 def decode(self, tensor): self.calls += 1 return self.first_stage_model.decode(tensor) def process_output(self, x): return x class NestedFixture: is_nested = True def __init__(self, video, audio): self.parts = video, audio def unbind(self): return self.parts @pytest.fixture def vae(): return FakeVAE() @pytest.fixture def samples(): return {"samples": torch.arange(120, dtype=torch.float32).reshape(5, 3, 4, 2) / 120}