"""One-physical-group Sampling Engine for the Continuum runtime.""" from __future__ import annotations from typing import Any, Callable def sample_physical_group( *, sample_callable: Callable[..., Any], model: Any, conditioning: Any, latent: Any, sampler: Any, sigmas: Any, seed: int, enable_preview: bool, physical_group: int, logical_chunks: tuple[int, ...], terminal_atomic: bool, layout_validation_profiler: Any = None, packed_row_planner: Any = None, planner_event: Any = None, diagnostic_policy: Any = None, context_frames: int = 0, ) -> tuple[Any, Any]: """Run exactly one physical sample with unchanged observer ordering.""" normalized_chunks = tuple(int(value) for value in logical_chunks) layout_profile_token = ( layout_validation_profiler.begin_sampling_group( physical_group=int(physical_group), logical_chunks=normalized_chunks, sampling_steps=max(0, int(sigmas.shape[-1]) - 1), ) if layout_validation_profiler is not None else None ) diagnostic_sample_token = None if diagnostic_policy is not None: diagnostic_sample_token = diagnostic_policy.before_sampling( physical_group=int(physical_group), logical_chunks=normalized_chunks, context_frames=int(context_frames), latent=latent, conditioning=conditioning, seed=int(seed), sigmas=sigmas, ) if planner_event is not None: planner_event( packed_row_planner, "begin_group", physical_group=int(physical_group), logical_chunks=normalized_chunks, terminal_atomic=bool(terminal_atomic), ) sampled = sample_callable( model=model, conditioning=conditioning, latent=latent, sampler=sampler, sigmas=sigmas, seed=int(seed), enable_preview=bool(enable_preview), ) if planner_event is not None: planner_event(packed_row_planner, "finish_group") if layout_validation_profiler is not None and layout_profile_token is not None: layout_validation_profiler.finish_sampling_group(layout_profile_token) return sampled, diagnostic_sample_token