daniel/h3-blackwell-runtime
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Compare H3 sampler pre-step states

Browse source
This commit is contained in:
Daniel Maddern 2026-08-13 14:58:04 +07:00
parent 20e18d9d0c
commit 416a7d9697
1 changed files with 10 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

16
tools/compare_fl2va_steps.py
View file

@ -21,15 +21,15 @@ text = H3TokenRefiner(checkpoint)(Qwen3VLPromptConditioner("/text-encoders/qwen3
model = H3PackedDenoiser.from_checkpoint(checkpoint, attention_backend="sage2").eval()
packer = H3PromptPacker(checkpoint)
packed_initial = initial["initial_x"].to("cuda").reshape(-1)
video_shape = (1, 24, 7, 12, 20)
audio_shape = (1, 32, 2, 37)
video_count = torch.tensor(video_shape).prod().item()
video = packed_initial[:video_count].reshape(video_shape)
audio_carried = packed_initial[video_count:].reshape(audio_shape)
old_video = old_audio = old_sigma = None
for index, reference in enumerate(steps):
sigma, sigma_down = sigmas[index], sigmas[index + 1]
packed_state = reference["x"].to("cuda").reshape(-1)
video = packed_state[:video_count].reshape(video_shape)
audio_carried = packed_state[video_count:].reshape(audio_shape)
native_audio = audio_carried * (_audio_sigma(sigma) / sigma)
hidden, times, segments, positions, video_segment, audio_segment = packer(text, video, native_audio, float(sigma))
raw_video, raw_audio = model(hidden, times, positions, segments, video_segment, audio_segment)
@ -44,7 +44,11 @@ for index, reference in enumerate(steps):
previous_sigma = sigmas[index - 1] if index else None
video = res_multistep_update(video, denoised[0], sigma, sigma_down, old_video, old_sigma, previous_sigma)
audio_carried = res_multistep_update(audio_carried, denoised[1], sigma, sigma_down, old_audio, old_sigma, previous_sigma)
reference_latent = reference["x"].to("cuda").reshape(-1)[:video_count].reshape(video_shape)
latent_delta = (video.float() - reference_latent.float()).abs()
print(f"step={index:02d} x0_video_mean={denoised_delta.mean().item():.6g} x0_video_max={denoised_delta.max().item():.6g} latent_video_mean={latent_delta.mean().item():.6g} latent_video_max={latent_delta.max().item():.6g}")
if index + 1 < len(steps):
reference_latent = steps[index + 1]["x"].to("cuda").reshape(-1)[:video_count].reshape(video_shape)
latent_delta = (video.float() - reference_latent.float()).abs()
latent_text = f"latent_video_mean={latent_delta.mean().item():.6g} latent_video_max={latent_delta.max().item():.6g}"
else:
latent_text = "latent_video_mean=final-unobserved latent_video_max=final-unobserved"
print(f"step={index:02d} x0_video_mean={denoised_delta.mean().item():.6g} x0_video_max={denoised_delta.max().item():.6g} {latent_text}")
old_video, old_audio, old_sigma = denoised[0], denoised[1], sigma_down