Ethanfel/ComfyUI-SelVA
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

debug: add latent and audio stats logging to V2A sampler

Browse Source 
Match the diagnostic output already in text_only.py to compare
V2A vs T2A latent distributions and diagnose conditioning issues.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Ethanfel
2026-03-27 22:28:08 +01:00
parent c0b7ccbcee
commit 477fe0f08f
1 changed files with 7 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nodes/sampler.py
+7 -1
View File
@@ -118,6 +118,9 @@ class PrismAudioSampler:
batch_cfg=True, batch_cfg=True,
) )
fakes_f = fakes.float()
print(f"[PrismAudio] latent stats: shape={tuple(fakes_f.shape)} mean={fakes_f.mean():.4f} std={fakes_f.std():.4f} min={fakes_f.min():.4f} max={fakes_f.max():.4f}", flush=True)
# Offload diffusion model and conditioner before VAE decode # Offload diffusion model and conditioner before VAE decode
if strategy == "offload_to_cpu": if strategy == "offload_to_cpu":
diffusion.model.to(get_offload_device()) diffusion.model.to(get_offload_device())
@@ -127,7 +130,7 @@ class PrismAudioSampler:
# VAE decode in fp32 (snake activations overflow in fp16) # VAE decode in fp32 (snake activations overflow in fp16)
with torch.amp.autocast(device_type=device.type, enabled=False): with torch.amp.autocast(device_type=device.type, enabled=False):
audio = diffusion.pretransform.decode(fakes.float()) audio = diffusion.pretransform.decode(fakes_f)
# Offload VAE # Offload VAE
if strategy == "offload_to_cpu": if strategy == "offload_to_cpu":
@@ -136,8 +139,11 @@ class PrismAudioSampler:
# Peak normalize then clamp (matching reference: div by max abs before clamp) # Peak normalize then clamp (matching reference: div by max abs before clamp)
audio = audio.float() audio = audio.float()
pre_norm_std = audio.std().item()
pre_norm_peak = audio.abs().max().item()
peak = audio.abs().max().clamp(min=1e-8) peak = audio.abs().max().clamp(min=1e-8)
audio = (audio / peak).clamp(-1, 1) audio = (audio / peak).clamp(-1, 1)
print(f"[PrismAudio] audio stats (pre-norm): std={pre_norm_std:.4f} peak={pre_norm_peak:.4f}", flush=True)
# Return as ComfyUI AUDIO: {"waveform": [B, channels, samples], "sample_rate": int} # Return as ComfyUI AUDIO: {"waveform": [B, channels, samples], "sample_rate": int}
return ({"waveform": audio.cpu(), "sample_rate": SAMPLE_RATE},) return ({"waveform": audio.cpu(), "sample_rate": SAMPLE_RATE},)