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

fix(bigvgan-trainer): resolve device mismatch in _save_sample after offload

Browse Source 
After the finally block, offload_to_cpu moves the vocoder to CPU while
ref_mel stays on GPU. Fix: detect vocoder's current device via
next(vocoder.parameters()).device and move ref_mel there before vocoding.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Ethanfel
2026-04-09 01:35:07 +02:00
parent 9fdeb65182
commit 81ff0d46c9
1 changed files with 4 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nodes/selva_bigvgan_trainer.py
+4 -1
View File
@@ -182,8 +182,11 @@ class SelvaBigvganTrainer:
def _save_sample(label): def _save_sample(label):
"""Vocode the reference mel and save as .wav.""" """Vocode the reference mel and save as .wav."""
try: try:
# Vocoder may have been offloaded to CPU after training — match its device.
voc_device = next(vocoder.parameters()).device
mel = ref_mel.to(voc_device)
with torch.no_grad(): with torch.no_grad():
wav = vocoder(ref_mel) # [1, 1, T] or [1, T] wav = vocoder(mel) # [1, 1, T] or [1, T]
if wav.dim() == 2: if wav.dim() == 2:
wav = wav.unsqueeze(1) wav = wav.unsqueeze(1)
wav = wav.float().cpu().clamp(-1, 1) wav = wav.float().cpu().clamp(-1, 1)