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

fix: cast discriminators to model dtype to match vocoder output

Browse Source 
Discriminators are constructed as float32 but receive bfloat16 tensors
from the vocoder. Cast to model dtype on load to prevent conv dtype
mismatch in feature matching loss.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Ethanfel
2026-04-10 00:25:04 +02:00
parent 187b2e3169
commit 0854bd2638
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nodes/selva_bigvgan_trainer.py
+2 -2
View File
@@ -954,8 +954,8 @@ def _do_train(vocoder, mel_converter, clips,
f"MPD loaded={mpd_loaded}, MRD loaded={mrd_loaded}. " f"MPD loaded={mpd_loaded}, MRD loaded={mrd_loaded}. "
f"Available keys: {list(ckpt_d.keys())}" f"Available keys: {list(ckpt_d.keys())}"
) )
mpd.to(device).eval() mpd.to(device, dtype).eval()
mrd.to(device).eval() mrd.to(device, dtype).eval()
for p in mpd.parameters(): for p in mpd.parameters():
p.requires_grad_(False) p.requires_grad_(False)
for p in mrd.parameters(): for p in mrd.parameters():