Ethanfel

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 14:26:02 +02:00

8a85819f97 feat: register audio dataset pipeline nodes in __init__.py

f1c4654bab feat: add SelvaDatasetItemExtractor node

2d06cb2f52 fix: pass device to hann_window in _check_hf_shelf to avoid GPU mismatch

0731addea9 feat: add SelvaDatasetInspector node (codec artifacts, SNR, clipping)

7eb9bd5745 feat: add SelvaDatasetLUFSNormalizer node (pyloudnorm BS.1770-4)

Compare 9 commits »

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 13:49:52 +02:00

2b10205657 fix: raise segment_seconds max from 4s to 30s

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 13:45:27 +02:00

8166c56552 perf: gradient checkpointing on vocoder forward to reduce activation memory

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 13:40:05 +02:00

eece79ccae fix: correct MRD channel width to 128 and unload models before training

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 12:18:28 +02:00

357b875e5e fix: strip inference tensor flags in DITTO optimizer

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 12:10:07 +02:00

211494a91c fix: DITTO gradient never reached x0, remove unused imports and dead code

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 12:04:10 +02:00

1e9551152e feat: add DITTO optimizer, upgrade BigVGAN trainer, document all nodes

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 03:05:48 +02:00

f17f6f0863 feat: save ground truth spectrogram once for direct comparison

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 03:03:29 +02:00

304d9d01bf feat: save mel spectrogram PNG alongside each eval sample

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 03:01:56 +02:00

0128a81cc2 fix: use full first clip for eval samples instead of 1s segment

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:58:08 +02:00

710261f5be fix: add soundfile fallback for torchaudio.save in sample writing

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:54:43 +02:00

5df2abd6dd fix: handle all three inference-tensor sources in vocoder sanitization

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:46:18 +02:00

b243908873 debug: inspect conv_pre parametrizations and _parameters keys

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:41:53 +02:00

9df855ee0e debug: print is_inference() status before failing conv_pre call

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:35:51 +02:00

78f8aa98ad fix: clone inference tensors at thread entry to strip the inference flag

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:30:54 +02:00

e870446b0f fix: run BigVGAN training in a fresh thread to escape inference_mode

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:14:15 +02:00

df63b147e9 fix: sanitize all submodule buffers of mel_converter + guarantee target_mel output

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:09:27 +02:00

51ac099073 fix: sanitize target_flat — clips are inference tensors from outer inference_mode

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 02:05:40 +02:00

b7565ec458 fix: sanitize inference tensors in BigVGAN trainer via zeros+copy_ pattern

Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA

2026-04-09 01:58:59 +02:00

0fcb6d3106 fix(bigvgan-trainer): replace parameter objects to fully strip inference tensor flag