Ethanfel
  • Joined on 2026-01-04
Ethanfel pushed to feature/cupy-fallback at Ethanfel/ComfyUI-Tween 2026-04-11 10:13:35 +02:00
83e4b5dd98 perf: add torch.compile to PyTorch fallback kernels
Ethanfel created branch feature/cupy-fallback in Ethanfel/ComfyUI-Tween 2026-04-11 02:21:08 +02:00
Ethanfel pushed to feature/cupy-fallback at Ethanfel/ComfyUI-Tween 2026-04-11 02:21:08 +02:00
2e75e2d076 fix: handle None from cupy.cuda.get_cuda_path() in cuda_launch
c08fe58fe7 feat: make cupy optional in install.py
9e84890877 feat: remove cupy requirement gate from model loading
2e98e453a4 Add pure-PyTorch fallback for BIM-VFI cost volume kernel
daf0304243 Add pure-PyTorch fallback for GIMM-VFI softsplat forward warp
Compare 7 commits »
Ethanfel pushed to master at Ethanfel/ComfyUI-Omnivoice 2026-04-10 22:17:23 +02:00
b4c1cb2955 chore: bump version to 1.0.6
Ethanfel pushed to master at Ethanfel/ComfyUI-Omnivoice 2026-04-10 22:12:05 +02:00
2e3c357e5a fix: raise speed minimum from 0.1 to 0.3 to prevent VRAM explosion
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 16:48:27 +02:00
40d29bcaf8 feat: add experiment configs for logit+cosine combo and BigVGAN decoder fine-tuning
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 15:04:09 +02:00
65dc549494 feat: add reference audio comparison metrics to LoRA trainer eval
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 13:44:41 +02:00
f745e241c4 chore: sanitize tooltips/comments + add experiment configs
Ethanfel pushed to master at Ethanfel/ComfyUI-Omnivoice 2026-04-10 12:14:14 +02:00
aa986fd534 fix: set transformers>=5.3.0 as required by omnivoice
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 12:13:57 +02:00
082a2da438 fix: restore dtype after float32 STFT in discriminator spectrogram
Ethanfel pushed to master at Ethanfel/ComfyUI-Omnivoice 2026-04-10 12:13:22 +02:00
3fee610050 fix: bump transformers minimum to >=4.57.0
Ethanfel pushed to master at Ethanfel/ComfyUI-Omnivoice 2026-04-10 12:08:46 +02:00
d4638aa785 chore: bump version to 1.0.5
Ethanfel pushed to master at Ethanfel/ComfyUI-Omnivoice 2026-04-10 12:07:11 +02:00
e9c947b613 fix: strip title and heading tags from EPUB text output
197bcc554e chore: bump version to 1.0.4
9f2683cd54 fix: add torchcodec to requirements for torchaudio backend
f8657aca80 fix: update tests for chapter_title output and guidance_scale param
0f2bcc4c0e chore: bump version to 1.0.3
Compare 28 commits »
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 11:36:04 +02:00
c28e090196 fix: cast discriminator inputs to match bfloat16 dtype in BigVGAN FM loss
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 11:32:04 +02:00
af6c225f53 feat: add dataset pipeline nodes + latent augmentation for LoRA trainer
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 02:40:05 +02:00
30127c13ca feat: add BigVGAN vocoder sweep scheduler node
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 01:50:09 +02:00
4226297735 chore: remove debug VRAM logging
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 01:48:06 +02:00
4297715a08 debug: add driver-level VRAM reporting + offload video_enc
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 01:42:47 +02:00
9af4bbdd91 fix: force torch.cuda.empty_cache() after pre-generation and CLIP encoding
Ethanfel pushed to feature/lora-timestep-sampling at Ethanfel/ComfyUI-SelVA 2026-04-10 01:35:56 +02:00
89d6fccd28 debug: add per-operation VRAM logging in first training step