2.87 GB

Ctrl+K

1 contributor

History: 14 commits

Mothersuperior

Publish recovered joint_v6 audio-loss trainer and required neighbor tables

e2e63d8 verified 18 days ago

assets
Publish recovered joint_v6 audio-loss trainer and required neighbor tables 18 days ago
scripts
Publish recovered joint_v6 audio-loss trainer and required neighbor tables 18 days ago
.gitattributes

1.52 kB
initial commit 24 days ago
README.md

12 kB
Publish recovered joint_v6 audio-loss trainer and required neighbor tables 18 days ago
nar_lora_joint_v4.bf16.safetensors

70.3 MB
xet

Add safetensors releases (fp32 + bf16) with named LoRA tensors; scripts load either format 22 days ago
nar_lora_joint_v4.pt

141 MB
xet

rename weights to v4 24 days ago
nar_lora_joint_v4.safetensors

141 MB
xet

Add safetensors releases (fp32 + bf16) with named LoRA tensors; scripts load either format 22 days ago
nar_lora_joint_v4_comfyui.safetensors

108 MB
xet

Add ComfyUI-native NAR LoRA (fused qkv/gate_up layout, vae2llm/llm2vae as diffs) 22 days ago
nar_lora_joint_v5.bf16.safetensors

70.3 MB
xet

Add v5 joint pair: co-trained head + NAR LoRA on a mixed 3-artist real set (real decoder loss 1.056->0.981) 21 days ago
nar_lora_joint_v5.safetensors

141 MB
xet

Add v5 joint pair: co-trained head + NAR LoRA on a mixed 3-artist real set (real decoder loss 1.056->0.981) 21 days ago
nar_lora_joint_v5_comfyui.safetensors

108 MB
xet

Add v5 joint pair: co-trained head + NAR LoRA on a mixed 3-artist real set (real decoder loss 1.056->0.981) 21 days ago
nar_lora_joint_v8.bf16.safetensors

70.3 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
nar_lora_joint_v8.safetensors

141 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
nar_lora_joint_v8_comfyui.safetensors

108 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
nar_lora_joint_v9.bf16.safetensors

70.3 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
nar_lora_joint_v9.safetensors

141 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
nar_lora_joint_v9_comfyui.safetensors

108 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
tokenizer_head_joint_v4.bf16.safetensors

85.6 MB
xet

Add safetensors releases (fp32 + bf16) with named LoRA tensors; scripts load either format 22 days ago
tokenizer_head_joint_v4.pt

171 MB
xet

rename weights to v4 24 days ago
tokenizer_head_joint_v4.safetensors

171 MB
xet

Add safetensors releases (fp32 + bf16) with named LoRA tensors; scripts load either format 22 days ago
tokenizer_head_joint_v5.bf16.safetensors

85.6 MB
xet

Add v5 joint pair: co-trained head + NAR LoRA on a mixed 3-artist real set (real decoder loss 1.056->0.981) 21 days ago
tokenizer_head_joint_v5.safetensors

171 MB
xet

Add v5 joint pair: co-trained head + NAR LoRA on a mixed 3-artist real set (real decoder loss 1.056->0.981) 21 days ago
tokenizer_head_joint_v8.bf16.safetensors

85.6 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
tokenizer_head_joint_v8.safetensors

171 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
tokenizer_head_joint_v9.bf16.safetensors

85.6 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
tokenizer_head_joint_v9.safetensors

171 MB
xet

Add v8 and v9 joint pairs trained with an audio-domain loss (log-mel + MR-STFT + stereo vs original recordings); benchmark table in README 21 days ago
tokenizer_head_v5_30k.bf16.safetensors

85.6 MB
xet

Add v5 tokenizer head (14.5k tracks, optimi 5e-4 x 30k): top-1 18.9% 21 days ago
tokenizer_head_v5_30k.safetensors

171 MB
xet

Add v5 tokenizer head (14.5k tracks, optimi 5e-4 x 30k): top-1 18.9% 21 days ago