github-featured/openvpi--diffsinger
2
0
Fork 0
Code Issues Pull Requests Deployments Models Agent Notes Packages Projects Releases Wiki Activity

Commit Graph

  • 1b0a6da195 some minor fixes (#325) main yxlllc 2026-08-19 16:43:09 +08:00
  • 3d372a8119 fill zero-frame words with correct pitch values fix yxlllc 2026-08-19 16:10:51 +08:00
  • 4560fe2399 fill zero-frame phones with correct pitch values yxlllc 2026-08-19 15:12:23 +08:00
  • 1802dfb877 some minor fixes (#322) yxlllc 2026-08-16 14:45:51 +08:00
  • cf9a34e63b fix naming error yxlllc 2026-08-16 13:47:33 +08:00
  • e0063c5d01 fix the off-by-one issue yxlllc 2026-08-15 19:00:21 +08:00
  • 0672052cbc fix preder_ds data loading yxlllc 2026-08-15 16:44:19 +08:00
  • 27761a0796 correct the length calculation yxlllc 2026-08-14 15:48:29 +08:00
  • e2307b1080 feat: Triton fused SoftSignGLU kernel for LYNXNet2 (#312) Kakaru 2026-08-14 00:06:49 +08:00
  • 96bb14dae9 some minor fixes (#321) yxlllc 2026-08-11 15:15:43 +08:00
  • b3d54d3b91 fix yxlllc 2026-08-09 21:27:08 +08:00
  • 25bb493188 some minor fixes yxlllc 2026-08-09 20:24:47 +08:00
  • 8a07f76596 fix: pad mel with spec_min instead of 0.0 in acoustic collater (#313) Kakaru 2026-08-04 13:37:48 +08:00
  • ee7463f551 some minor fixes (#318) yxlllc 2026-08-03 19:39:05 +08:00
  • ad64d79737 some minor fixes yxlllc 2026-08-03 00:44:42 +08:00
  • 4764cab403 fix: preserve timestep rank in sinusoidal embedding (#316) dual-timestep Kakaru 2026-08-02 21:13:27 +08:00
  • 36da95303e fix: pool frame-level speaker mixes for Mix-LN (#317) Kakaru 2026-08-02 20:55:07 +08:00
  • 72981e4d6a support dual-timestep reflow yxlllc 2026-07-30 22:51:09 +08:00
  • 753b7cc622 fix: Roll back the changes related to onnxslim in #308 and #305 (#310) Kakaru 2026-06-28 15:35:56 +08:00
  • 289f0cb64b Update ignored patterns in variance exporter (#308) Kakaru 2026-06-25 15:58:04 +08:00
  • e73056f880 Hardcode reset_iterations yqzhishen 2026-06-25 14:14:34 +08:00
  • 5c14181606 Pre-release audit fixes yqzhishen 2026-06-25 03:00:06 +08:00
  • 6bd3317856 Update Chinese tutorial link yqzhishen 2026-06-24 18:22:26 +08:00
  • 401045b976 Remove unnecessary assertion yqzhishen 2026-06-24 17:08:00 +08:00
  • e49240b2ef Config schemas: add new keys, restructure backbone_args, fix outdated defaults yqzhishen 2026-06-23 00:21:40 +08:00
  • 94ecd9cc3e Mix-LN: Mixed LayerNorm with speaker-conditioned mixup for acoustic encoder (#307) yqzhishen 2026-06-22 22:48:00 +08:00
  • 823547e3e6 Fix swapped initial bias mix-ln yqzhishen 2026-06-22 22:41:33 +08:00
  • f001ebd24a Add reference yqzhishen 2026-06-22 22:01:33 +08:00
  • 8d881570cb Fix duplicate layer_norm1 yqzhishen 2026-06-22 21:40:44 +08:00
  • bb0d1426ed Fix bad merging yqzhishen 2026-06-22 20:54:48 +08:00
  • 16e8a0bfc0 Default false yqzhishen 2026-06-22 20:43:16 +08:00
  • 17452039f1 Format yqzhishen 2026-06-22 20:35:12 +08:00
  • fbe5a146b5 Merge branch 'main' into mix-ln yqzhishen 2026-06-22 20:07:23 +08:00
  • 0482c1fbb3 Final landing of LYNXNet2 backbone and Muon optimizer (#306) yqzhishen 2026-06-22 17:47:13 +08:00
  • 20d129159a Add duration template muon_lynxnet2 wolfgitpr 2026-06-22 16:40:47 +08:00
  • 76c048849c Adjust configs && Add duration config wolfgitpr 2026-06-22 16:14:19 +08:00
  • 41d1cfd05d Revert "Adjust configs" wolfgitpr 2026-06-22 14:05:38 +08:00
  • fe2df4506d Adjust configs wolfgitpr 2026-06-22 13:47:15 +08:00
  • 40f646f1f0 Merge branch 'main' into muon_lynxnet2 yqzhishen 2026-06-21 00:28:56 +08:00
  • 2a8a013162 Unify training and ONNX-export environment, lift PyTorch 1.13 pin (#305) Kakaru 2026-06-21 00:18:58 +08:00
  • 5771a3bb5f perf: optimize Muon with GramNS and fix fp16 spectral norm stability (#301) Kakaru 2026-06-20 23:02:53 +08:00
  • 05c196ff82 Formatter fix v3 yqzhishen 2026-05-30 17:08:18 +08:00
  • 39fd90bf44 perf: change 'plt.pcolor' to 'plt.pcolormesh' (#302) Kakaru 2026-05-24 22:11:36 +08:00
  • 589c51d64e fix: Fixed the bug where torch.ceil caused sdp_dur_loss to lose its gradient (#303) sdp Kakaru 2026-05-24 22:11:20 +08:00
  • 5534e285a4 Code smells: sync framework yqzhishen 2026-05-21 04:47:17 +08:00
  • 36cd2df17c Fix typo yqzhishen 2026-05-09 20:55:19 +08:00
  • c2ae8afcfe fix typo (#300) Kakaru 2026-05-09 20:16:35 +08:00
  • 38dc794758 fix(core): correct vmin/vmax type assertion in repetitive flows (#299) 依旧在星空下等你 2026-05-09 19:08:34 +08:00
  • ae18130982 Fix config base model and formatter yqzhishen 2026-05-02 17:03:59 +08:00
  • 6317c37c10 Sync framework update yqzhishen 2026-05-02 15:38:05 +08:00
  • f50cfdd6bc Gram Newton Schulz (#297) polar_express Kakaru 2026-04-13 11:17:05 +08:00
  • bb0535f6ab Fix the issue where DP loss becomes infinite after SDP is enabled / Provide an intuitive plot method to observe the effect of SDP (#296) Kakaru 2026-03-27 21:09:34 +08:00
  • 6e144e5f2a Fix the missing parameter (use_sdp) (#295) Kakaru 2026-03-27 13:31:46 +08:00
  • ec9f2f8848 update SDP and Partial RoPE (#294) Kakaru 2026-03-26 15:32:32 +08:00
  • cbc2a1ce1e Fix scope context in __init__ yqzhishen 2026-03-16 15:24:00 +08:00
  • 423afe5137 Add type hint yqzhishen 2026-02-07 15:27:31 +08:00
  • 87ee89ad28 Support metric collection yqzhishen 2026-01-28 21:13:46 +08:00
  • 2d80e15d67 Reset metrics yqzhishen 2026-01-28 21:13:29 +08:00
  • b03d3e606a fix "input Linear" and "output Linear" optim by Muon (#285) Kakaru 2026-01-25 18:06:59 +08:00
  • fe72a38382 4 decimal places for metric key yqzhishen 2026-01-23 23:36:17 +08:00
  • 272b29280c Simplify logger yqzhishen 2026-01-23 23:35:57 +08:00
  • c770fd4ac3 Simplify RoPE yqzhishen 2026-01-23 23:31:43 +08:00
  • d2bada6c67 Use longest consecutive frames check-uv-voicing yqzhishen 2026-01-15 05:30:29 +08:00
  • 7aa2b84091 Check embed dim for RoPE yqzhishen 2026-01-14 21:54:01 +08:00
  • 1a96f26dfb Check embed dim for RoPE yqzhishen 2026-01-14 21:54:01 +08:00
  • ebc3805f94 Check embed dim for RoPE yqzhishen 2026-01-14 21:54:01 +08:00
  • e6f6a844af Change OpenUtau link yqzhishen 2026-01-14 21:52:31 +08:00
  • 12df7b3f03 Fix link to DiffScope project in README (#272) Crindzebra Sjimo 2026-01-14 21:27:19 +08:00
  • 45214f0585 Restore configs yqzhishen 2026-01-14 20:59:36 +08:00
  • 4f89f95b8b Disable phoneme coverage check yqzhishen 2026-01-14 16:45:55 +08:00
  • f46a4d8b1e Disable validation set check yqzhishen 2026-01-14 16:24:44 +08:00
  • 85d0754176 Add ph_map path yqzhishen 2026-01-14 16:11:55 +08:00
  • 106530513a Fix extra info yqzhishen 2026-01-14 01:00:19 +08:00
  • 1740c45f16 Fix language-specified phoneme check yqzhishen 2026-01-13 23:43:09 +08:00
  • 6697e08edd Add check_ph_voicing.py yqzhishen 2026-01-13 22:31:24 +08:00
  • 54b274fffe Fix strict=True yqzhishen 2026-01-13 16:04:10 +08:00
  • 40d69254ba Fix checkpoint compatibility yqzhishen 2026-01-13 15:51:51 +08:00
  • e1536e6636 Compute RoPE cache in init and make it read-only yqzhishen 2026-01-13 15:51:25 +08:00
  • faf7c2ba9b Keep T dim to avoid confusing yqzhishen 2026-01-12 22:01:50 +08:00
  • 8ad692bf3b Merge pull request #288 from KakaruHayate/fix_rope Kakaru 2026-01-12 21:59:00 +08:00
  • 3fe251ced8 fix yxlllc 2026-01-12 21:27:15 +08:00
  • 69ebaefdb7 fix padding of rmvpe and vr inferences yxlllc 2026-01-12 21:14:12 +08:00
  • 3f02382e2e Adapt frozen_spk_embed to Mix-LN yqzhishen 2026-01-12 21:32:13 +08:00
  • c84b457c00 Fix strict=True yqzhishen 2026-01-13 16:04:10 +08:00
  • e7ab9cd00b Fix strict=True yqzhishen 2026-01-13 16:03:41 +08:00
  • 8aa988691f Fix check version yqzhishen 2026-01-13 16:01:05 +08:00
  • 8f0dd994cd Fix checkpoint compatibility yqzhishen 2026-01-13 15:51:51 +08:00
  • 0d7f0545c8 Compute RoPE cache in init and make it read-only yqzhishen 2026-01-13 15:51:25 +08:00
  • 446494cb5d Fix checkpoint compatibility yqzhishen 2026-01-13 15:51:51 +08:00
  • 9bb3a9fe8a Compute RoPE cache in init and make it read-only yqzhishen 2026-01-13 15:51:25 +08:00
  • 74e8d32b4f Fix checkpoint compatibility yqzhishen 2026-01-13 15:51:51 +08:00
  • 4e298fe7e7 Compute RoPE cache in init and make it read-only yqzhishen 2026-01-13 15:51:25 +08:00
  • d47796c8cd Keep T dim to avoid confusing yqzhishen 2026-01-12 22:01:50 +08:00
  • 9ce87119f7 Keep T dim to avoid confusing yqzhishen 2026-01-12 22:01:50 +08:00
  • d8b38e66d8 Keep T dim to avoid confusing yqzhishen 2026-01-12 22:01:50 +08:00
  • 4ec3ea31a1 Merge pull request #288 from KakaruHayate/fix_rope Kakaru 2026-01-12 21:59:00 +08:00
  • 1c184abf2b Merge pull request #288 from KakaruHayate/fix_rope Kakaru 2026-01-12 21:59:00 +08:00
  • 0363b180c1 fix yxlllc 2026-01-12 21:27:15 +08:00
  • 629379dba5 fix padding of rmvpe and vr inferences yxlllc 2026-01-12 21:14:12 +08:00
  • 5ed5a922dd Restore configs yqzhishen 2026-01-12 21:43:37 +08:00