github/ai-toolkit
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Commit Graph

  • 9d6a9a0803 Fixed embedding scale for offloading for ltx 2.3. Was a new bug added today. main Jaret Burkett 2026-08-31 15:53:58 -06:00
  • 6940ebf533 Skip files where caching fails in a dataset instead of crashing. Jaret Burkett 2026-08-30 16:49:12 -06:00
  • e98109f213 Version bump Jaret Burkett 2026-08-30 11:02:22 -06:00
  • 74ed5fddb0 Fix issue with ltx caching Jaret Burkett 2026-08-30 11:01:52 -06:00
  • 764b5064fb Migrate to a new DTO for latents to carry more information that a normal tensor such as audio. Jaret Burkett 2026-08-30 10:30:21 -06:00
  • 2a69c1e7de Add initial support for Minimax H3 VSA sparse attention Jaret Burkett 2026-08-30 08:24:43 -06:00
  • be995185f5 Fixed issue with offloading with new model loader on nvfp4 weights. Jaret Burkett 2026-08-29 12:01:40 -06:00
  • 7380476b9c Version bump Jaret Burkett 2026-08-29 11:24:16 -06:00
  • 683fe8afc0 Stability improvements to model offloading. Added D-OPSD bleed loss as well. Jaret Burkett 2026-08-29 11:23:47 -06:00
  • 64a20f51a6 Merge pull request #1025 from ostris/models/v2 Jaret Burkett (Ostris) 2026-08-29 09:06:19 -06:00
  • bb55d38958 Version bump Jaret Burkett 2026-08-29 08:58:15 -06:00
  • 7195abc32b Performance fixes and bug fixes Jaret Burkett 2026-08-28 22:21:09 -06:00
  • 5ddc5f8ca7 Optimize quantization path Jaret Burkett 2026-08-28 11:47:29 -06:00
  • ce3df32101 Fix speed on wan conv patch Jaret Burkett 2026-08-28 10:54:26 -06:00
  • 92df289931 Fix quantization issues Jaret Burkett 2026-08-28 10:48:34 -06:00
  • 85a6880643 Move the offload and quantization to the base model Jaret Burkett 2026-08-28 08:37:56 -06:00
  • 702254688d Added legacy paths Jaret Burkett 2026-08-28 07:55:19 -06:00
  • c3bc8b0b4e Full test of ui arch compatability Jaret Burkett 2026-08-28 06:56:05 -06:00
  • 520d96aac3 Phase 2 Jaret Burkett 2026-08-27 18:18:31 -06:00
  • 45886f01b2 Phase 2 Jaret Burkett 2026-08-27 16:16:35 -06:00
  • 9113420b61 Phase 2 Jaret Burkett 2026-08-27 15:37:13 -06:00
  • 9ed2e0b8e7 Test model loading Jaret Burkett 2026-08-27 14:31:53 -06:00
  • 8db198ec0a Phase 1 Jaret Burkett 2026-08-27 11:53:08 -06:00
  • e8d9cf6d35 Models v2 - phase 0 Jaret Burkett 2026-08-27 10:51:43 -06:00
  • 5497a001cb Fix issue where errors on qwen image omni were not being fully surfaced. Jaret Burkett 2026-08-26 20:54:24 -06:00
  • da79ebce99 Add D-OPSD as a distillation handeling option for MiniMax H3 ref2va Jaret Burkett 2026-08-26 11:06:33 -06:00
  • 8a912564ce Fix video DOP methods Jaret Burkett 2026-08-26 10:05:14 -06:00
  • 8436c407f6 Add ability to delete the loss logs from a selected range on the loss graph Jaret Burkett 2026-08-23 09:57:47 -06:00
  • 27a03a91f2 Rework progress bar and ui speed string so it gets the same value at a more accurate and snappier wall clock time Jaret Burkett 2026-08-21 13:31:02 -06:00
  • b96476a841 Allow overriding the batch size on a per dataset level Jaret Burkett 2026-08-21 12:56:04 -06:00
  • 89102f76dc Allow user to pick if reference images are sent in the image or video reference stream for Hidream H3 ref2va Jaret Burkett 2026-08-19 13:32:06 -06:00
  • afd1d92722 Fix issue with double sampling qwen vl video frames for hidream h3 for reference videos Jaret Burkett 2026-08-17 15:39:07 -06:00
  • 42dfe9c661 Improve video frame reader Jaret Burkett 2026-08-16 10:21:11 -06:00
  • b982a03ae4 Add training adapter for MiniMax H3 ref2va and default to contrastive loss with adapter Jaret Burkett 2026-08-16 09:11:07 -06:00
  • 2042481914 On MiniMax H3 set the default to use both contrastive guidance and the training adapter as it is yielding better results than one or the other. Jaret Burkett 2026-08-16 08:43:31 -06:00
  • 61310c6397 Inprove cpu data acuisition on backend Jaret Burkett 2026-08-16 08:42:36 -06:00
  • 2cbc2bb097 On minimax h3, trim cached text embeddings to max tokens if they are longer than specified Jaret Burkett 2026-08-16 08:42:01 -06:00
  • 0f788923ae Animate action bar so starting a job has a 'working' indicator. Jaret Burkett 2026-08-15 10:33:40 -06:00
  • e6cffbc002 Adjust encoded paths so filenames resolve as a basename properly when downloading with things like wget. Jaret Burkett 2026-08-15 08:56:55 -06:00
  • 151ad0e959 Adjust MiniMah h3 sizing for videos to downscale to match Jaret Burkett 2026-08-15 07:50:42 -06:00
  • 70b1089359 Version bump Jaret Burkett 2026-08-15 07:14:13 -06:00
  • 127d6f626d Rework img/video reference in Minimax H3 to more closely match the comfy ui implementation. Jaret Burkett 2026-08-15 07:13:54 -06:00
  • f1faa7725b version bump Jaret Burkett 2026-08-15 06:20:32 -06:00
  • 97bf49edad Add support for video references in MiniMax H3 ref2va Jaret Burkett 2026-08-15 06:18:09 -06:00
  • 4900e5e866 Version bump Jaret Burkett 2026-08-14 19:54:51 -06:00
  • 5f53ecde54 Make the viewer thumbnails pul the thumbs Jaret Burkett 2026-08-14 18:14:12 -06:00
  • 247cb45c3e Improve gpu cpu monitor efficiency. Jaret Burkett 2026-08-14 17:14:00 -06:00
  • 695b0baccf Switch to a live always active gpu and cpu device monitor that can be connected to with an SSE connection for real time device information stream. Much more efficient and faster than previous method of polling. Jaret Burkett 2026-08-14 12:40:01 -06:00
  • 5261d3fcca Add checkpointing to the Wan2.1 encoder Jaret Burkett 2026-08-14 07:31:16 -06:00
  • 0e4b6e8695 Version bump Jaret Burkett 2026-08-13 12:36:34 -06:00
  • 6ea281973d Add support for MiniMax H3 Ref2Vid training Jaret Burkett 2026-08-13 12:36:03 -06:00
  • ab18528fdb Let omni captioner handle just images as well. Set it as the new default captioner. Jaret Burkett 2026-08-13 10:30:17 -06:00
  • 6b7fb60a22 Make contrastive guidance loss do constant instead of sigma loss schedule by default Jaret Burkett 2026-08-13 09:52:02 -06:00
  • 4e91fb2d0a Add thinking and abliterated versions of qwen omni. Jaret Burkett 2026-08-13 08:56:31 -06:00
  • 6c88e3d138 Add layer offloading to omni captioner Jaret Burkett 2026-08-13 07:21:54 -06:00
  • a69f3e8710 Version bump Jaret Burkett 2026-08-12 21:38:00 -06:00
  • 742a4c8cef Added a Download Full Dataset and Download Captions option to the datasets page so they captions or dataset can be easily downloaded. Jaret Burkett 2026-08-12 21:37:35 -06:00
  • 81adcc2176 Add caption prompt template picker to the ui. A few styling improvements thrown in for a good time. Jaret Burkett 2026-08-12 21:23:28 -06:00
  • ca42a72f4c Fixed issue with flash attention on omni captioner Jaret Burkett 2026-08-12 21:14:23 -06:00
  • 7f9a142dfd Generate thumbnails for dataset items so things load faster and the ui is more stable. Especially good for videos. Jaret Burkett 2026-08-12 20:33:42 -06:00
  • 175cc1e151 Add Qwen 3 Omni for captioning videos with sound. Jaret Burkett 2026-08-12 19:57:01 -06:00
  • 4b00b61257 When doing an audio loss, show the img and audio loss in the loss log Jaret Burkett 2026-08-12 18:58:54 -06:00
  • e16e04f123 On the ui, keep the last caption job bar visible at the bottom of a dataset so it can be viewed and edited Jaret Burkett 2026-08-12 17:32:48 -06:00
  • 18645d93b7 Fix trailing slashes in path for settings Jaret Burkett 2026-08-12 17:04:22 -06:00
  • a1ddeeef13 Fixed issue with offloading text encoder on ltx 2.5 Jaret Burkett 2026-08-12 10:46:58 -06:00
  • 0fd3e61c4c Expand loras to match if they are not the same size in lora merger Jaret Burkett 2026-08-12 10:04:44 -06:00
  • 0bacc88e47 Version Bump Jaret Burkett 2026-08-12 07:55:21 -06:00
  • 7eb65b837a Switch MiniMax H3 to default to contrastive guidance loss. Add a custom model select toggle to automatically fill out settings for the model. Jaret Burkett 2026-08-12 07:54:57 -06:00
  • cbf910ac02 Add support for LTX 2.5 Jaret Burkett 2026-08-12 05:50:30 -06:00
  • 924c426675 Resolve the new ltx 2.5 path subfolder for ltx 2.5. Sill checking compatability. Jaret Burkett 2026-08-11 13:30:17 -06:00
  • 62017a915a Version bump Jaret Burkett 2026-08-11 08:30:09 -06:00
  • 21dc65972d Only add a blank control if the model has to have it, for unconditionals. Previously it always encoded a blank control on unconditional if the model could take one. Affects DOP and blank prompt preservations. Jaret Burkett 2026-08-11 08:25:23 -06:00
  • f421542df4 When dropping out, caching, and doing DOP, make sure we select a cahced trigger word when dropped out so DOP matches the drop out embeddings. Jaret Burkett 2026-08-11 06:59:20 -06:00
  • 8d4beedd04 Move guidance loss target selection when given a range before preservation so it is avaliable during preservation. Jaret Burkett 2026-08-11 06:57:33 -06:00
  • ab5fef8970 Fix caption dropout so it now works identically when caching text embeddings. Jaret Burkett 2026-08-10 18:53:33 -06:00
  • 356ce7e84e Version bump Jaret Burkett 2026-08-10 09:21:27 -06:00
  • 257da9b586 Rework DOP so it works with caching text embeddings Jaret Burkett 2026-08-09 22:13:49 -06:00
  • 5ff8a0435a Make v1 the default training adapter Jaret Burkett 2026-08-09 14:56:39 -06:00
  • 61da9c95d3 Version bump Jaret Burkett 2026-08-09 12:39:30 -06:00
  • 72623ed3d6 When doing auto frame count. Ensure the time is not squeezed or expanded to fit tempooral spacing. trime the few extra frames. Also fixed frame counts of buckets. Jaret Burkett 2026-08-09 12:31:02 -06:00
  • 682b27c6ee Reworked freeing memory manager for removing text encoder completly when not needed. Jaret Burkett 2026-08-09 07:08:24 -06:00
  • 3a28c4b1b7 Keep alive cron server to prevent failed connections Jaret Burkett 2026-08-08 20:27:01 -06:00
  • 8c1a4082fd Allow images to work with auto frame count, and include images in video datasets if they exist. Jaret Burkett 2026-08-08 20:26:28 -06:00
  • 6d8afa5684 Version bump Jaret Burkett 2026-08-08 18:58:11 -06:00
  • c596d4ab27 Merge pull request #1002 from whatsthisaithing/codex/fix-convrot-offload-stream-lifetime Jaret Burkett (Ostris) 2026-08-08 18:55:18 -06:00
  • d184c6c622 Fix offload buffer stream lifetime Fitzy 2026-08-08 19:59:23 -04:00
  • f4e9130547 Fix race condition that can corrupt grads under certain conditions. Jaret Burkett 2026-08-07 14:52:47 -06:00
  • 817f3dcbcb Fix finite check on inverted masked prior Jaret Burkett 2026-08-07 09:49:12 -06:00
  • 685ce37a8d Adjust sample timestep sigmas to be model evals for h3 for 1 extra step Jaret Burkett 2026-08-06 19:54:04 -06:00
  • 9171d5ec1d Update training adapter path Jaret Burkett 2026-08-06 11:21:28 -06:00
  • 71625d1207 Add the alpha version of the MiniMax H3 trianing adapter and set it as the new default training method. Jaret Burkett 2026-08-06 11:11:31 -06:00
  • b904b99705 varsion bump Jaret Burkett 2026-08-06 09:25:35 -06:00
  • b811636ae4 Check for finite vs isnan on loss before backpropigating. to catch infinity overflows Jaret Burkett 2026-08-06 09:18:56 -06:00
  • edacd406b3 Handle minimax loading of non pruned model Jaret Burkett 2026-08-06 09:18:00 -06:00
  • 7309db4d74 Fix layer offloaded adaln projection layer move Jaret Burkett 2026-08-06 08:10:20 -06:00
  • 139a38f5bd Upcast adaln pruned layers to fp32 on h3 to prevent overflow. Jaret Burkett 2026-08-06 07:51:15 -06:00
  • 1e1418b22c Apply sigma to contrastive guidance to balance loss better. Prevent noise grads on images/non audio datasets. Prep for training adapters on MiniMax H3 Jaret Burkett 2026-08-05 13:27:12 -06:00
  • 9065951da3 Version bump Jaret Burkett 2026-08-04 22:14:54 -06:00
  • 3afa270ab5 Fix audio losses for DOP and other preservation losses Jaret Burkett 2026-08-04 22:08:48 -06:00
  • 0f9094db95 Fix audio loss when doing do_guidance_loss Jaret Burkett 2026-08-04 21:43:31 -06:00