Commit Graph
841 Commits
Author SHA1 Message Date
Jaret Burkett b764c4a94e DFE cleanup 2026-09-09 06:42:24 -06:00
Jaret Burkett af82197907 Fix gradient checkpointing for sapiens DFE 2026-09-08 12:24:56 -06:00
Jaret Burkett af89a27e0f Fixed issue with partial offloading torchao quants 2026-09-08 05:15:26 -06:00
Jaret Burkett 3e1cb5f67f Rework DFEs to work better with x0 models. Show their loss in the graph 2026-09-07 13:26:01 -06:00
Jaret Burkett 7690ea62c8 Add variable precision adam convrot optimizer with stochastic rounding and accumulation. 2026-09-06 10:51:35 -06:00
Jaret Burkett (Ostris) 2a2ca26ee8 Merge branch 'main' into fix/dataloader-pin-memory 2026-09-05 08:08:39 -06:00
Jaret Burkett ecb2bfbe5e Fixed issue where some unmanaged modules could be lefs on cpu 2026-09-02 09:22:39 -06:00
Jaret Burkett 9d6a9a0803 Fixed embedding scale for offloading for ltx 2.3. Was a new bug added today. 2026-08-31 15:53:58 -06:00
Jaret Burkett 6940ebf533 Skip files where caching fails in a dataset instead of crashing. 2026-08-30 16:49:12 -06:00
Jaret Burkett 74ed5fddb0 Fix issue with ltx caching 2026-08-30 11:01:52 -06:00
Jaret Burkett 764b5064fb Migrate to a new DTO for latents to carry more information that a normal tensor such as audio. 2026-08-30 10:30:21 -06:00
Jaret Burkett be995185f5 Fixed issue with offloading with new model loader on nvfp4 weights. 2026-08-29 12:01:40 -06:00
Jaret Burkett 683fe8afc0 Stability improvements to model offloading. Added D-OPSD bleed loss as well. 2026-08-29 11:23:47 -06:00
Jaret Burkett 7195abc32b Performance fixes and bug fixes 2026-08-28 22:21:09 -06:00
Jaret Burkett 5ddc5f8ca7 Optimize quantization path 2026-08-28 11:47:29 -06:00
hitoshic1982 5fc8fd76de Fix cache_text_encoder device preset missing in BaseModel
BaseModel.set_device_state_preset() has no branch for the
'cache_text_encoder' preset, so when TextEmbeddingCachingMixin calls
set_device_state_preset('cache_text_encoder') the active_modules list
stays empty and the text encoder is placed on CPU. With
cache_text_embeddings enabled on a BaseModel arch (e.g. chroma), every
caption is then encoded on CPU — with a quantized T5-XXL this took
~6.5 minutes per caption (~21 hours for a 205-image dataset) while the
GPU sat idle. StableDiffusion.set_device_state_preset() already has
this branch; this adds the same two lines to BaseModel.
2026-08-29 01:37:39 +08:00
Jaret Burkett ce3df32101 Fix speed on wan conv patch 2026-08-28 10:54:26 -06:00
Jaret Burkett 92df289931 Fix quantization issues 2026-08-28 10:48:34 -06:00
Jaret Burkett 85a6880643 Move the offload and quantization to the base model 2026-08-28 08:37:56 -06:00
Jaret Burkett 702254688d Added legacy paths 2026-08-28 07:55:19 -06:00
Jaret Burkett c3bc8b0b4e Full test of ui arch compatability 2026-08-28 06:56:05 -06:00
Jaret Burkett 520d96aac3 Phase 2 2026-08-27 18:18:31 -06:00
Jaret Burkett 45886f01b2 Phase 2 2026-08-27 16:16:35 -06:00
Jaret Burkett 9113420b61 Phase 2 2026-08-27 15:37:13 -06:00
Jaret Burkett 9ed2e0b8e7 Test model loading 2026-08-27 14:31:53 -06:00
Jaret Burkett 8db198ec0a Phase 1 2026-08-27 11:53:08 -06:00
Jaret Burkett e8d9cf6d35 Models v2 - phase 0 2026-08-27 10:51:43 -06:00
Jaret Burkett da79ebce99 Add D-OPSD as a distillation handeling option for MiniMax H3 ref2va 2026-08-26 11:06:33 -06:00
Jaret Burkett 8a912564ce Fix video DOP methods 2026-08-26 10:05:14 -06:00
Jaret Burkett 27a03a91f2 Rework progress bar and ui speed string so it gets the same value at a more accurate and snappier wall clock time 2026-08-21 13:31:02 -06:00
Jaret Burkett b96476a841 Allow overriding the batch size on a per dataset level 2026-08-21 12:56:04 -06:00
Jaret Burkett afd1d92722 Fix issue with double sampling qwen vl video frames for hidream h3 for reference videos 2026-08-17 15:39:07 -06:00
Jaret Burkett 127d6f626d Rework img/video reference in Minimax H3 to more closely match the comfy ui implementation. 2026-08-15 07:13:54 -06:00
Jaret Burkett 97bf49edad Add support for video references in MiniMax H3 ref2va 2026-08-15 06:18:09 -06:00
Jaret Burkett 5261d3fcca Add checkpointing to the Wan2.1 encoder 2026-08-14 07:31:16 -06:00
Jaret Burkett 6b7fb60a22 Make contrastive guidance loss do constant instead of sigma loss schedule by default 2026-08-13 09:52:02 -06:00
Jaret Burkett 7f9a142dfd Generate thumbnails for dataset items so things load faster and the ui is more stable. Especially good for videos. 2026-08-12 20:33:42 -06:00
Jaret Burkett a1ddeeef13 Fixed issue with offloading text encoder on ltx 2.5 2026-08-12 10:46:58 -06:00
Jaret Burkett 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. 2026-08-11 08:25:23 -06:00
Jaret Burkett 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. 2026-08-11 06:59:20 -06:00
Jaret Burkett ab5fef8970 Fix caption dropout so it now works identically when caching text embeddings. 2026-08-10 18:53:33 -06:00
Jaret Burkett 257da9b586 Rework DOP so it works with caching text embeddings 2026-08-09 22:13:49 -06:00
Jaret Burkett 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. 2026-08-09 12:31:02 -06:00
Jaret Burkett 682b27c6ee Reworked freeing memory manager for removing text encoder completly when not needed. 2026-08-09 07:08:24 -06:00
Jaret Burkett 8c1a4082fd Allow images to work with auto frame count, and include images in video datasets if they exist. 2026-08-08 20:26:28 -06:00
Fitzy d184c6c622 Fix offload buffer stream lifetime 2026-08-08 19:59:23 -04:00
Jaret Burkett f4e9130547 Fix race condition that can corrupt grads under certain conditions. 2026-08-07 14:52:47 -06:00
Jaret Burkett 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 2026-08-05 13:27:12 -06:00
Jaret Burkett 3afa270ab5 Fix audio losses for DOP and other preservation losses 2026-08-04 22:08:48 -06:00
Jaret Burkett 0f9094db95 Fix audio loss when doing do_guidance_loss 2026-08-04 21:43:31 -06:00