Commit Graph
1624 Commits
Author SHA1 Message Date
Jaret Burkett c22f21fa95 Fix potential rotation mismatch issue and add webp support for the inactive fast image size checker 2026-10-21 07:00:36 -06:00
Jaret Burkett 04733873cf Improve performance of encoding Moss prompts with a batch size > 1 2026-10-19 10:15:45 -06:00
Jaret Burkett 2d4c89d392 Fix issue with flash linear attention in docker image 2026-10-18 20:12:28 -06:00
Jaret Burkett 31f4d92f88 Version Bump 2026-10-18 13:46:14 -06:00
Jaret Burkett 4370d60904 Remove the compile for qwen3.8 since the kernels are optimized already 2026-10-18 13:30:04 -06:00
Jaret Burkett 85d01c6a04 Prevent possible traversal upstream through caption tags on captioner 2026-10-18 13:05:41 -06:00
Jaret Burkett 7c6b168c2c Add tokens that will pull files with extensions git status 2026-10-18 13:04:58 -06:00
Jaret Burkett e72468b456 Added initial support for fine tuning and captioning with Qwen 3.8 27B 2026-10-18 12:25:28 -06:00
Jaret Burkett 8cf43dd2cc Fix issue that left text encoder on gpu after caching text embeddings. 2026-10-17 17:22:42 -06:00
Jaret Burkett 7b94dc5d8d Add info about the Comfy UI node for the MOSS image edit loras 2026-10-17 15:46:40 -06:00
Jaret Burkett 577004a9cd Publish libdevice as a module global so older triton builds without closure capture can compile the convrot kernels 2026-10-17 14:54:32 -06:00
Jaret Burkett 9d408eb64d Fixed issue with Flux.3 controls where references were not scaled properly and more than one reference could be ignored in certain instances 2026-10-17 14:43:02 -06:00
Jaret Burkett 88360c68b2 Speed up MOSS-Image edit kv_cache training: reference isolation as unmasked flash attention instead of a boolean mask 2026-10-17 12:10:31 -06:00
Jaret Burkett 9a1828f201 Save qkv lora weights for MOSS Image as unfused LoRA, so they are more compatable with both fused and unfused weights. 2026-10-17 09:55:04 -06:00
Jaret Burkett 82e6311ce7 Add MOSS-Image SFT edit training 2026-10-16 17:07:13 -06:00
Jaret Burkett dcebc7442d Version Bump 2026-10-16 12:19:35 -06:00
Jaret Burkett b7cd8a13af Add support to train MOSS Image Base as an edit model 2026-10-16 12:19:06 -06:00
Jaret Burkett 0d19b26481 Version bump 2026-10-16 09:55:06 -06:00
Jaret Burkett c6a9b56ae9 Add support for Moss Image Base 2026-10-16 09:54:38 -06:00
Jaret Burkett 80cb872a84 Fall back to an online load when a cached tokenizer/processor loads incomplete 2026-10-16 09:38:30 -06:00
Jaret Burkett 29f5ff452b Fix MOSS-Image tokenizer loading on a fresh HF cache (missing chat template) 2026-10-16 09:33:05 -06:00
Jaret Burkett e05a3a4f2c Add support for MOSS Image 2026-10-16 08:09:45 -06:00
Jaret Burkett 12395a103f Load comfy files if they exist in flux.3 image 2026-10-15 11:18:14 -06:00
Jaret Burkett 438595b78b Fixed issue with caching text embeddings on Flux.3 Image 2026-10-15 09:56:40 -06:00
Jaret Burkett 0947356c62 Fix issue with loading flux3 text encoders in comfy format 2026-10-15 09:42:15 -06:00
Jaret Burkett ef75577394 Version bump 2026-10-15 09:23:26 -06:00
Jaret Burkett (Ostris) 91bdc6dd15 Merge pull request #1078 from ostris/flux3-image
Add Support for FLUX.3 Image Klein Base
2026-10-15 09:18:34 -06:00
Jaret Burkett 9390ca463e Merge branch 'main' into flux3-image 2026-10-15 08:26:39 -06:00
Jaret Burkett 831ccfa7aa Added experimental band loss 2026-10-15 08:26:12 -06:00
Jaret Burkett 4f5b5cc7ad flux3: store DiT and VAE in the comfy layout with renamed downloads 2026-10-14 14:06:19 -06:00
Jaret Burkett 3530876838 Merge branch 'main' into flux3-image 2026-10-14 13:41:20 -06:00
Jaret Burkett 7045370346 Keep cached pixel tensors as uint8 in the dataloader workers and only materialize to fp16 on the main thread 2026-10-14 13:40:49 -06:00
Jaret Burkett dbbb7ee8f4 flux3: multi-reference sampling, match_target_res reference sizing, UI sections 2026-10-14 13:37:40 -06:00
Jaret Burkett d4ffbcad70 flux3: drop dead gate_type path, reject flow-map packages, note refreshed base checkpoint 2026-10-14 13:03:00 -06:00
Jaret Burkett 7a64fdafad Merge in from main 2026-10-14 11:27:16 -06:00
Jaret Burkett f7a1fb9a77 Version bump 2026-10-09 21:09:34 -06:00
Jaret Burkett f370e7441d Add support for Qwen Image 2 Turbo with training adapter 2026-10-09 21:09:09 -06:00
Jaret Burkett 2fde764876 Version bump 2026-10-09 05:14:08 -06:00
Jaret Burkett 2fa0bd87e6 Add support for Iris-3B 2026-10-09 05:13:39 -06:00
Jaret Burkett cb51e9a511 Add flux3 image 2026-10-09 03:08:46 -06:00
Jaret Burkett a949f21f8e Add ability to merge dataset buckets. Useful when using numerous different datasets. 2026-10-08 14:04:30 -06:00
Jaret Burkett 0cbac2c1f3 Add threadding to bucket resizing of a batch. Helpful to speed up large batch gathering. 2026-10-08 11:26:48 -06:00
Jaret Burkett ecee894ed2 Key table rows by id so deleted rows don't leak pending state to neighbors 2026-09-27 10:19:02 -06:00
Jaret Burkett 6468a2ffa7 Fix issue loading new comfy checkpoints with ming 2026-09-26 13:32:05 -06:00
Jaret Burkett 16ae8d2999 Fixed issue on DGX spark that caused unnessary moves of models causing slow down on unified memory 2026-09-26 11:56:42 -06:00
Jaret Burkett 60d0c28c4d Switch to new training adapters for all minimax models that no longer need contrastive guidance 2026-09-24 18:48:04 -06:00
Jaret Burkett a776394353 Add support to train on Fast H3 v2 8 step directly with training adapter 2026-09-24 18:40:17 -06:00
Jaret Burkett 5aa3d0c9a8 Update ui to show DGX spark gpu usage correctly 2026-09-24 17:55:11 -06:00
Jaret Burkett 60ca869b38 Add proper support for DGX spark with easy manager. Thank you NVIDIA! 2026-09-24 17:44:40 -06:00
Jaret Burkett cbc6bbdd70 Update ming image repo to new one 2026-09-24 08:40:16 -06:00