ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
486 commits 391 branches 58 tags 304 MiB
Commit graph

486 commits

This branch
This branch
All branches
Author SHA1 Message Date
Roland Tannous
a7fe8a388c Filter GGUF models from training page model selectors 
GGUF models can't be fine-tuned, so hide them from the training/studio
page while keeping them available for inference on the chat page.

- Add "gguf" to EXCLUDED_TAGS in HF model search hook
- Filter local models with .gguf extension or -GGUF in ID
 2026-02-25 15:47:45 +04:00
Roland Tannous
299ce77467 added vision.py patch for vision processor from PR#260 2026-02-25 11:39:17 +00:00
Roland Tannous
6f0b7bc38a fix: use raw github URL for vision.py patch + add VLM processor diagnostic logging 2026-02-25 10:29:05 +00:00
Roland Tannous
7adb69581e Fix GGUF export cwd confusion: remove os.chdir, use absolute paths 
Remove os.chdir(save_directory) from export.py which was causing all of
unsloth-zoo's relative-path internals (check_llama_cpp, use_local_gguf,
_download_convert_hf_to_gguf) to resolve against the export directory
instead of the repo root. This caused llama.cpp to be cloned inside each
export dir and destroyed the repo root's llama-server build on cleanup.

Now passes absolute paths to save_pretrained_gguf so unsloth resolves
llama.cpp from the repo root where setup.sh already built it.

Also builds llama-quantize in setup.sh (needed by unsloth-zoo's export
pipeline) and symlinks it to llama.cpp root for check_llama_cpp().
 2026-02-25 03:30:54 +04:00
Roland Tannous
0e7c8a2e5e Switch GGUF backend from /v1/completions to /v1/chat/completions 
Fixes two bugs:
1. Chat template tags (<|im_start|>, <|im_end|>) leaking into output
   because /v1/completions treated them as literal text
2. Image hallucination because image_b64 was never passed to llama-server

Now llama-server handles chat templates natively and receives images
as OpenAI-format multimodal content parts for vision models.
 2026-02-24 19:21:01 +04:00
Roland Tannous
ef1cd3ac98 Use llama-server -hf mode, add GGUF variant selector, fix vision detection 
Replace Python-side GGUF download with llama-server's native -hf flag for
HuggingFace repos. Add frontend variant picker so users can choose
quantization (Q4_K_M, Q8_0, BF16, etc.) with file sizes. Fix vision
detection via mmproj files instead of hardcoding is_vision=False.
 2026-02-24 19:03:06 +04:00
Roland Tannous
08aeeaee4b Fix llama-server: build in-tree, fix path resolution, add LD_LIBRARY_PATH 2026-02-24 18:19:29 +04:00
Roland Tannous
4e88092452 Preflight llama-server check before downloading remote GGUF files 2026-02-24 18:02:43 +04:00
Roland Tannous
a900eb9ad7 Fix GGUF detection for HuggingFace repo IDs (not just local paths) 2026-02-24 17:49:09 +04:00
Roland Tannous
70c912d788 Fix CUDA detection for llama-server build on multi-GPU machines 2026-02-24 17:45:17 +04:00
Roland Tannous
a40ebb1aab Add GGUF model inference via llama-server backend 2026-02-24 17:40:05 +04:00
Roland Tannous
77b0978d5f Merge pull request #228 from unslothai/fix/cap-num-proc-multigpu-deadlock 
Cap dataset.map num_proc on multi-GPU machines to prevent fork deadlocks
 2026-02-23 19:04:22 +04:00
Roland Tannous
d74174f7f5 Cap dataset.map num_proc on multi-GPU machines to prevent fork deadlocks 2026-02-23 14:25:31 +00:00
Roland Tannous
4cb0cfdaf5 Remove firebase-debug.log and setup_leo.sh from tracking and add to .gitignore 2026-02-23 17:38:22 +04:00
Roland Tannous
2e2aa54ad2 Merge pull request #225 from unslothai/fix/fix-response-on-completion-truncation 
fix: error on >30% sample drop after `train_on_responses_only` instead of silent DataLoader crash
 2026-02-23 16:28:32 +04:00
Roland Tannous
3015916d26 fix: error on >30% sample drop after train_on_responses_only instead of silent DataLoader crash 2026-02-23 12:21:06 +00:00
Roland Tannous
b03938f6ad Merge nightly into main Brings main up to date with nightly, including chat attachments, model-per-thread persistence, speech recognition, VRAM recommendations, MoE model configs, VLM fixes, and compile cache cleanup. Conflicts resolved by taking nightly's version for all diverged files (main-only changes were a feature add + immediate revert with net zero effect). 2026-02-23 14:52:57 +04:00
Roland Tannous
bfb84221fe Merge pull request #221 from unslothai/feature/clear-unsloth-compile-cache 
feat: clear unsloth_compiled_cache on startup, shutdown, and between …
 2026-02-23 11:30:46 +04:00
Roland Tannous
dbbcdb4f09 feat: clear unsloth_compiled_cache on startup, shutdown, and between model loads 2026-02-23 07:26:22 +00:00
Roland Tannous
2a117f57a9 Merge pull request #220 from unslothai/feature/moe-training-models-configs 
Add model defaults for MoE models (Qwen3 MoE, GLM Flash) and GLM response mapping
 2026-02-23 10:09:27 +04:00
Roland Tannous
fb1c321ad3 Add GLM, Qwen3 MoE, TinyQwen3 MoE, and Ministral 3 VL model defaults and GLM train_on_responses_only mapping 2026-02-23 05:51:43 +00:00
Roland Tannous
bb1bd49a68 Merge pull request #217 from unslothai/fix/update-config-yamls 
Fix vision LoRA defaults for VLMs and clean up text-only model configs
 2026-02-22 19:14:54 +04:00
Roland Tannous
132cdb0547 fix: correct vision LoRA defaults for VLMs and remove vision fields from text-only model configs 2026-02-22 15:09:10 +00:00
Roland Tannous
7ac391d1e0 Merge pull request #215 from unslothai/fix/vlm-processing-class 
fix: pass full Processor as processing_class for VLM SFTTrainer
 2026-02-22 18:14:14 +04:00
Roland Tannous
202b7cdfa7 fix: pass full Processor as processing_class for VLM SFTTrainer 2026-02-22 14:11:12 +00:00
Roland Tannous
a4346b954e Merge pull request #213 from unslothai/fix/fix-clear-chat-new-model 
Fix: Clear Chat and Manage Model Lifecycle on Model Switch
 2026-02-22 17:54:27 +04:00
Roland Tannous
761953b50e feat(chat): persist model per thread and auto-load on thread switch 2026-02-22 13:35:45 +00:00
Roland Tannous
536a735acc feat(chat): eject current model and start fresh thread on model switch 2026-02-22 13:17:03 +00:00
Roland Tannous
a490dca8a0 Merge pull request #211 from unslothai/fix/param-count-display 
Fix: Remove download count fallback when model param count is unavailable
 2026-02-22 16:52:25 +04:00
Roland Tannous
d3fbaa2256 Merge pull request #204 from unslothai/fix/image-preview-thumbnail 
Fix: Resolve image preview thumbnail not rendering before send
 2026-02-22 16:09:24 +04:00
Roland Tannous
36f10ea90e Merge pull request #212 from unslothai/fix/stop-unclickable 
Updated stop button to be unavailable during cancel training
 2026-02-22 16:08:47 +04:00
Roland Tannous
c5a88ef90b Merge pull request #208 from unslothai/revert-207-feature/attachment-restore 
Revert "fix(chat): persist + hydrate user attachments in IndexedDB history"
 2026-02-22 12:22:25 +04:00
Roland Tannous
df8e4bdc40 Merge pull request #203 from unslothai/feat/sort-unsloth-models-first 
Feat: Sort unsloth models first in HF search dropdowns
 2026-02-22 12:21:42 +04:00
Roland Tannous
75f3e5e2a1 feat: dual-query HF model search to surface all unsloth size variants first 2026-02-22 08:20:25 +00:00
imagineer99
4fe3772aa2 fix: remove download count fallback when model param count is unavailable 2026-02-22 06:30:50 +00:00
Wasim Yousef Said
4779ae8e61 Merge pull request #209 from unslothai/feature/attachment-restore 
fix(chat): persist + hydrate user attachments in IndexedDB history
 2026-02-21 22:17:18 -08:00
Roland Tannous
ac83e8b668 Revert "fix(chat): persist + hydrate user attachments in IndexedDB history" 2026-02-22 10:15:24 +04:00
Wasim Yousef Said
4f2e434bc3 Merge pull request #207 from unslothai/feature/attachment-restore 
fix(chat): persist + hydrate user attachments in IndexedDB history
 2026-02-21 22:09:53 -08:00
Shine1i
c726cff4c8 feat: add utils for deep cloning content and attachments in thread messages 2026-02-22 07:08:18 +01:00
imagineer99
52a2bfe016 fix: resolve image preview thumbnail not rendering before send 2026-02-22 02:39:56 +00:00
imagineer99
c815fc045d feat: sort unsloth models first in HF search dropdowns 2026-02-22 01:45:51 +00:00
Roland Tannous
000da20237 Merge pull request #202 from unslothai/fix/rename-downloading-model-to-loading-model 
renaming downloading model to loading model
 2026-02-21 17:32:32 +04:00
Roland Tannous
9bc32789a6 renaming downloading model to loading model 2026-02-21 13:30:41 +00:00
Roland Tannous
44828f582f Merge pull request #200 from unslothai/fix/sloth-z-index-overlay 
fix: enable navbar z-index by adding relative position
 2026-02-21 17:15:05 +04:00
Roland Tannous
ea324f2806 Merge pull request #193 from unslothai/feature/vram-fit-chat 
Added vram fit indicator to models in chat
 2026-02-21 17:09:31 +04:00
Roland Tannous
3a4a576128 Merge pull request #196 from unslothai/fix/compare-dictate-attachment-buttons 
Added dictate and add attachments feature in chat page
 2026-02-21 16:52:56 +04:00
Roland Tannous
9b3cab6d5f move microphone icon in compare page to be next to send button 2026-02-21 12:51:38 +00:00
samit
3fb9b4c056 added SpeechRecognition declarations and missing type packages 2026-02-21 01:00:08 -08:00
imagineer99
33b31c6be6 fix: enable navbar z-index by adding relative position 2026-02-21 08:37:47 +00:00
samit
f4a888cddb Added VRAM fit indicator to recoomended models 2026-02-20 23:55:10 -08:00