ceos/unsloth
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
1,025 commits 391 branches 58 tags 304 MiB
Commit graph

209 commits

Author SHA1 Message Date
Manan17
5ca623a166 fixing gguf export for gemma with text 2026-03-11 00:58:22 +00:00
Manan17
bc5a72dd8c fix: local directory dataset loading 2026-03-10 21:29:51 +00:00
Manan Shah
f696ef81e8 Merge branch 'nightly' into feat/embedding-models 2026-03-10 14:16:05 -07:00
Roland Tannous
08d9c84f1f Merge pull request #359 from unslothai/fix/stream-manual-slice-dataset 
fix: stream HF dataset when manual slice is specified
 2026-03-11 01:13:51 +04:00
Manan17
3b0b002b34 fixing logging for each step 2026-03-10 20:32:40 +00:00
Roland Tannous
21ef22a9ff fix: skip streaming when dataset_slice_start > dataset_slice_end 
Prevents training on the wrong row range when start exceeds end by
falling back to full download where existing clamping handles it.
 2026-03-10 20:21:34 +00:00
Roland Tannous
226f251589 fix: guard against negative dataset_slice_end before streaming 
Fall back to full download when dataset_slice_end is negative,
avoiding an empty stream.take(0) that would produce a broken dataset.
 2026-03-10 20:12:42 +00:00
Roland Tannous
970a029108 fix: stream HF dataset when manual slice is specified 
Instead of downloading the full dataset and then slicing, use
streaming mode to only fetch the rows needed (up to slice_end + 1)
when a manual dataset slice is configured.
 2026-03-10 19:50:53 +00:00
Roland Tannous
c986174c56 fix: preserve zero-valued dataset slice boundaries in embedding worker 
Use explicit None checks instead of falsy `or` for slice_start and
slice_end so that a valid slice_end=0 is not replaced with the full
dataset length.
 2026-03-10 19:33:10 +00:00
Roland Tannous
b84202e8db fix: restrict shard siblings to exact basename and total count 
startswith(prefix) could match unrelated split variants whose names
extend the selected file's prefix (e.g. model-Q8_0-v2-00001-of-...).
Now builds an exact regex from the chosen file's base prefix and shard
total so only true siblings are downloaded.
 2026-03-10 19:28:26 +00:00
Roland Tannous
b8678a3ed6 fix: pass hf_token for gated embedding models and key cache by token 
- Forward hf_token to FastSentenceTransformer.from_pretrained() so
  private/gated embedding repos authenticate correctly
- Key _embedding_detection_cache by (model_name, hf_token) tuple so
  unauthenticated lookups don't shadow subsequent authenticated ones
 2026-03-10 19:20:12 +00:00
Roland Tannous
d635846b8d fix: use exact variant matching and shard-prefix discovery for split GGUFs 
Substring matching (e.g. "Q8_0" in filename) could match superset
variants like "IQ8_0", causing wrong quantizations to be downloaded.
Now uses word-boundary regex for variant matching and discovers split
shards by shared filename prefix rather than treating all variant
matches as shards.
 2026-03-10 19:13:03 +00:00
Roland Tannous
d6ae910edc fix: propagate is_embedding into worker subprocess config 
start_training() cherry-picks kwargs into a config dict but was missing
is_embedding, so config.get("is_embedding", False) in worker.py always
returned False and embedding training never ran.
 2026-03-10 19:05:47 +00:00
Roland Tannous
defa761fb2 fix: download all GGUF shards for split models (e.g. 7B Q8_0) 
LlamaCppBackend.load_model() only downloaded the first matching GGUF
file. For split models (e.g. 7B Q8_0 with 3 shards), llama-server
needs all shards present. Now collects and downloads all matching files.
 2026-03-10 19:04:10 +00:00
Roland Tannous
5a086353ab feat: add embedding model training support 
Add end-to-end embedding/sentence-transformer training pipeline using
FastSentenceTransformer, SentenceTransformerTrainer, and
MultipleNegativesRankingLoss with BatchSamplers.NO_DUPLICATES.

Backend:
- Add is_embedding_model() detection via HF tags + pipeline_tag
- Add /check-embedding/ API route and EmbeddingCheckResponse
- Extend derive_model_type() to return "embeddings"
- Add _run_embedding_training() in worker.py with progress callbacks,
  stop handling, LoRA (task_type=FEATURE_EXTRACTION), and model saving
- Add is_embedding field to TrainingStartRequest and ModelDetails
- Add YAML configs for 5 models: all-MiniLM-L6-v2, bge-m3,
  embeddinggemma-300m, gte-modernbert-base, Qwen3-Embedding-0.6B

Frontend:
- Wire isEmbeddingModel flag through store, API types, and mappers
- Force packing=false, train_on_completions=false, warmup_ratio=0.03
- Hide packing and train_on_completions checkboxes for embedding models
- Auto-set modelType to "embeddings" from backend model_type response
 2026-03-10 18:10:09 +00:00
Roland Tannous
a26a5cc6be Merge pull request #352 from unslothai/fix/cancel-training 
Fix/cancel training
 2026-03-10 14:38:30 +04:00
Manan17
fd7ca8bda8 distinguish cancel and stop for force terminate 2026-03-10 02:35:32 +00:00
Manan17
9be55f0c1b fixing cancel training 2026-03-10 02:20:56 +00:00
Roland Tannous
daa50d0756 Revert "Merge pull request #347 from unslothai/feature/studio-storage-roots" 
This reverts commit 6b43e33ff1, reversing
changes made to 9edadaf21f.
 2026-03-10 01:52:47 +00:00
Shine1i
5301514775 feat(studio): studio storage roots path utilities 2026-03-09 23:48:31 +00:00
Roland Tannous
d882678fe4 Add AGPL-3.0 SPDX headers to all source files 2026-03-09 20:17:45 +00:00
Shine1i
2ccb75f2b7 Merge remote-tracking branch 'origin/nightly' into feature/fixes-client 2026-03-09 19:07:42 +01:00
Roland Tannous
b6811bc5c4 Merge pull request #342 from unslothai/local-dataset 
dataset upload
 2026-03-09 21:22:23 +04:00
Roland Tannous
41351e1566 fix: split dataset 80/20 when eval split matches train split 2026-03-09 16:36:44 +00:00
Shine1i
542d9126cc chore(data-recipe): bump data-designer to 0.5.2 and pin duckdb<1.5 2026-03-09 17:27:02 +01:00
Roland Tannous
2a11e79b8b fix: restore eval_enabled early signal for subprocess training 2026-03-09 15:35:49 +00:00
Roland Tannous
c3185d5d98 fix: allow eval-only progress events through worker callback filter 2026-03-09 14:39:49 +00:00
Roland Tannous
56412f2362 include all candidate files when scanning a directory, not just the first 2026-03-09 13:52:45 +00:00
Roland Tannous
91dd7fc762 merge nightly, resolve conflict in use-chat-model-runtime 2026-03-09 13:19:17 +00:00
Roland Tannous
c719f1ba54 training: restore YAML fallback for trust_remote_code (no UI toggle) 2026-03-09 13:10:24 +00:00
Roland Tannous
7989cd4567 respect trust_remote_code toggle, return helpful error when required 2026-03-09 13:06:55 +00:00
Manan17
a49638c504 dataset upload 2026-03-09 05:50:18 +00:00
Shine1i
3b1663b1e9 feat(recipe-studio, datasets): improve dataset handling and update metadata logic 2026-03-09 02:47:32 +01:00
samit
662cb1c440 Adding trust_remote_code to the orchestrator and worker 2026-03-08 16:44:41 -07:00
Shine1i
a2dde15367 merge nightly 2026-03-09 00:32:33 +01:00
samit
86e94b5844 exposed trust_remote_code through the UI 2026-03-08 16:28:56 -07:00
Roland Tannous
a1778d6655 fix: replace is_dataset_multimodal with is_dataset_image/is_dataset_audio in training orchestrator 2026-03-08 19:40:00 +00:00
Manan17
80b704d7b7 Audio_VLM bug fix 2026-03-08 19:14:07 +00:00
Roland Tannous
7ee81dd7df feat: route audio inference (TTS, ASR, Whisper) through orchestrator/worker subprocess 2026-03-08 18:25:27 +00:00
Roland Tannous
1435dbaf59 merge nightly into audio branch (mock test) 2026-03-08 10:23:44 +00:00
Manan17
ef714f010e adding export support 2026-03-08 04:18:20 +00:00
Roland Tannous
a7c34b42be fix: clear stale model state on failed inference subprocess reload 2026-03-07 23:32:53 +00:00
Roland Tannous
4f766bbe25 fix: reset checkpoint metadata on failed export checkpoint reload 2026-03-07 23:29:34 +00:00
Manan17
6487f81113 check fir gated repo 2026-03-07 21:32:50 +00:00
Roland Tannous
8454e6dd2b fix: scope dataloader_num_workers=0 to Windows + transformers 5.x only 2026-03-07 17:55:59 +00:00
Roland Tannous
ef9184c731 fix: prevent training hang on Windows by adding triton-windows support 2026-03-07 17:53:36 +00:00
Roland Tannous
e25705a211 fix: propagate PYTHONPATH to child subprocesses, revert tokenizer patching 2026-03-07 11:28:24 +00:00
Roland Tannous
9330588015 fix: patch TokenizersBackend in export output after save_pretrained 2026-03-07 10:57:51 +00:00
Roland Tannous
76c78afb8f fix: patch TokenizersBackend by model name - Qwen3.5→Qwen2Tokenizer, GLM→PreTrainedTokenizer 2026-03-07 10:29:59 +00:00
Roland Tannous
d60cd2843f fix: patch Qwen3.5 broken tokenizer_class TokenizersBackend across all backends 2026-03-07 09:43:25 +00:00