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

222 commits

Author SHA1 Message Date
Roland Tannous
d69431fa57 Scale dataset num_proc dynamically to cpu_count//3 instead of hardcap 8 2026-02-18 08:38:53 +00:00
Manan17
76cd1dc24c fixing the hangup of training after multiple back to back training processes 2026-02-18 08:18:13 +00:00
Manan17
c37bf686a6 Dividing the total cpu_count // 3 2026-02-18 07:59:57 +00:00
Roland Tannous
14edb08cf5 Merge pull request #148 from unslothai/fix/linear 
linear fix
 2026-02-18 11:10:16 +04:00
Manan17
58116e7e7a fix the linear path on backend 2026-02-18 07:08:32 +00:00
Roland Tannous
d2f7eaf085 Merge pull request #145 from unslothai/fix/check-format-sample 
fix: stream HF datasets in check-format endpoint to avoid full d…
 2026-02-18 10:49:17 +04:00
Roland Tannous
713d4a6ddb Merge pull request #146 from unslothai/feature/fix-cuda-fork-deadlock 
fix: cap dataset.map() num_proc to 8 to prevent CUDA fork deadlocks
 2026-02-18 10:47:46 +04:00
imagineer99
064cd56a21 feat: disable navbar navigation while training is active 
Disable Export and Chat nav items (desktop + mobile) when
isTrainingRunning is true, keeping only Studio clickable.
 2026-02-18 01:30:13 +00:00
Manan17
42ee6178ae linear fix 2026-02-18 00:11:27 +00:00
Roland Tannous
3d0d1c7020 fix: cap dataset.map() num_proc to 8 to prevent CUDA fork deadlocks 2026-02-17 23:12:45 +00:00
Roland Tannous
5864dece26 fix: fix: stream HF datasets in check-format endpoint to avoid full downloads; add info logging to model config endpoints 2026-02-17 22:53:29 +00:00
Roland Tannous
63196042ea Merge pull request #144 from unslothai/fix/remove-hardcoded-port-in-dataset-preview 
fix: remove hardcoded port from dataset preview error message
 2026-02-18 02:44:17 +04:00
imagineer99
e3fb46a4c6 fix: remove hardcoded port from dataset preview error message 2026-02-17 22:37:52 +00:00
Shine1i
dc0cec772d feat: enhance training stop and reset flow with detailed checks 2026-02-17 23:32:22 +01:00
Wasim Yousef Said
765e1cfee2 Merge pull request #143 from unslothai/feature/local-models 
feat: add schemas for local model discovery and listing
 2026-02-17 13:10:04 -08:00
Shine1i
972cde7971 feat: add schemas for local model discovery and listing 2026-02-17 21:53:42 +01:00
Roland Tannous
f88f0bc047 Merge pull request #142 from unslothai/feature/wsl-gguf-sudo-fix 
fix: skip sudo check on WSL during GGUF export to prevent password pr…
 2026-02-18 00:46:15 +04:00
Wasim Yousef Said
43fac18d40 Merge pull request #141 from unslothai/feature/uxui-heuristics 
style: improve layout consistency and responsiveness across components
 2026-02-17 12:34:37 -08:00
Shine1i
68f0321404 style: improve layout consistency and responsiveness across components 
- Adjusted padding, spacing, and grid configurations for better alignment and scaling across screen sizes.
- Enhanced mobile responsiveness by updating flex and grid layouts, ensuring optimal display on smaller devices.
- Tuned container dimensions and card styling to maintain design consistency.
 2026-02-17 21:27:02 +01:00
Wasim Yousef Said
d9869dece1 Merge pull request #140 from unslothai/feature/uxui-heuristics 
ux: improve training-to-chat flow, param defaults UX, and guided onboarding/export polish
 2026-02-17 12:10:27 -08:00
Shine1i
9c353f7bc2 refactor: wrap splash screen content in a card for improved layout and consistency 2026-02-17 21:08:45 +01:00
Shine1i
08437def80 feat: add session-based storage for chat training comparison handoff 
- Implemented a utility to manage `training-compare-handoff` data in `sessionStorage` with strict validation and expiration logic.
- Added methods to set, retrieve, and clear handoff data for improved chat training flow.
 2026-02-17 20:43:08 +01:00
Shine1i
a0235025af fix chat compare handoff: auto-load trained lora, stop refresh loop, add debug logs 2026-02-17 20:42:43 +01:00
Roland Tannous
c9fdce63e7 fix: skip sudo check on WSL during GGUF export to prevent password prompt hang 2026-02-17 19:30:02 +00:00
Shine1i
ff458a36e9 feat: enhance training flow with new runtime hints, adjustable steps/epochs 
- Added halfway/completed training hints with actionable links.
- Introduced sliders for adjusting max steps and epochs dynamically.
- Refined tooltip explanations for configuration parameters.
- Enabled custom overlay styling for `AlertDialogContent`.
 2026-02-17 20:08:02 +01:00
Roland Tannous
39b072d2ee move requirements/ to studio/backend/ and update paths in setup.sh 2026-02-17 19:02:25 +00:00
Roland Tannous
c13e1594de Merge pull request #138 from unslothai/fix/reset-max-steps-epoch-defaults 
Override Model Defaults for num_epochs and max_steps
 2026-02-17 22:51:40 +04:00
Roland Tannous
299bc65e36 chore: override model defaults to use max_steps=30, save_steps=30, num_epochs=0 for testing 2026-02-17 18:46:23 +00:00
Wasim Yousef Said
7874fa9066 Merge pull request #137 from unslothai/feature/chart-fixes 
rafactor: fix training charts: sticky full-window follow latest + grad line render + stopped metric fallback
 2026-02-17 10:41:01 -08:00
Shine1i
4e2b567c29 refactor: simplify chart view logic by removing pan controls and enhancing window size handling 2026-02-17 19:32:32 +01:00
Shine1i
2d90210a69 feat: add reusable chart components for training metrics visualization 
- Introduced `EvalLossChartCard`, `GradNormChartCard`, `LearningRateChartCard`, and `TrainingLossChartCard` components.
- Implemented shared chart settings via `SharedChartSettings` to manage scale, outliers, and view configuration.
- Added utilities for metrics formatting, step tick generation, data compression, and smoothing (`utils.ts`).
- Created types and structures for chart data handling (`types.ts`).
 2026-02-17 19:10:24 +01:00
Roland Tannous
b8171a86ac Merge branch 'nightly' into integrate/exports-page 2026-02-17 22:09:36 +04:00
Roland Tannous
e818d97f24 Merge pull request #136 from unslothai/setup/update-setup-sh-dependencies 
setup.sh: Replace inline pip installs with pinned requirements files
 2026-02-17 22:04:03 +04:00
Shine1i
0be3e6f525 feat: integrate gradient norm tracking in training runtime and metrics 
- Enhanced chart logic to filter and visualize finite gradient norm values.
 2026-02-17 18:26:59 +01:00
Shine1i
7203766fac refactor: streamline vision model detection and improve state persistence logic 
- Removed redundant vision-check controllers.
- Added `NON_PERSISTED_STATE_KEYS` to manage persisted training state.
- Introduced `partializePersistedState` for cleaner state filtering.
 2026-02-17 18:01:45 +01:00
Shine1i
3badf6649c feat: add default model configuration mapping and auto-apply logic 
- Implemented backend model configuration mapping to training state.
- Added auto-apply logic for default configurations when models are selected.
- Introduced utilities for type conversion and validation within training configuration.
 2026-02-17 17:58:36 +01:00
Roland Tannous
e803e13d3e add full dependency chain for unsloth + unsloth-extras 2026-02-17 14:21:18 +00:00
Roland Tannous
8a4a1554b0 fix: vite build fail - suppress unused estimatedSize prop in export dialog 2026-02-17 07:21:23 +00:00
Roland Tannous
bbd7d6d122 Merge pull request #129 from unslothai/fix/adding-meta-data-for-checkpointing-api 
Adding metadata for checkpoints
 2026-02-17 11:17:24 +04:00
imagineer99
e1fbccfc57 fix: subset param name 2026-02-17 05:53:21 +00:00
Manan17
6f9dd90d56 Integration of the api with the EXPORT page with UI changes 2026-02-17 00:34:48 +00:00
Manan17
c7b7ecab4f Adding metadata for checkpoints 2026-02-16 23:46:17 +00:00
Wasim Yousef Said
7b8220598e Merge pull request #124 from unslothai/feature/bug-fixes 
feat: support disabling top-k sampling with -1 and standardize normalization
 2026-02-16 13:21:12 -08:00
Roland Tannous
ff0aec180a Merge branch 'nightly' into feature/eval-split-auto-detection 2026-02-17 01:11:30 +04:00
Shine1i
0db7da96cc feat: support disabling top-k sampling with -1 and standardize normalization logic 
- Updated top-k parameter range to accept -1 in models and frontend.
- Added utility to normalize top-k for backend compatibility.
 2026-02-16 21:33:24 +01:00
Roland Tannous
fa0ca59215 feat: auto-detect model+dataset compatibility to select VLM vs LLM training path 2026-02-16 19:18:49 +00:00
Roland Tannous
09f3b6bce5 feat(frontend): auto-detect vision models via backend, separate search filter from model classification 2026-02-16 18:24:44 +00:00
Roland Tannous
f4f2f50364 fix: show Back to configuration breadcrumb when training completes 2026-02-16 17:34:52 +00:00
Roland Tannous
b32ad350c5 feat: apply default chat template for base models without tokenizer chat_template 2026-02-16 15:56:06 +00:00
Roland Tannous
5df3a0b250 feat: add eval_enabled flag and format-first-then-split for eval dataset 2026-02-16 14:13:55 +00:00