Skip to content

Pull requests: unslothai/unsloth-zoo

New pull request New
110 Open 677 Closed
110 Open 677 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Forward declared tools to the DiffusionGemma visual server
#864 opened Jul 3, 2026 by oobabooga Member Loading…
5
Add a grouped bnb 4-bit training forward for gpt-oss experts
#862 opened Jul 3, 2026 by danielhanchen Member Loading…
2
Add a banded sliding-window SDPA fast path for Gemma-4 style layers
#861 opened Jul 3, 2026 by danielhanchen Member Loading…
9
Route gemma-4 sliding-window layers through FlashAttention-2
#860 opened Jul 3, 2026 by danielhanchen Member Loading…
4
Keep DeepSeek-V4 hyper-connection mixers eager to stop backward inf
#859 opened Jul 3, 2026 by danielhanchen Member Loading…
1
Optional gate gradient identity for the grouped MoE combine
#858 opened Jul 3, 2026 by danielhanchen Member Loading…
6
Cap the fused cross entropy chunk target so chunking stays active on large GPUs
#857 opened Jul 3, 2026 by danielhanchen Member Loading…
3
Load prequantized bnb 4-bit MoE expert checkpoints under transformers v5
#856 opened Jul 3, 2026 by danielhanchen Member Loading…
12
Fix MLX notebook generate input/output compatibility
#855 opened Jul 3, 2026 by Lyxot Contributor Loading…
6
feat(mlx): support distributed inference loading
#854 opened Jul 3, 2026 by Lyxot Contributor Loading…
11
dataset_utils: add mask_out_tokens to train_on_responses_only (fixes unslothai/unsloth#6695)
#852 opened Jul 3, 2026 by pjordanandrsn Loading…
7
mlx: fixes for bnb dequant, report_to, early stopping, and NEFTune
#851 opened Jul 3, 2026 by oobabooga Member Loading…
9
feat(mlx): load GPTQ/AWQ pre-quantized checkpoints on Apple Silicon
#848 opened Jul 2, 2026 by BardiaKoopah Contributor Loading…
11
fix(gguf): preserve Qwen3.5/3.6 MTP config so nextn tensors convert
#847 opened Jul 2, 2026 by LeoBorcherding Contributor Loading…
2 tasks done
3
fix(mlx): normalize legacy tokenizer sidecars for VLM loads
#846 opened Jul 2, 2026 by Lyxot Contributor Loading…
8
MLX: clear error when mlx-lm/mlx-vlm is too old for a QK-norm arch (gemma4/qwen3_5)
#845 opened Jul 2, 2026 by danielhanchen Member Loading…
28
fix(mlx): preserve inference norm dtypes
#844 opened Jul 1, 2026 by Lyxot Contributor Loading…
8
feat(mlx): add DDP training support
#841 opened Jun 29, 2026 by Lyxot Contributor Loading…
23
Optional recompute-in-backward for the v5 grouped MoE path
#838 opened Jun 28, 2026 by danielhanchen Member Loading…
16
Grouped-GEMM MoE forward for transformers<5 ModuleList experts
#837 opened Jun 28, 2026 by danielhanchen Member Loading…
40
dataset_utils: add mask_thinking_tokens to mask the </think> token
#833 opened Jun 26, 2026 by Sushankthatipally Loading…
10
Feat(mlx): GRPO support for MLX
#832 opened Jun 26, 2026 by BardiaKoopah Contributor Loading…
35
Feat(mlx): Add ORPO (loss_type='orpo') for text models to MLXTrainer
#830 opened Jun 24, 2026 by BardiaKoopah Contributor Loading…
8
Add shared Xet to HTTP stall fallback (hf_xet_fallback, hf_cache_state)
#829 opened Jun 24, 2026 by danielhanchen Member Loading…
431
Support DoRA (use_dora=True) in the safetensors LoRA merge
#828 opened Jun 24, 2026 by danielhanchen Member Loading…
1
ProTip! Updated in the last three days: updated:>2026-06-30.