Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
Commits on main
Studio: keep the last full-attention layer unquantized in the MLX KV cache (#11083)
164f9c8 Long Yixing committed 1 month ago
Pass gradients through compressed-tensors activation quantization so W8A8 checkpoints train with LoRA (#11585)
9d15a59 Daniel Han committed 1 month ago
Count the MLX grammar engine slot in the Apple Silicon step totals (#12135)
ef63a38 Daniel Han committed 1 month ago
Studio: show the Hub error for an unreadable GGUF repo instead of routing it to Transformers (#12117)
3ab8cc6 Daniel Han committed 1 month ago
Unsloth Desktop: Create, Edit, and Delete Skills from the Skills Menu (#11800)
c0ebb4c Leo Borcherding committed 1 month ago
Studio: read more chat attachment formats, and hand the rest to the python tool (#11379)
57a8849 Long Yixing committed 1 month ago
Keep the Triton MoE grouped GEMM in compiled graphs and index weights past 2^31 elements (#12114)
b5bae41 Daniel Han committed 1 month ago
Keep ORPO / CPO rows within max_length on TRL 0.29+ (#12115)
88ac167 Daniel Han committed 1 month ago
Trim comments in the DeepSeek-V4 grouped LoRA, remote-code shims and Mistral-format loader code (#12089)
10f2575 Daniel Han committed 1 month ago
Treat a reaped grandchild as dead in the formatter timeout test (#12107)
32d57fc Daniel Han committed 1 month ago
Unsloth Studio installer (AMD/Linux): explain why a newer ROCm gets ROCm 7.2 PyTorch (#11651)
84e97f1 Leo Borcherding committed 1 month ago
Studio: cache and coalesce nvidia-smi reads in the backend (#11995)
0d2905f Daniel Han committed 1 month ago
Studio: quantize the KV cache of sliding-window MLX models such as Gemma 4 (#11082)
c4f0b0b Long Yixing committed 1 month ago
Patch the TRL trainers that moved to trl.experimental (ORPO, CPO, Online DPO, GKD, ...) (#12097)
69623ef Daniel Han committed 1 month ago
Default GRPO to TRL's dapo loss with beta 0, and cap CISPO weights at 5.0 (#12088)
a9982dc Daniel Han committed 1 month ago
Studio: share MLX VLM prompt-cache snapshot buffers and replay exact prompts (#11659)
32a35ae Long Yixing committed 1 month ago
Carry the resolved attention implementation to nested configs a remote config baked flash attention into (Nemotron 3 Nano Omni) (#12099)
f53b103 Daniel Han committed 1 month ago
Read settings.py as UTF-8 in the unknown-palette test (#12101)
23b0807 Daniel Han committed 1 month ago
Read settings.py as utf-8 in the palette filter test (#12105)
23e3fe9 Daniel Han committed 1 month ago
Studio: return an error when a model can't use the tools sent to it (#11960)
dbba692 Nilay committed 1 month ago
Upload all GGUF files in one commit so create_pr opens one pull request (#11857)
65f27c0 Nilay committed 1 month ago
Studio: let full-scope keyless callers auto-switch models, and explain the refusal elsewhere (#11180)
7be1d14 Long Yixing committed 1 month ago
Studio: keep a pinned context as a request limit instead of refusing KV cache quantization (#11084)
1759e64 Long Yixing committed 1 month ago
Studio: refuse a hand-set Metal context only past the GPU wired limit (#10804)
e8c5187 Long Yixing committed 1 month ago
Studio: regionally compile Lumina-2 and HiDream-I1, and re-decide their auto precision by measurement (#12039)
eec898c Daniel Han committed 1 month ago
Unsloth Studio / Desktop: let the GPUs picker say how much of the model each card gets (#12015)
3cbe541 Leo Borcherding committed 1 month ago
Load the repo AutoProcessor for AutoModel-only repo-code VLMs (#12037)
7c8a51f Daniel Han committed 1 month ago
Keep Llama 3.2 Vision off flash attention (vision and cross attention have no is_causal) (#12033)
7b33e85 Daniel Han committed 1 month ago
Studio: keep a chat's start date in the system prompt, note a new date on the latest user turn (#12096)
7e80a43 Daniel Han committed 1 month ago