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: use the macOS system font (SF Pro) as the default UI font on Mac (#16402)
76033a7 Michael Han committed 19 days ago
Tests: run the RoPE 64-bit indexing metadata check on CPU-only runners (#16226)
e64e494 Daniel Han committed 19 days ago
Add unsloth.precision: one registry for low-precision recipes, refusals and high-precision layers (#16387)
0bfae01 Daniel Han committed 19 days ago
Tests: pin bf16 support and the HF cache location in the GGUF preflight fixtures (#15967)
921a6ce Daniel Han committed 19 days ago
Studio: make Show in folder work on Linux, and report when it can't (#16198)
f077473 Debojyoti Banerjee committed 19 days ago
Studio: search earlier turns for the user's question, not the tool loop's own notices (#16204)
f4ee098 Debojyoti Banerjee committed 19 days ago
Tests: GQA routing oracle expects expanded K/V on torch without enable_gqa (#16225)
ff23cfc Daniel Han committed 19 days ago
GRPO PrefixGrouper: check the host-syncs layout test against the reference (#16244)
9f0a000 Daniel Han committed 19 days ago
Stub the device context in the packed fp32 routing test (#16229)
0e150c5 Daniel Han committed 19 days ago
Studio tests: compare the hub source restore against the inherited source (#16236)
3df6e15 Daniel Han committed 19 days ago
Studio: decode with CUDA graphs by default in the transformers chat backend (#15993)
9e31326 Daniel Han committed 19 days ago
FSDP2 float16 with CPU offload: unscale CPU DTensor grads as local shards (#16398)
a098d99 Daniel Han committed 19 days ago
Studio tests: data settings driver expects the named single-chat delete for one archived result (#16401)
5eb2c26 Daniel Han committed 19 days ago
Studio: respawn a llama-server that ran out of GPU memory mid-session with more VRAM headroom (#16388)
7418155 Daniel Han committed 19 days ago
CLI: let train load models that need trust_remote_code (#16332)
3435afe Yakshith committed 19 days ago
Studio: exit the inference worker instead of waiting to be killed (#16338)
5362460 Yakshith committed 19 days ago
Studio: answer a collapsed On Device row from disk, not the Hub (#16271)
6c95897 Yannick Loth committed 19 days ago
Studio: run and hide MiniMax-M2 tool calls that llama-server leaves in content (#16103)
bc66d79 Daniel Han committed 19 days ago
Studio: keep Trick or Treat as a year-round theme, remove the Halloween extras (#16392)
8451d54 Michael Han committed 19 days ago
Studio tests: pin the delete chat placeholder, not its wording (#16397)
3a7798b Daniel Han committed 19 days ago
Decode CUDA graphs: opt-in whole-step capture for batch 1 (UNSLOTH_DECODE_CUDA_GRAPHS_FULL=1) (#15976)
eed6da8 Daniel Han committed 19 days ago
Gemma 3: generate on the non-flex attention path instead of flex_attention's fallback (#16041)
3740f83 Daniel Han committed 19 days ago
Flex attention on torch < 2.14: fitting tiles for short decode calls so a batch-size change does not leave the process on eager attention (#16073)
d5d0200 Daniel Han committed 19 days ago
Studio: serve the Decision API's pinned GGUFs from their ModelScope commits (#16285)
4348ac8 Debojyoti Banerjee committed 19 days ago
CLI: inference and chat take images for vision models (#16347)
ac8e86e Debojyoti Banerjee committed 19 days ago