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
Kaggle CI: use both accounts, weighted by their own weekly quota (#10121)
dddf214 Daniel Han committed 2 months ago
Kaggle T4: Latest_compile and Vision_FLA_compile legs, plus gpt-oss completions and GGUF (#9700)
40f452e Daniel Han committed 2 months ago
Studio: keep a Downloads entry when the list is empty (#9849)
1b596b1 Souravrajvi0 committed 2 months ago
Report a trainer kwarg that neither the trainer nor the config takes (#9944)
5d3acce Vineeth Sai Varikuntla committed 2 months ago
Studio: reuse the MLX prompt cache on hybrid recurrent models (#10062)
68348bf Long Yixing committed 2 months ago
Apply the context length and quant settings when attaching to a running server (#10093)
9aeefd5 Nilay committed 2 months ago
Studio: report the GPUs the OS sees, not just the ones PyTorch opened (#9858)
c303050 Daniel Han committed 2 months ago
Studio: send tool results to the model in the right order (#10091)
6153a6f Nilay committed 2 months ago
Linux: fix ROCm detection on split Debian stacks (#8886)
3703a66 Natani L. Mayday committed 2 months ago
fix(amd): gate training GPU selection on the installed torch build's arch coverage (#8791)
ad73e50 Leo Borcherding committed 2 months ago
Studio: tell the browser about the bootstrap shutdown deadline (#10104)
166a4a9 Daniel Han committed 2 months ago
Studio: report a context size for models loaded through MLX (#10098)
dfd0d2e Nilay committed 2 months ago
Give the unsloth_zoo pytest step a per-test timeout (#10108)
893a1fb Daniel Han committed 2 months ago
fix(studio): show the confirmation card the argument text the tool is run with (#10023)
d12aedd Long Yixing committed 2 months ago
Stop the torchao ROCm test from depending on the host's GPU (#10099)
9954357 Daniel Han committed 2 months ago
Run the llama.cpp keep-install back-compat tests on Windows (#10106)
d983613 Daniel Han committed 2 months ago
Scope the llama.cpp update test's Popen patch to the installer (#10105)
454a5ab Daniel Han committed 2 months ago
Use the public fastmcp call_tool so the MCP guard test survives fastmcp 4 (#10110)
6a8b387 Daniel Han committed 2 months ago
Add GGUF fit-target control and wire to llama-server --fit-target (#4882)
bb69768 Mithilesh Kumar committed 2 months ago
Wrap Accelerator.distributed_type patch in property() (#10087)
e05a62e mihiii1337 committed 2 months ago
Fix Accelerator.distributed_type bound-method bug on single-GPU patch (#10072)
e23deac Mayuri committed 2 months ago
Move the studiobench attribution config in with the harness (#10100)
1d2679f Daniel Han committed 2 months ago
Studio: split parallel tool-call arguments at JSON object boundaries (#10059)
fa6f06e Daniel Han committed 2 months ago
Stop two installer tests answering about the runner instead of the code (#10079)
bd09cc3 Daniel Han committed 2 months ago
Stop charging a chat the whole cache because Max Tokens says "Max" (#10070)
6416393 Daniel Han committed 2 months ago
Patch Accelerator.distributed_type as a property, not a bare function (#10028)
bb83bbb Vineeth Sai Varikuntla committed 2 months ago