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: let unsloth start agents send images to a vision model (#17540)
de45d3f Nilay committed 8 days ago
Studio: keep angle-bracket text in user turns on the safetensors fallback formatter (#17562)
37282c3 oobabooga committed 8 days ago
Studio: ask before Python deletes files through a passed function (#17541)
2b62ba1 Nilay committed 8 days ago
Studio: stop tool images breaking chats on text-only connection models (#17561)
ded777a Nilay committed 8 days ago
Studio: show an unloading toast when the image or video model is ejected (#17560)
95dd0cf Nilay committed 8 days ago
Studio: answer on Mistral reasoning models instead of an internal error (#17542)
7cda598 Nilay committed 8 days ago
Studio: keep the audio Eject during a running task so it can say to stop first (#17564)
ee1e9ec Nilay committed 8 days ago
Studio: show the context DiffusionGemma is really running with (#17566)
bcd9201 Nilay committed 8 days ago
Bump the npm-oxc-validator group across 1 directory with 3 updates (#16771)
758caa4 dependabot[bot] committed 8 days ago
Studio: web fetch reads forum threads and docs that keep their text in noscript (#17545)
2c4610c Nilay committed 8 days ago
Bump the cargo-tauri group across 1 directory with 7 updates (#16772)
f3899c4 dependabot[bot] committed 8 days ago
Dependabot: bump every sub-path of codeql-action and cache in one PR (#17587)
ecf4aaf Daniel Han committed 8 days ago
Studio: tooltips switch to the hovered icon immediately (#17574)
c6712db Michael Han committed 8 days ago
[Bug] Unsloth Studio: chat-only installs run llama.cpp on the CPU because the CUDA bundles ship no cudart/cublas (#17463)
e91e7be Leo Borcherding committed 8 days ago
Studio: Steer now stops the response for a parked attachment send (#17583)
2f0da1e Etherl committed 8 days ago
studiobench: step back over a gap the completeness traversal skipped (#17581)
13ef434 Daniel Han committed 8 days ago
Keep vendored Studio files byte-identical on Windows checkouts (#17579)
381fbff Daniel Han committed 8 days ago
Studio: fix MiniMax-H3-Image downloads, explicit FP8/INT8 and 16 MP sizes (#17563)
20686a0 oobabooga committed 8 days ago
Notebooks CI: refresh the Colab GPU oracle snapshot for cuDNN 9.28.1.4 (#17578)
af69d1f Daniel Han committed 8 days ago
Studio: keep the on-demand kernels install importable next to transformers (#17576)
9c4e41c Daniel Han committed 8 days ago
studiobench: overlap the completeness traversal stops on a windowed thread (#17577)
b5e447c Daniel Han committed 8 days ago
Studio: let the background diffusers import finish before MiniMax-H3-Image imports diffusers (#17569)
b1a466f oobabooga committed 8 days ago
studiobench: residual stabilizer padding settles the windowed pin check (#17571)
ea64e56 Daniel Han committed 8 days ago
Studio: mount only the messages near the reader in long chats (#17504)
d6841bb Wasim Yousef Said committed 8 days ago
Tests: skip the installer's transient filtered copies in the sentence-transformers pin scan (#17547)
2d3deb5 Daniel Han committed 8 days ago
Studio H3-Image: prompt cache, pinned streaming and the H3 attention levers (#17495)
5d5c138 Daniel Han committed 9 days ago