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 RAG: delete a retired folder's or scope's chunks in batches instead of one index scan per document (#17281)
78abc68 Daniel Han committed 10 days ago
Studio chat: skip the display strip chain for history text with no tool markup (#17264)
0ee859b Daniel Han committed 10 days ago
Studio Metal: read the Apple wired headroom once per GGUF placement (#17275)
a037a9f Daniel Han committed 10 days ago
Patch sentence_transformers when it is first imported instead of importing it with unsloth (#17267)
082595c Daniel Han committed 10 days ago
unsloth studio stop: return as soon as the server exits instead of on the next half-second step (#17283)
64795e5 Daniel Han committed 10 days ago
Studio: import HfApi on first use instead of at backend start (#17273)
f40d2b1 Daniel Han committed 10 days ago
Studio: pause display-only training polls while the tab is hidden (#17294)
b1b92f7 Daniel Han committed 10 days ago
Studio dataset detection: read only the still-missing columns when filling a sparse first row (#17248)
4052d2e Daniel Han committed 10 days ago
Unsloth Studio: list Cosmos3-Nano in the Video picker and route it from the Hub (#16743)
8059310 Leo Borcherding committed 10 days ago
Studio Data Recipes: keep unstructured seed upload file IO off the event loop (#17266)
1c387c7 Daniel Han committed 10 days ago
Studio: keep marked out of startup by moving the export format names to their own module (#17250)
44e86a0 Daniel Han committed 10 days ago
Studio: memoize the nonce-free Content-Security-Policy on its inputs (#17246)
5a1bf45 Daniel Han committed 10 days ago
Studio Windows tools: stop the trusted git lookup at the first trusted candidate (#17270)
e28d146 Daniel Han committed 10 days ago
Studio: skip superseded wallpaper writes and no-op history / bookmark persists (#17298)
406c485 Daniel Han committed 10 days ago
Studio GGUF variants: reuse the request's Hub listing for the MLX fallback (#17279)
e172bd3 Daniel Han committed 10 days ago
Studio chat search: skip metadata_json when building the search index batch (#17249)
8043277 Daniel Han committed 10 days ago
Studio dataset preview: hand a JSON file that is not JSON Lines to the loader without reading it whole (#17277)
d2a69e4 Daniel Han committed 10 days ago
Studio embeddings: parse the GGUF body only when read and once, and serialize the Studio embedder payload off the event loop (#17286)
947973b Daniel Han committed 10 days ago
Studio: reduce Liquid Glass paint cost when WebKit paints on the CPU (#16960)
6acc773 oobabooga committed 10 days ago
Studio: keep the browser feature index out of startup (#17280)
e77b167 Daniel Han committed 10 days ago
Unsloth Studio: Cosmos3-Nano video continuation (extend a clip from its last or first frames) (#16741)
34905ec Leo Borcherding committed 10 days ago
Studio research: run each step's knowledge base and MCP lookups alongside its web search (#17287)
1cc11b2 Daniel Han committed 10 days ago
Studio library: cache Intl formatters for per-row dates and sizes (#17251)
d72e1dc Daniel Han committed 10 days ago
Studio debug logs: list log names instead of building a Path per file on each poll (#17302)
4d880f6 Daniel Han committed 10 days ago
Studio Images X/Y plot: shrink grid cells off the event loop (#17265)
a28ec1a Daniel Han committed 10 days ago
unsloth CLI: check a Studio pid with psutil before spawning tasklist on Windows (#17271)
89d4c20 Daniel Han committed 10 days ago
Studio: reuse Windows ROCm Get-Counter readings for display polls like amd-smi (#17306)
3b6339d Daniel Han committed 10 days ago
Studio diffusion: price a Qwen-Image-2.1 edit's reference on the group tier so 16 GiB edits stop hooking the resident transformer (L4 first edit -19%, -8.4 GiB host RAM) (#17364)
15bd320 Daniel Han committed 10 days ago
Decode CUDA graphs on by default for every FastModel model (2-8x on T4, tokens equal to eager) (#16587)
00d1056 Daniel Han committed 10 days ago