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: convert only the cached unstructured seed rows a caller reads (#15667)
913e0c2 Daniel Han committed 24 days ago
Studio: read each PDF page's words once when locating chunk regions (#15665)
a061fb1 Daniel Han committed 24 days ago
Studio: match audio.cpp CUDA runtime to bundle architecture (#15491)
1ced939 Daniel Han committed 24 days ago
Studio: tokenize a repeated chunk text once in the sentence-transformers counter (#15664)
04a9937 Daniel Han committed 24 days ago
Studio: copy SNAC codes to the host once per example in TTS preprocessing (#15660)
c56162e Daniel Han committed 24 days ago
Studio: reuse the size probe's repo listing for scoped download blob hashes (#15663)
3c8e1fd Daniel Han committed 24 days ago
Studio: copy generated SNAC codes to the host once per decode (#15666)
0b618df Daniel Han committed 24 days ago
Studio desktop: give port validation probes the health watchdog's 10 s budget (#15670)
b2e2ea8 Daniel Han committed 24 days ago
Studio: overlap model checkpoint downloads with dataset preparation (#15483)
e4251c5 Daniel Han committed 24 days ago
Model loading: skip the root index probe when the repo listing shows no index (#15662)
87c3a31 Daniel Han committed 24 days ago
Studio: clarify LoRA and training hyperparameter labels (#15479)
71ad126 Daniel Han committed 24 days ago
Studio: calmer login and setup pages with an accent sign-in button (#15731)
8cc613a Michael Han committed 24 days ago
[bugfix] Snapshot in-memory DPO policies for the default reference (#15537)
24e5736 cherry77-cloud committed 24 days ago
Unsloth Studio / Desktop: log out, show the owner's profile in Accounts, and let the owner sign in with a chosen name (#15619)
ded2b14 Leo Borcherding committed 24 days ago
UNSLOTH_PP_SIZE: install the Trainer wrapper at import, so encoders refuse instead of training as DDP (#15710)
cbc5fbb Daniel Han committed 24 days ago
fast_inference under UNSLOTH_FSDP2=1 torchrun: keep vLLM instead of refusing it (#15709)
a27be31 Daniel Han committed 24 days ago
Native FSDP2: clip gradients through Unsloth's clip, not the mixed mesh path (#15708)
d546b0c Daniel Han committed 24 days ago
Studio: answer 422 to a timestamp or page cursor sqlite cannot store (#15389)
11f6310 Vineeth Sai Varikuntla committed 24 days ago
Studio: reject tool calls from truncated local generations (#15428)
6a6b9d0 oobabooga committed 24 days ago
[Bug] AMD: Unsloth Studio Images retries the ROCm sd.cpp build on hosts whose ROCm comes only from the torch wheel (#15430)
8e42faf Leo Borcherding committed 24 days ago
Studio video: park MiniMax-H3's rotating VAEs by re-pointing at their pageable host copy when pinning is not allowed (#15715)
c432bc3 Daniel Han committed 24 days ago
Studio: only trust file search and dictation servers Studio started (#15600)
c969323 Nilay committed 24 days ago
Studio: keep Qwen-Image-2.1 prompt-length dims dynamic under group offload hooks (#15652)
3701d52 Daniel Han committed 24 days ago
Studio: keep ticked PDF form check boxes in knowledge bases (#15572)
db9d4f0 Nilay committed 24 days ago
Studio: count ASCII in context estimates without a per-character loop (#15659)
86f509a Daniel Han committed 24 days ago
Studio: parse each seed glob and matched file once when applying loader skip rules (#15668)
6f19166 Daniel Han committed 24 days ago
Studio: load local JSON datasets whose later rows add keys (#15555)
8222149 Daniel Han committed 24 days ago