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
Check a dataset for bad rows and over-length examples before training (#16818)
d06ee9d Nilay committed 14 days ago
Studio: make links in the collapsed update notes clickable (#16829)
2a3eae7 Nilay committed 14 days ago
Studio: ask to update Codex when it is too old for unsloth start codex (#16843)
e006948 Nilay committed 14 days ago
Studio: test that the picker keeps a Hub search tab when a local model lands (#14180)
643b3b4 Mrityunjay Kumar committed 14 days ago
Studio: ask before Python runs blocked shell commands via __import__ (#16840)
3463dae Nilay committed 14 days ago
Studio chat UI test: wait for the monitor size limits before the resize, and report them (#16885)
dff470f Daniel Han committed 14 days ago
Studio GGUF: fuse Q8_0 dequant and cast into one kernel (LTX-2.5 Q8_0 denoise -12%, bit-identical) (#16884)
a0197ca Daniel Han committed 14 days ago
Load path: read an adapter's base config through the probe memo, skip the remote mapper for unknown LoRA bases in Studio (#16793)
90e1e73 Daniel Han committed 14 days ago
Studio: decode the Transformers chat stream incrementally instead of re-decoding the current line per token (#16800)
91c2e80 Daniel Han committed 14 days ago
Studio: make the claude mcp add command work in every project (#16839)
1f5b335 Nilay committed 14 days ago
Unsloth Studio: keep chat search fast on long histories by not shipping tool images and widget seeds to the index (#16848)
5cc8572 Leo Borcherding committed 14 days ago
install.ps1: delete the committed venv rollback in a background process (#16796)
27d9b87 Daniel Han committed 14 days ago
RL class source: parse the code above a class with ast.parse, not a dynamic compile() (#16896)
7e4586c Daniel Han committed 14 days ago
Studio RAG: batch llama-server token counts and encode the auto-inject query once (#16812)
070e482 Daniel Han committed 14 days ago
TextPreprocessor.add_structure_tokens: leave fenced code blocks alone (#16697)
0e241f2 Vineeth Sai Varikuntla committed 14 days ago
Studio: index FK child columns, partial index for the run lease sweep, keep the inventory clean on Deep Research start (#16795)
252af16 Daniel Han committed 14 days ago
RL patch pass: skip the whole-module ast.parse when reading trainer class source on Python < 3.13 (#16806)
306aad9 Daniel Han committed 14 days ago
Legacy CLI: accept --dtype None on every backend, not only MLX (#15775)
69854d6 Vineeth Sai Varikuntla committed 14 days ago
Studio: let unsloth start dsh install DeepSeek Harness on Node 22 (#16844)
55662d5 Nilay committed 14 days ago
Studio: run Chroma's ComfyUI fp8 files in fp8, as ComfyUI does (#16872)
6f0692a oobabooga committed 14 days ago
get_chat_template: apply mapping when the role key is not renamed (#16425)
e1034a1 Vineeth Sai Varikuntla committed 14 days ago
Studio: update the NPU runtime, harden its pin updates, and polish the NPU UI (#16834)
1928a2c oobabooga committed 14 days ago
Studio: resume paused and interrupted model downloads on Windows (#16828)
628b0c9 Nilay committed 14 days ago
Studio audio analyze: screen the true peak in float32 and convolve only the candidate outputs (#16794)
418e88d Daniel Han committed 14 days ago
Studio: read unread scheduled replies aloud when opened (#16847)
b2f8daf oobabooga committed 14 days ago
Studio: keep fine-tuning jobs in order when a batch starts ahead of them (#16832)
b0072c7 Nilay committed 14 days ago