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
Keep padding-free off for attention that mixes packed documents (DeepSeek V4, Zaya, RecurrentGemma) (#16119)
a4242ba Daniel Han committed 21 days ago
Fix Images and Video header controls clipping in narrow windows (#15798)
d697fef Lee Jackson committed 22 days ago
Studio: run a second tool call glued to the first in DeepSeek envelopes or one Hermes tag (#16026)
1100377 Daniel Han committed 22 days ago
fix(cli): make `unsloth doctor` MLX-aware on Apple Silicon (#14457)
1cd3674 charlie committed 22 days ago
Tests: read the decision prefetch test's frontend source as UTF-8 (#16123)
999aa3f Daniel Han committed 22 days ago
Studio: decode Gemma quoted values in a tool call without its wrapper (#16009)
36ec318 Daniel Han committed 22 days ago
Studio: heal tool-call arguments sampled as near-JSON instead of passing them whole as text (#16005)
4e95539 Daniel Han committed 22 days ago
Studio: MOSS-Image, SDXL-Turbo and MiniMax-H3 default steps / guidance follow the reference workflows (#16097)
c569b23 Daniel Han committed 22 days ago
Scan baseline: re-review torch 2.15.0, botocore 1.43.122 and unsloth-zoo 2026.10.10 (#16114)
7977cc1 Daniel Han committed 22 days ago
Studio: mark the root when the chat wallpaper is ready instead of html:has() (#16117)
1952d23 Daniel Han committed 22 days ago
Studio: follow the chat background from #15973 through three frontend guards (#16116)
cedf0ce Daniel Han committed 22 days ago
Studio: run a tool call whose name differs only in case or hyphens (#16007)
99c79ae Daniel Han committed 22 days ago
Studio diffusion: Ada FP8 unit-scale dispatcher installs only where it engages, chunked eager scaling (#14665)
487ee60 Daniel Han committed 22 days ago
Studio: prepare decision datasets faster, in less memory, with row progress (#16106)
3f8e4d3 Daniel Han committed 22 days ago
Studio: retire service workers other apps left on the Studio origin (#16107)
4b94a56 Daniel Han committed 22 days ago
Studio: do not hand agents a tool call cut off at the token limit (#16035)
9d195d4 Nilay committed 22 days ago
Studio: keep commands and modules from a second sandboxed pip install (#16018)
f28d1c8 Nilay committed 22 days ago
Studio: place automatic native audio loads on one GPU instead of refusing a pooled pick (#16022)
05aa797 oobabooga committed 22 days ago
Studio: scheduled tasks use Memory and project instructions (#16038)
77ecead Nilay committed 22 days ago
AV lint: record the sd.cpp missing-DLL probe test's python.exe copy (#16108)
2199d9f Daniel Han committed 22 days ago
Studio: rewind puts back files edit_file changed inside dot folders (#16017)
9d203e6 Nilay committed 22 days ago
Studio diffusion: download a partially cached snapshot before sizing it for small hosts (#16080)
a00356e Daniel Han committed 22 days ago
Studio: recover a closed tool call whose single code argument has unescaped quotes (#16006)
8d80884 Daniel Han committed 22 days ago
Studio: stop offering the VibeVoice q4-ios GGUF that audio.cpp cannot load (#16048)
9c3a78c oobabooga committed 22 days ago
Studio: compile the SDXL U-Net per block so the 4th image no longer waits for the whole compile (#16085)
8a02150 Daniel Han committed 22 days ago
Studio: add OpenCode Zen and OpenCode Go to Connections (#16010)
1f3eb9d Nilay committed 22 days ago
Studio: keep a bare-JSON tool call out of the final answer after the tool loop (#16003)
0285476 Daniel Han committed 22 days ago
Studio: add customizable chat background images and effects (#15973)
694cac1 Wasim Yousef Said committed 22 days ago