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
MoE autotune cache: version disk entries and refit loaded configs (#14617)
465263e Daniel Han committed 1 month ago
Studio: OpenAI-compatible fine-tuning jobs API (/v1/files, /v1/fine_tuning/jobs) (#14496)
6bbe268 Daniel Han committed 1 month ago
Studio: resize the sliding pill with transforms, so it keeps pace with the move (#14361)
28f11ed Etherl committed 1 month ago
Fused decode RoPE: drop the per-layer host drain on split models (#14611)
8b4bb4c Daniel Han committed 1 month ago
Fast LayerNorm backward: no aliased gradient buffers when only one of weight / bias trains (#14609)
72da63a Daniel Han committed 1 month ago
Studio Run settings: restore saved settings on a fresh editor, show save state, group Advanced rows (#14589)
7fd4988 Michael Han committed 1 month ago
Studio: more accurate connection checks and clearer errors (#14594)
c12ba61 Michael Han committed 1 month ago
FLUX.3 upsampler: name older NVIDIA Vulkan devices as NVIDIA, probe an empty answer once per window (#14604)
d4c220b Daniel Han committed 1 month ago
Studio: shortcut fixes for non-US layouts, Esc and Cmd-K, plus clearer settings (#14597)
8bcbf83 Michael Han committed 1 month ago
Studio: token activity shows a year of quartile-shaded days, with accurate counts behind it (#14600)
665e7a6 Michael Han committed 1 month ago
MoE grouped GEMM: no lock on cached autotuner launches; read the SM count once per card (#14605)
b2ba86a Daniel Han committed 1 month ago
unsloth start: fix Studio discovery, proxy, Hermes context, Claude hooks and Codex login (#14591)
cb2590f Michael Han committed 1 month ago
QAT: qat_scheme = "mxfp8", and let MX QAT linears take float32 activations under autocast (#14607)
890ea71 Daniel Han committed 1 month ago
Studio: Projects recency by chat activity, less polling behind projects and sources (#14586)
265aa0d Michael Han committed 1 month ago
Studio: chat fork fixes and clearer fork affordances (#14587)
a38820e Michael Han committed 1 month ago
Studio: show relative and blob Hugging Face images in Hub model cards (#14602)
114ee5f Michael Han committed 1 month ago
Honor requested languages for Studio Qwen3-ASR transcription (#14548)
f15be5d Tom Pipe committed 1 month ago
Fast decode: rotate Q and K in one launch, and fuse Qwen3 q/k norm with RoPE (#14599)
c5d885f Daniel Han committed 1 month ago
Keep Unsloth Studio data after docker rm on CPU-only Docker hosts (#14540)
9488507 Nilay committed 1 month ago
RoPE cache: build cos/sin outside inference_mode so GRPO keeps the compiled decoder block (#14584)
77e742c Daniel Han committed 1 month ago
Studio diffusion: latch failed fused fp16 RoPE and AdaLN launches for the load (#14583)
db6e309 Daniel Han committed 1 month ago
Studio video: export GIFs with one clip-wide palette (about 3x faster) (#14581)
2749eee Daniel Han committed 1 month ago
Studio diffusion: turn on the measured-safe optimizations by default (fused FLUX RoPE, CFG batching, Blackwell native attention, BF16 small DiTs) (#14566)
5a955ef Daniel Han committed 1 month ago
Studio: keep CFG-parallel on the default speed tier and run Z-Image's guidance branches on two cards (#14564)
5433863 Daniel Han committed 1 month ago
Studio diffusion: keep fp8 single-file weights stored in fp8 where widening would offload or crowd host RAM (#14549)
4ff4c66 Daniel Han committed 1 month ago
MoE fallback kernel configs: fit the device shared memory for float32 operands (#14576)
0391814 Daniel Han committed 1 month ago
MoE autotune cache: compare device capability as a tuple so saved configs are reused (#14578)
3c8e64e Daniel Han committed 1 month ago
Skip the compiled MoE autotune test on the recorded real CUDA, not the spoofable probe (#14577)
463e3f3 Daniel Han committed 1 month ago