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: batch MoE fine-tuning jobs with per-job expert LoRA (#15809)
2a9f230 Daniel Han committed 23 days ago
Block CUDA graphs: turn on automatically for fixed-shape, launch-bound LoRA runs (#15768)
04d3788 Daniel Han committed 23 days ago
Keep image and video models loaded beside chat models on Apple Silicon (#15693)
556b28b Long Yixing committed 23 days ago
Studio: size FLUX.1 Kontext guard, preview and compile shape from its ~1 MP render (#15797)
7df6a08 Daniel Han committed 23 days ago
FP8: skip the compiled wrapper for rowwise FP8 inference linears (#15584)
772bff9 Daniel Han committed 23 days ago
MoE heuristic dW config: 128-wide N tile for gpt-oss K = 2880 (#15505)
3a360b3 Daniel Han committed 23 days ago
SDPA decode: keep the caller's backend policy when avoiding cuDNN (#15525)
c1dc69d Daniel Han committed 23 days ago
Flex small-smem tiles: keep float32 Gemma 3 off flex on sm86 / sm89, tile forced no-grad calls (#15529)
67d3f81 Daniel Han committed 23 days ago
studiobench: seed the browser with update notifications off (#15805)
56278f5 Daniel Han committed 23 days ago
Tests: allow the zoo's padded logits row stride in the fused linear adapter check (#15803)
2d4ac1c Daniel Han committed 24 days ago
Run packed rows of head-dim-routed flex decoders through varlen SDPA (#15514)
92ecb52 Daniel Han committed 24 days ago
Cross entropy: 8192-wide chunks at 4 warps for the chunked forward on sm120 (#15744)
fa96c94 Daniel Han committed 24 days ago
Frontend tests: follow the desktop quit hiding the window (#15801)
fb02958 Daniel Han committed 24 days ago
Studio: list a Qwen-Image-2.1-Turbo base pipeline cached in its FP8 mirror On Device (#15772)
56744bf Daniel Han committed 24 days ago
Studio: give each batched fine-tuning job its own fp16 loss scale (#15741)
795da3b Daniel Han committed 24 days ago
CI: hold the unpinned version-compat lanes below torch 2.15 until torchvision 0.30 is published (#15774)
6a7c2c6 Daniel Han committed 24 days ago
Tensor parallel: batch_rebalance and group_by_length samplers give tp peers the same rows (#15752)
25fd237 Daniel Han committed 24 days ago
Studio video: read an LTX-2 checkpoint header once per file when probing for 2.3 (#15738)
f1dedb6 Daniel Han committed 24 days ago
Studio export: skip the GPU cache clear when nothing was loaded since the last one (#15737)
a121a07 Daniel Han committed 24 days ago
Installer downloads: report progress on whole 1 MiB chunks again (#15789)
64d0e77 Daniel Han committed 24 days ago
FSDP2 block CUDA graphs: free the gathered weights of flex BlockMask blocks (#15756)
85676c3 Daniel Han committed 24 days ago
Flash impl: warn about an explicit fa3 / fa4 fallback on the Trainer path too (#15750)
e21cfc0 Daniel Han committed 24 days ago
Tensor parallel x FSDP2: average trainable TP parameters left out of FSDP2 over the data parallel ranks (#15748)
f27ae8a Daniel Han committed 24 days ago
Bump install.sh / install.ps1 pins to unsloth>=2026.10.10, unsloth-zoo>=2026.10.10 (#15785)
d82698a Daniel Han committed 24 days ago
Import: check for a broken openai install on first use instead of at import unsloth (#15702)
549b140 Daniel Han committed 24 days ago
Tests: follow the thread-list guard flag and the RAG WAL keeper (#15778)
a16fb4f Daniel Han committed 24 days ago
Studio: install the released Diffusers 0.42.1 instead of the pinned main commit (#15761)
3cced21 Daniel Han committed 24 days ago