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 diffusion: keep CFG batching off for a resident denoiser with no device timer (CPU, MPS) (#14652)
cc10707 Daniel Han committed 1 month ago
CFG-parallel: treat a changed Python number, storage offset or autocast as a cold call (#14614)
a07c510 Daniel Han committed 1 month ago
Unsloth Studio (Vulkan): fill the discrete GPU before a shared-memory iGPU in a mixed pin (#12650)
3f11b2e Leo Borcherding committed 1 month ago
feat(studio): MLX PLE and expert offload in Unsloth Studio (#14326)
9880bc0 Long Yixing committed 1 month ago
NVFP4: keep fast decode callers in the activation dtype under autocast (#14750)
b64900b Daniel Han committed 1 month ago
Studio: upgrade a Desktop Chat only install to the full install from Settings (#14593)
625cced oobabooga committed 1 month ago
FP8 dequant: multiply per-tensor and rowwise scales in fp32 for fp16 output (#14677)
22f5633 Daniel Han committed 1 month ago
tests: restore transformers' Fp8Dequantize after the fp8 save policy tests (#14741)
0811e55 Daniel Han committed 1 month ago
Studio: let the owner turn off the Decision API while a model loads (#14557)
c5a9fe0 Nilay committed 1 month ago
Studio: transcribe audio through a saved connection on the Audio page (#14289)
2cfa64f Nilay committed 1 month ago
Studio: warn when a GGUF is tight for Apple unified memory (#14598)
3569dfa Michael Han committed 1 month ago
MoE grouped GEMM: keep the legacy dW kernel when an explicit tile overflows per-expert shared memory (#14723)
e66f408 Daniel Han committed 1 month ago
Studio: lower the VRAM budget floor to 10% for multi-GPU spreading (#13427)
80d23aa Shooter3k committed 1 month ago
Studio settings: fix API, Connections, Remote access and LAN issues (#14596)
a29487c Michael Han committed 1 month ago
Studio: lighter system polling and clearer unified-memory monitor (#14592)
870f377 Michael Han committed 1 month ago
Attention routing tests: follow torch's own GQA kernel choice (fixes two failures on torch 2.14) (#14672)
0c001fc Daniel Han committed 1 month ago
FP8: keep every scoped converter and clean up every dequantizing load before save (#14678)
1b76d5f Daniel Han committed 1 month ago
Studio: extract ebooks, Office files and email in Data Recipes document seeds (#14499)
fee26fe Daniel Han committed 1 month ago
Studio: stop the SDPA probe reporting fused attention on MPS (#14488)
caf7fe7 Long Yixing committed 1 month ago
FastSentenceTransformer: non-reentrant gradient checkpointing under DDP for encoder models (#13639)
8334168 Daniel Han committed 1 month ago
Studio: unload an idle decision model when another model loads (#14555)
2466ce2 Nilay committed 1 month ago
Studio: cut polled-probe cost and release memory sooner (#14595)
173153b Michael Han committed 1 month ago
Studio: keep the image info stamp off by default, recipe stays in the metadata (#14669)
f76dd22 Daniel Han committed 1 month ago
Chat: the composer dock stops searching the whole thread on every com… (#14585)
8920e38 Eyera committed 1 month ago
Native FSDP2 float16: fix fused AdamW crashing at the first step (#14663)
630bdfa Daniel Han committed 1 month ago
Studio: speak through a saved connection on the Audio page (#14288)
b4f3e06 Nilay committed 1 month ago
Windows UI lane: retry the permission suite's auth wipe while the stopped server releases auth.db (#14658)
588e1c4 Daniel Han committed 1 month ago