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
Gate the fused decode RoPE tests on a real CUDA device, not the spoofable probe (#14661)
d4c1f46 Daniel Han committed 1 month ago
Studio: open localhost addresses typed in the desktop browser panel (#14290)
65db750 Nilay committed 1 month ago
Studio composer icon check: re-measure a case when the dev page reloads mid-case (#14647)
b20f180 Daniel Han committed 1 month ago
Unsloth Studio: serve an MLX model through the library that loaded it (#14612)
f029025 Long Yixing committed 1 month ago
Studio: stop a Clef decision model from outliving a crashed Unsloth (#14556)
458190c Nilay committed 1 month ago
RMSNorm: enable the narrow-row kernels on ROCm with the DPP reduction order (#14637)
9dfba3b Daniel Han committed 1 month ago
Fix save_pretrained_torchao for full finetuning QAT on transformers 5 (#14639)
a81cb77 Daniel Han committed 1 month ago
Studio: load GGUF models faster by skipping the vocabulary read (#14573)
79db5ad Nilay committed 1 month ago
Studio diffusion: test First Block Cache on a two-card FLUX.2 layer split (#14620)
c7e5f20 Daniel Han committed 1 month ago
Studio: stop long MLX chats from re-reading the whole chat every turn (#14574)
a981fa8 Nilay committed 1 month ago
Studio Settings: stop forcing a GitHub release lookup on every Settings open (#14649)
c2f75c4 Michael Han committed 1 month ago
Studio: engine versions and updates in Settings, Use via API for Images and Video, shorter Notifications, rounder waveforms (#14590)
e24fa71 Michael Han committed 1 month ago
MoE LoRA: widen gate_up_proj to split gate_proj / up_proj experts when down_proj already matches (#14628)
5f13f12 Daniel Han committed 1 month ago
NVFP4 fused GEMM: honor autocast dtype and check K alignment for dX (#14625)
d302a58 Daniel Han committed 1 month ago
Studio: default scheme-less public provider URLs to https (follow-up to #14594) (#14629)
cd81b03 Michael Han committed 1 month ago
unsloth start: hard-link only auth.json into the Codex parent overlay (follow-up to #14591) (#14635)
2cc6b8e Michael Han committed 1 month ago
Studio: name an untitled chat's fork after its first reply (follow-up to #14587) (#14636)
c10feb7 Michael Han committed 1 month ago
Studio: keep folder identity lookups inside the scan budget (follow-up to #14588) (#14638)
5f27002 Michael Han committed 1 month ago
Studio: keep Projects page renames and deletes from waiting out a streaming reply (#14634)
7bbaa74 Michael Han committed 1 month ago
Studio: busiest day keeps the darkest token activity shade (follow-up to #14600) (#14630)
a493452 Michael Han committed 1 month ago
Studio: shortcut follow-up to #14597 for Swiss/Turkish digit chords and Esc-to-stop (#14633)
d8d2281 Michael Han committed 1 month ago
Give the single-row cached attention harness CPU stand-ins for the fused decode RoPE kernels (#14632)
d7fe9f2 Daniel Han committed 1 month ago
Studio: refuse Laya choice questions whose options would be cut short (#14558)
817a818 Nilay committed 1 month ago
Studio: keep Decision API models out of the chat model picker (#14580)
0ae76cc Nilay committed 1 month ago
Studio: accept skills whose description has an unquoted colon (#14537)
57f4e85 Nilay committed 1 month ago
Studio: say when an attached image is over the 20 MB limit (#14538)
3ba06aa Nilay committed 1 month ago
NF4 bitsandbytes GEMM gate: return before its lazy setup under torch.compile (#14608)
356721b Daniel Han committed 1 month ago
Studio Images: SDXL prompt weighting, CLIP skip, CFG rescale and variation seeds (#14562)
d277aa6 Daniel Han committed 1 month ago