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: validate each dataset snapshot once per provenance attestation (#16790)
18b4c0d Daniel Han committed 14 days ago
Studio MiniMax-H3: opt-in FlashInfer SM120 fused block kernels for FP8 and NVFP4 (#16875)
790595d Daniel Han committed 14 days ago
Studio: build the video backend off the event loop in /video/status (#16792)
c4a3025 Daniel Han committed 14 days ago
Studio: round the MLX warmup and eval ratios up like transformers (#16827)
d90dae6 Vineeth Sai Varikuntla committed 14 days ago
Studio: probe STT Transformers availability off the event loop (#16783)
d44fd45 Daniel Han committed 14 days ago
Studio: draw nothing for the glass lens when it is Off (glass 1.4 to 2.1x faster without GPU rendering) (#16852)
767372b oobabooga committed 14 days ago
Studio: CI gate that fails when typing in a long chat gets slower as the chat grows (#16801)
8f4206c Daniel Han committed 14 days ago
Context parallel: accept Unsloth's fast Qwen3 attention and compiled FastModel attention (#16870)
5c9d1af Daniel Han committed 14 days ago
Pipeline parallel: refuse the MoE router aux loss over several micro-batches at Trainer init (#16822)
41c4f6f Daniel Han committed 14 days ago
Studio: stop read aloud from waiting on a model catalog rescan (#16831)
941722b oobabooga committed 14 days ago
Studio: clear the API key when the connection type changes (#16842)
f8eedf8 Nilay committed 14 days ago
Studio: name the runtime and its release in the runtime update card (#16880)
ed76754 Michael Han committed 14 days ago
Context parallel: give decoder layers no mask inside a CP step so block compile keeps one graph on packed rows (#16862)
6953b1c Daniel Han committed 14 days ago
Studio diffusion: measured activations for Lumina-Image-2.0 (int8 stays resident on 16-24 GB cards, ~3% per render) (#16808)
b75a4cd Daniel Han committed 14 days ago
Studio LTX-2.5: keep a streamed denoiser partly resident per request (#16845)
fdd45f4 Daniel Han committed 14 days ago
Studio LTX-2.5: release the server's cached GPU blocks before an encode, so later prompts stop encoding on the CPU (#16804)
2821cbb Daniel Han committed 14 days ago
Pipeline parallel on CPU: run the chunked cross entropy on the last stage (#16846)
a2d21d9 Daniel Han committed 14 days ago
Installer mirror test: bound the probe speed by the batch time, not a fixed 1 KiB/s (#16871)
9c1cffb Daniel Han committed 14 days ago
Mac CI: run the permission and loaded-models indicator suites alongside the UI suites (#15703)
e002fda Jonathan Chen committed 14 days ago
Studio: prompt More menu for Fork and Bookmark, and no composer scrollbar flash (#16866)
300baed Michael Han committed 14 days ago
Studio diffusion: protect the first and last NVFP4 steps by default on sm_100-class cards (W4A16 auto) (#16759)
e79f756 Daniel Han committed 14 days ago
Studio chat UI test: drag the monitor grip again when a native resize stops early (#16861)
f988934 Daniel Han committed 14 days ago
Studio MiniMax-H3: price the slab compaction against the host reserve so a 16 GiB L4 render is not OOM-killed (#16849)
8d54207 Daniel Han committed 14 days ago
Studio video memory plan: count the forward's headroom before keeping whole-module offload (#16805)
44d3664 Daniel Han committed 14 days ago
Studio: let the floating browser go anywhere, and float per chat (#16859)
2193ee8 Michael Han committed 14 days ago
Studio diffusion training: no per-micro-step host syncs, one decode per cached image, streamed int8 base load (#16763)
da412c0 Daniel Han committed 14 days ago
Studio bookmarks test: click Follow link again when the click does not navigate (#16858)
d87c41c Daniel Han committed 14 days ago
Studio sd.cpp: read the linked CUDA major from DT_NEEDED so a CUDA 13 build gets no cu12 cuDNN (#16789)
c1cb7db Daniel Han committed 14 days ago
Studio sd.cpp installer: retry a transient download failure before stepping down to the CPU build (#16788)
9d40777 Daniel Han committed 14 days ago