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: modal opens stop re-laying out long chats in WebKit (#17284)
68399c7 Daniel Han committed 10 days ago
Studio: key group hover / focus-within / active on the group class (WebKit long-chat menus and dialogs) (#17323)
433317f Daniel Han committed 10 days ago
Studio: colour chat code blocks from the worker while scrolling and streaming (#17160)
c71abe2 Daniel Han committed 10 days ago
Studio ROCm TunableOp: label the results file with the HIP build string, not a raw read (#17325)
1ae69bf Daniel Han committed 11 days ago
Windows install: keep typer below 0.29 so studio setup does not hold pydantic_core open (#17319)
f3ebd48 Daniel Han committed 11 days ago
Studio: browser pane opens without per-frame thread work on long chats (#17282)
ab9845e Daniel Han committed 11 days ago
Studio: stream a reply without re-rendering every message client per token (#17305)
c5c4162 Daniel Han committed 11 days ago
Studio: keep the view_image picture visible when its tool card is collapsed (#17174)
b335310 Keyi Yu committed 11 days ago
Studio: Glass Settings pane scrolls on the compositor; API monitor fade pauses its blur (software compositing) (#17308)
fefb942 Daniel Han committed 11 days ago
Studio: title tooltips stop walking the whole thread on every pointer move (#17304)
b0e84bf Daniel Han committed 11 days ago
Use one glass surface for the docked Windows browser (#16754)
820042f Lee Jackson committed 11 days ago
Studio diffusion: keep the fused AdaLN census out of compiled FLUX.1 blocks (recompile limit on first render) (#17310)
e1e9a7c Daniel Han committed 11 days ago
Studio native diffusion: run the text encoders on a second CUDA card when one card cannot hold the load (#17276)
c38dc70 Daniel Han committed 11 days ago
Long-chat gate: measure every appearance setting or allowlist it with a reason (#17262)
655b309 Daniel Han committed 11 days ago
Long-chat gate: measure streaming a reply into the chat (#17269)
0739e88 Daniel Han committed 11 days ago
Long-chat gate: settle pointer targets before a trusted hover or click (#17274)
c2eedcb Daniel Han committed 11 days ago
Studio ROCm: use the measured fastest GEMM solution table for diffusion denoisers on gfx1151 (#17299)
a784ee5 Daniel Han committed 11 days ago
Studio tests: skip the FLUX.3 group-offload test where diffusers is not installed (#17317)
671d9d5 Daniel Han committed 11 days ago
Studio diffusion: split the batch across two cards for LTX-2, Chroma, Qwen-Image, HunyuanImage and Krea-2, and pair Cosmos 3 guidance (#17278)
14a59fd Daniel Han committed 11 days ago
Studio tests: pin the H3 cached-prequant lookup offline and cover the online twin rule (#17309)
19cc609 Daniel Han committed 11 days ago
Studio diffusion: auto attention keeps native flash SDPA on A100 / L4 class cards (#17290)
a691227 Daniel Han committed 11 days ago
Studio: serve the hosted Wan2.2-A14B int8 pair to an offloaded load as torchao-free W8A8 layers (#15240)
1b88784 Daniel Han committed 11 days ago
Studio: give FLUX.3's per-mode block stacks their own offload groups (5 GiB L4 int8: -17.6% per image) (#17122)
d17d054 Daniel Han committed 11 days ago
Studio tests: stub stt-context when mic-target loads the voice settings store (#17293)
0f6c2e5 Daniel Han committed 11 days ago
Studio: load image / video / text-encoder weights from .safetensors first, reuse a cached .pt / .bin when its sha256 matches the Hub (#17061)
2fb404b Daniel Han committed 11 days ago
Studio CFG-parallel: treat an NVFP4 precision-state flip as a cold call so the two branches never compile at once (#14854)
6d2e6bf Daniel Han committed 11 days ago
Studio ROCm VAE attention: math SDPA in the VAE's bf16/fp16 dtype, not an fp32 upcast (#17252)
1caab2b Daniel Han committed 11 days ago
Studio ROCm layer split: probe secondary cards read-only, at the attention dtype (#14655)
10c52ed Daniel Han committed 11 days ago
gloo-tp tests: a rank that finished is not failed by an abort in interpreter teardown (#17244)
c405205 Daniel Han committed 11 days ago