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: drop the empty bubble under attachment-only user messages (#12799)
61afc99 Michael Han committed 1 month ago
Load the compiled class when auto_model is a concrete class (#12757)
56a520c Daniel Han committed 1 month ago
scan-packages baseline: re-review datasets 5.1.0's readline loop (#12784)
87f0a8e Daniel Han committed 1 month ago
MLX backend tests: answer zoo's _fusion_modules in the inference stub (#12783)
0700115 Daniel Han committed 1 month ago
Studio: tidy the model picker's On Device columns and list inset (#12746)
7093e0c Michael Han committed 1 month ago
Studio: use the edge Gemma 4 template for the QAT E2B/E4B repos (#12715)
7f4945c Pundarikaksh Narayan Tripathi committed 1 month ago
Provider URL validation tests: count only the lookups the code under test made (#12771)
e3e67e0 Daniel Han committed 1 month ago
version-compat latest lanes: install requests, stop shadowing torchcodec (#12769)
0095a66 Daniel Han committed 1 month ago
Transcript stream: never drop a phase change behind a newer update (#12760)
858381b Daniel Han committed 1 month ago
Re-measure the Studio startup budget after the Sandbox tab and action bar rework (#12763)
d68515a Daniel Han committed 1 month ago
SDPA packed-segment test: gate the CUDA memory test on a real GPU (#12761)
268f88e Daniel Han committed 1 month ago
studiobench: delete through the reply's More menu, where #12735 moved it (#12748)
f423a8b Daniel Han committed 1 month ago
Recount harness: bind the loader's msgs the history restore now estimates from (#12747)
f865421 Daniel Han committed 1 month ago
Live Monitor models disk: translate its labels and count every tile (#12749)
cac19c3 Daniel Han committed 1 month ago
Studio: seam-free tiled decode for every image VAE whose stock tiles fall under the floor (HunyuanImage-2.1 grid, Qwen-Image, slivers) (#12736)
32ee4bf Daniel Han committed 1 month ago
Deep research contract: follow the research Delete gate into the More menu item (#12751)
36255a1 Daniel Han committed 1 month ago
Run bitsandbytes NF4 Linear4bit through Unsloth's NF4 kernels (#12739)
102dede Daniel Han committed 1 month ago
fix: enable XPU support for tests with agnostic test selection (#8319)
67e67ad JoshuaL3000 committed 1 month ago
Studio: keep Wan2.2-5B's leading blocks resident under group offload (#12691)
6f4d997 Daniel Han committed 1 month ago
Studio: experimental Int8 Prefill run setting for MLX models (#12550)
cc84c7d Long Yixing committed 1 month ago
Unsloth Desktop/Studio audio: follow-ups from testing the Clone, Transcribe, Edit, Separate and Convert pages (#12706)
0108742 Leo Borcherding committed 1 month ago
offload_layers='auto': plan for batch size 2, re-plan at Trainer init for the real batch (#12741)
01254e6 Daniel Han committed 1 month ago
SDPA: attend packed rows per segment instead of under a dense mask (2.9x less memory, 2.5x faster without flash-attn / xformers) (#12740)
5babb22 Daniel Han committed 1 month ago
Studio: int8 ConvRot text encoder for Qwen-Image-2.1, fp8 fallback (#12684)
9bf3dec Daniel Han committed 1 month ago
Studio: race address families on sync connects (RFC 8305 Happy Eyeballs) (#10803)
954bf81 Wenrui LI committed 1 month ago
Studio: whole-step CUDA graphs under offload on top of per-block graphs (L4 FLUX.1 16 GB 10% faster, HunyuanVideo-1.5 1 GiB lighter) (#12707)
8de2cde Daniel Han committed 1 month ago
perf(studio): keep the MLX GPU warm for a bounded time after generation (#12721)
d4bfbd9 Long Yixing committed 1 month ago
Studio: per-block CUDA graphs under every offload mode, and compile below the offload hooks (#12694)
81f1a43 Daniel Han committed 1 month ago