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
Keep Mamba x_proj and out_proj unquantized for the fused mamba_ssm training path (#15377)
bea21e2 Daniel Han committed 25 days ago
Studio: tighter gap under the Chat with files header (#15523)
c4ac41a Michael Han committed 25 days ago
fix(studio): center sidebar header icons in hover circles (#15528)
98172b7 Wasim Yousef Said committed 25 days ago
Scan baseline: re-review the seven unsloth-zoo sites 2026.10.9 moved (#15524)
ca640b8 Daniel Han committed 25 days ago
Studio ROCm: tell self from cross attention without guarding the query length on the key length (#15458)
88ff2b0 Daniel Han committed 25 days ago
Studio: account hosted quantized components of every image and video family in cache cleanup (#15487)
33ae272 Daniel Han committed 25 days ago
Studio: classify host pinning by the target device, not the current one (#15463)
4f84e8f Daniel Han committed 25 days ago
MXFP4 low memory test: gate the chunked dX budget test on a real CUDA device (#15516)
baa95ba Daniel Han committed 25 days ago
Precision conflict test: lift every loader.py helper from_pretrained reaches (#15515)
8df1aaf Daniel Han committed 25 days ago
Studio: use shared download manager for diffusion training (#15476)
c3df55b Daniel Han committed 25 days ago
tests: stub the tensor parallel guard in the for_training extraction harness (#15498)
945bf3f Daniel Han committed 25 days ago
Studio engines: certifi 2026.10.19 in the vLLM and SGLang locks (#15495)
e87c20b Daniel Han committed 25 days ago
Studio video: size the hosted H3 text encoder before the unified-memory guard (#15474)
a45de19 Daniel Han committed 25 days ago
for_training flag test: lift every _utils helper for_training reaches (#15493)
48b3e0b Daniel Han committed 25 days ago
Studio: keep timestamps off for every GPT transcription model on a connection (#15437)
07efd16 Daniel Han committed 25 days ago
Studio video: size the VAE decode from free memory after the denoiser offload (wider tiles, same clip whether the CUDA graph is kept or dropped) (#15464)
63939ab Daniel Han committed 25 days ago
Studio: orange Full access in permission menus, sandbox docs link, Confirm (#15504)
5933ddc Michael Han committed 25 days ago
MXFP4 packed matmul: bound the dX decode like the forward below sm_80 (#15376)
feef4b7 Daniel Han committed 25 days ago
FSDP2 sharded loading gloo test: barrier before a rank tears down its group (#15499)
ebf98cb Daniel Han committed 25 days ago
Settings contract tests: follow MonitorLink into the API page's access group (#15500)
8a1edb4 Daniel Han committed 25 days ago
fast_inference under torchrun: every rank takes the same vLLM-or-HF path (#15395)
5204566 Daniel Han committed 25 days ago
Routing replay: score named PEFT reference adapters with their own routing (#15387)
0993cc1 Daniel Han committed 25 days ago
QAT: fake quantize embeddings under FSDP1 and torch.func.functional_call (#15385)
4bd1155 Daniel Han committed 25 days ago
FP8 block decode: keep the fused small-M kernel for fp16 when torchao is selected (#15361)
bc0603f Daniel Han committed 25 days ago
Studio tests: probe int8 CPU quantize + flatten apart from the cache assertions (#15459)
da5bbf7 Daniel Han committed 25 days ago
Studio tests: turn Hub telemetry off where mocked Hub requests are counted (#15453)
8045b58 Daniel Han committed 25 days ago
Version compat: transformers 5.21.0 is tagged v5.21 upstream (#15438)
08400e2 Daniel Han committed 25 days ago
Save: compare fused 4-bit expert bytes bitwise for float quant_storage (#15378)
bf5448b Daniel Han committed 25 days ago
Studio diffusion: pin Qwen-Image-2.1's reduction-config filter (same seed, same image across processes) (#15455)
f991be7 Daniel Han committed 25 days ago