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: Skills pill stays with every skill unticked; its X or the + menu turns Skills off (#13687)
61d17ec Michael Han committed 1 month ago
Studio: train structured recipe columns as JSON, not as Python dicts (#13635)
ca725cf Nilay committed 1 month ago
Studio: one scrolling attachment row once a chat has messages; even the gap above the composer input (#13678)
5d1aa42 Michael Han committed 1 month ago
Studio: Chat with files search row lists its documents, which open in the attachment viewer (#13681)
f075667 Michael Han committed 1 month ago
Studio: Windows torch 2.11 kernels from our prebuilt wheels (#13564)
af9d978 Daniel Han committed 1 month ago
Model-config UI test: treat a /validate that never reports its end as a lost event (#13689)
07fea4d Daniel Han committed 1 month ago
Studio: make the first-token waiting limit configurable (#13390)
4e65841 Lee Jackson committed 1 month ago
Studio: size the composer to the message column at every width and zoom, and stop clipping thinking text (#13675)
2bb3b63 Michael Han committed 1 month ago
Studio: declare the Hub's /api/engines retry in the log budget (#13682)
a413834 Daniel Han committed 1 month ago
Studio: LTX-2.5 defaults follow ComfyUI (hosted INT8-ConvRot DiT, opt-in int8 text encoder) (#13618)
4e282d4 Daniel Han committed 1 month ago
End the last gpt-oss training turn with <|return|>, not <|end|> (#13657)
48ca6b7 Nilay committed 1 month ago
Studio UI test: wait for the chat shell to settle, not just the composer (#13672)
757ae1b Daniel Han committed 1 month ago
Studio CI: click the composer itself in the dismiss_on_composer probe (#13670)
cac08dc Daniel Han committed 1 month ago
Studio: key the i2v availability cache on the diffusers module, not its id (#13667)
95f51c8 Daniel Han committed 1 month ago
Native decode: read GQA KV heads in place instead of copying them per query head (#13271)
cc7e8aa Daniel Han committed 1 month ago
GRPO DDP gradient sync: copy each averaged bucket back with one _foreach_copy_ (#13268)
f440d28 Daniel Han committed 1 month ago
Studio: keep data recipe seed inspect inside a managed account's workspace (#13611)
1c3e50e Daniel Han committed 1 month ago
FP8 MoE + fast_inference: refuse unsupported checkpoints before vLLM loads, with a hint that works (#13277)
dc9b342 Daniel Han committed 1 month ago
Studio: live denoising canvas for transformers diffusion LMs (#13593)
add9d2a Daniel Han committed 1 month ago
Studio: point unsloth start at the right model when several are loaded (#13480)
b87ab5e Nilay committed 1 month ago
Tests: keep the zoo num_proc anchor tests meaningful once the zoo adopts the edit (#13269)
b054414 Daniel Han committed 1 month ago
QAT: fall back to the module forward for torchao NVFP4 / MX linears in fused LoRA (#13280)
7d132bb Daniel Han committed 1 month ago
tests: accept a positional encoding on Path.read_text / write_text (#13646)
7815d91 Daniel Han committed 1 month ago
Studio: send routine coding work to the local model with unsloth start --auto-route (#13478)
b0a7f1e Nilay committed 1 month ago
Studio: reload a Clef as soon as the training run that unloaded it ends (#13482)
9b30c4f Nilay committed 1 month ago
fast_cross_entropy_loss: all-masked labels give loss 0 instead of NaN (#13325)
464d369 Daniel Han committed 1 month ago
Diffusion LMs: fast inference, live denoising frames and in-loop cancellation (#13592)
f798045 Daniel Han committed 1 month ago