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: only resolve int8 text encoder files that are published (#17503)
0912949 Daniel Han committed 9 days ago
Studio: prebuilt CUTLASS int8 GEMM with the dequant epilogue fused, raced per arch (FLUX.3 klein INT8 -19.8% on L4) (#17486)
52cea66 Daniel Han committed 9 days ago
Bump install.sh / install.ps1 pin to unsloth>=2026.11.4 (#17524)
fcae458 Daniel Han committed 9 days ago
Studio: pin MiniMax-H3-Image (Fast FP8) at the top of the image picker (#17502)
475ceee Daniel Han committed 9 days ago
Studio: no RAM fallback mark in the model picker when there is no GPU (#17516)
0a12cb2 Michael Han committed 9 days ago
Studio: MLX chats with a picture keep going when the context fills (#17424)
31d7eb4 Nilay committed 9 days ago
Studio: train jobs targeting embed_tokens or lm_head alone, not batched (#16653)
68f2472 Daniel Han committed 9 days ago
RoPE: compile stock positions without full-context caches (#16182)
6b01508 Daniel Han committed 9 days ago
Studio desktop: serve the web UI behind an allowlisted reverse proxy (fixes #17420) (#17435)
cc655b1 Debojyoti Banerjee committed 9 days ago
Ulysses: dynamic and longrope RoPE see the whole sequence's positions (#17051)
ccbd70b Shubhankar Tripathy committed 9 days ago
Pipeline parallel on the CPU backend: weight the MoE router aux loss as TRL's SFTTrainer does (#17072)
e64acaa Debojyoti Banerjee committed 9 days ago
Mask GRPO completions cut by the vLLM server's context limit (#16650)
6f87cf8 Daniel Han committed 9 days ago
Context parallel: eval loss is the sequence's token mean, not cp_size times it (#17049)
76f3834 Shubhankar Tripathy committed 9 days ago
TP / FSDP2 saves: pass results between ranks as JSON, not pickle (#15510)
c5b13a0 Daniel Han committed 9 days ago
Studio sd.cpp: pin the prebuilt with the aarch64 K-quant GEMM and macOS Accelerate routing (#17489)
8da0386 Daniel Han committed 9 days ago
Gemma 3 generate test: find the non-flex scope by call, not by its exact text (#17497)
714073e Daniel Han committed 9 days ago
Studio: edit images through an external connection's /images/edits endpoint (#17448)
63ef0fa oobabooga committed 9 days ago
Studio MiniMax-H3: opt-in 4-bit GGUF text encoder for hosts that cannot keep int8 resident (#17479)
568e79e Daniel Han committed 9 days ago
Studio: reconnect a chat reply whose connection died silently (#17416)
cc38d58 Nilay committed 9 days ago
Studio: compact long chats on an Ollama connection to its loaded window (#17457)
e550e39 Nilay committed 9 days ago
Studio: keep built-in tools on when an MLX chat has several images (#17423)
d80e9a5 Nilay committed 9 days ago
Studio: let managed accounts use gated models with their own HF token (#17172)
d4d5fff Nilay committed 9 days ago
Studio: keep the LoRA on in Chat after a Base vs LoRA compare (#17433)
6e53779 Nilay committed 9 days ago
Studio: fit max_tokens to the window on vLLM and SGLang tool requests (#17426)
b347d06 Nilay committed 9 days ago
Studio: stop showing a stray </think> after an OpenAI model calls a tool (#17455)
13a92a3 Nilay committed 9 days ago
Studio: keep the seeded admin password out of pages relayed by a proxy (#17454)
a7a7186 Nilay committed 9 days ago