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: run FLUX.2 RoPE as one fused kernel on fp16 GPUs (klein step 5% faster on T4, bit-identical) (#12503)
864ab9b Daniel Han committed 1 month ago
studio: reuse ssl setup for local gguf requests (#12495)
6316710 Maheswar Kumar committed 1 month ago
Studio: plan image offload on measured activations so more VRAM is never slower (#12043)
856af9e Daniel Han committed 1 month ago
Studio: torch 2.13 for new Linux cu130 Python 3.13 installs, existing installs keep their torch (#12150)
eaef459 Daniel Han committed 1 month ago
Studio: keep GGUF weights and the last-run module on their host copy under whole-model offload (#12022)
75323d5 Daniel Han committed 1 month ago
Studio: let the compile cache keep denoiser graphs with sdpa_kernel blocks (restart first render 1-5.5 s faster) (#12510)
2eaf5b4 Daniel Han committed 1 month ago
Studio: faster cold start for image and video loads (first image on H100 7 to 37% sooner) (#12496)
01d9f5a Daniel Han committed 1 month ago
Studio: wait for the torch warm's dynamo import before a load's first download (#12472)
923b8f4 Daniel Han committed 1 month ago
Studio: fused int8 GEMM with the dequant epilogue, bf16 out (Qwen-Image-2.1 14-17% faster per step) (#12448)
66453bb Daniel Han committed 1 month ago
Name the matching build when flash-attn or causal-conv1d was built for another torch (#12211)
f9672f3 Daniel Han committed 1 month ago
Studio: resume NPU model downloads, keep their progress, and list NPU models in the Hub (#12512)
1fd53df oobabooga committed 1 month ago
Core: pip extras and auto-install support for torch 2.13 and 2.14 (#12151)
0eb7837 Daniel Han committed 1 month ago
SentenceTransformer: opt-in encoder unpadding via shared attention dispatch (#4460)
615eed1 Etherl committed 1 month ago
Studio: opt-in systemd user service for Linux auto-start (#9308)
fcd16f7 Sourav Rajvi committed 1 month ago
Studio: canonicalize and validate web_search arguments (#9716)
57ed76c Sourav Rajvi committed 1 month ago
Studio: resume a reply that stopped mid-thought on MLX (#12504)
95531ba Long Yixing committed 1 month ago
Studio: train vision models on images in a list column or inside the messages (#12484)
aff8dfb Nilay committed 1 month ago
fix(studio): block-split long backslash lines without inline lex (#11376) (#11501)
c5f1f79 Sourav Rajvi committed 1 month ago
Studio: extend the ROCm MIOpen cutoff and Qwen-Image-2.1 query chunking beyond gfx1151 (#12471)
96c8bd1 oobabooga committed 1 month ago
Studio: run MiniMax-H3's Diffusers INT8 path on 24 / 16 / 12 GB cards and near the host RAM floor (#12409)
5502355 Daniel Han committed 1 month ago
perf(studio): enter the MLX routed-experts and norm-handoff fusions per request (#12422)
569a7e5 Long Yixing committed 1 month ago
Add install_missing_dependencies to opt out of the FP8/FP4 llm-compressor auto-install (#9405)
aa56f66 Sourav Rajvi committed 1 month ago
Studio: restore markdown chat import and bulk export (#12513)
1670be3 Sourav Rajvi committed 1 month ago
Studio: stop leftover dataset columns becoming the system prompt (#12485)
efe3860 Nilay committed 1 month ago
Studio: put llama-server warnings and errors in the default session log (#10848)
8947133 Sourav Rajvi committed 1 month ago
Baseline the nine unsloth-zoo 2026.9.9 findings after review (#12515)
285d157 Daniel Han committed 1 month ago
Keep a Data Recipes publish private when the Hub repo already exists (#12482)
0479621 Nilay committed 1 month ago
fix(studio): allow Enter to send with idle macOS Pinyin IME (#12138)
05d2b8d Sourav Rajvi committed 1 month ago