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 video small host: price the encoder converted beside the cast misses, cast an oversized DiT from its source shards (#16700)
3c88a25 Daniel Han committed 16 days ago
Studio: read the SDXL v-prediction markers beside the weight load and once per commit, skip the UNet scheme probe (#16696)
5c2d611 Daniel Han committed 16 days ago
Studio: keep Qwen-Image's bf16 text encoder automatically when the card has room (#16678)
1ca9e5d Daniel Han committed 16 days ago
Kernels: keep sequence length out of Triton specialization in the RoPE kernels (#16690)
2a256a1 Daniel Han committed 16 days ago
Refuse tensor parallel on blocks that own parameters in the registry (#16638)
e2ffc74 Daniel Han committed 16 days ago
Studio: fold an unchanged LoRA set into a resident denoiser, restore it exactly before a change (#16675)
f5935a7 Daniel Han committed 16 days ago
Studio: send custom headers on remote connections (#15097)
14b0166 Sourav Rajvi committed 16 days ago
Carry the float8 training mode into the parallel capability query (#16637)
4e9a0c4 Daniel Han committed 16 days ago
Bump install.sh / install.ps1 pins to unsloth>=2026.11.2 (#16688)
29b2359 Daniel Han committed 16 days ago
Flex wrapper: keep the causal mask inside compiled blocks on torch < 2.10 (#16234)
2bf0053 Yufan Zhuang committed 16 days ago
Studio: drop the unfused QKV copies after the SDXL fuse, and decline the speed profile when kept copies would not fit (#16621)
914f10b Daniel Han committed 16 days ago
Studio: saving a preset no longer brings it back on the model it was tuned on (#14603)
f7769b4 Roydon Sequeira committed 16 days ago
Studio: forward a request's other chat_template_kwargs to llama-server (#16567)
accbbcc Debojyoti Banerjee committed 16 days ago
Studio i18n: translate the browser tool steps and program download prompts in every locale (#16677)
7d896d8 Daniel Han committed 16 days ago
Fused preference loss: fall back when a global forward hook is registered (#15226)
3e12d46 Shubhankar Tripathy committed 16 days ago
fix(studio): skip hidden HTML subtrees in RAG parsing (#16612)
74c9290 LimbC-C committed 16 days ago
Confine Windows glass wallpaper blur to the inner page pane (#16671)
b43a0df Lee Jackson committed 16 days ago
fix(studio): rewrite S3 audio paths in single-record JSON manifests (#16631)
14bd64f Max-Reisinger committed 16 days ago
Studio: MLX training from a saved adapter, Continued Pretraining, streamed and audio datasets, with exact resume (#16622)
8611f68 Long Yixing committed 16 days ago
Studio: let search_conversation find this turn's results on API providers (#16666)
d80856c Daniel Han committed 16 days ago
cuDNN SDPA: keep known-bad decode and NaN-dQ shapes off cuDNN before torch 2.14 / 2.15 (#15934)
8161e81 Daniel Han committed 16 days ago
Studio video: load a commit-pinned cached snapshot offline when refs/main is missing (#16640)
7e92b5b Daniel Han committed 16 days ago
Studio H3: release resident denoiser blocks' host copies and give the RAM to the conditioner (#16639)
4d74b01 Daniel Han committed 16 days ago
Model picker contracts: read the GGUF row and gear pins from variantSelectMeta (#16669)
ab96fb1 Daniel Han committed 16 days ago
Legacy CLI: apply --chunk_size and --stride to a text file passed as --dataset (#16424)
047030f Vineeth Sai Varikuntla committed 16 days ago
Block Swap blocker: Qwen3.8-Flash-Next-FP8 can't train with offload_layers, its FP8 n-gram table loads without its weight_scale (#15632)
2ed0e98 Leo Borcherding committed 16 days ago