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: explain failed and recurring llama.cpp runtime repairs (#12106)
ca5e2fd UmranPros committed 1 month ago
Studio: fix prompt timestamp layout, accessibility, and imported times (#12173)
501b15f Michael Han committed 1 month ago
feat(studio): default the model source to ModelScope where Hugging Face is restricted (#12197)
53b36ea Long Yixing committed 1 month ago
Library delete dialog: no dark mode border, cancel on outside click (#12222)
282d170 Michael Han committed 1 month ago
Images: smaller prompt box radius, inset scrollbar, taller default (#12223)
d670959 Michael Han committed 1 month ago
Studio: load Images strip tiles as thumbnails instead of full PNGs (#12228)
7965e20 oobabooga committed 1 month ago
Scope the runtime encoding scan past vendored laya while its loader supplies UTF-8 (#12230)
e7811ae Daniel Han committed 1 month ago
Studio: load Laya without randomly initialising its vocabulary embedding (#12210)
7e732ba Daniel Han committed 1 month ago
Studio: clamp unsupported reasoning effort to offered levels (#12208)
1a46fe1 oobabooga committed 1 month ago
Unsloth Studio (AMD/Windows): real torchao INT8 / FP8 export on Windows ROCm (#7102)
7b07d86 Daniel Han committed 1 month ago
fix(studio): reuse hardware monitor drag and resize handling for API monitor (#12204)
5141125 Lee Jackson committed 1 month ago
Studio: resolve repo-relative image paths in the simple image+text VLM converter (#11897)
52b822c breken committed 1 month ago
fix: align GKD Liger loss with the native objective (#12007)
87964a3 cherry77-cloud committed 1 month ago
Train Kimi-K3 with its MXFP4 experts kept packed and dequantized on the fly (#11750)
f7da293 Daniel Han committed 1 month ago
Studio: say why a generation or a model load failed, and offer the log (#8804)
3ae1c29 Daniel Han committed 1 month ago
Default loader.py's Mistral-format names in the precision-conflict harness (#12229)
e3a7cf7 Daniel Han committed 1 month ago
fix(studio): prevent terminal tool hangs in credential scan (#12048) (#12087)
1499c15 Sourav Rajvi committed 1 month ago
Load compressed-tensors packed INT4 checkpoints for QLoRA without a 16-bit copy (exact packed INT4 by default, NF4 fallback), and train remote DeepSeek-V3 MoE code (Kimi-K2.7-Code) (#11537)
6ee4c0c Daniel Han committed 1 month ago
Repair the checks red on main after today's Mistral-format, GKD, laya, llm-compressor and HF token merges (#12218)
eb91752 Daniel Han committed 1 month ago
Point the chat export contract at the Library and project menu, where #12122 moved export (#12203)
8d675fd Daniel Han committed 1 month ago
Give the Kimi processor test's stand-in tokenizer convert_tokens_to_ids (#12206)
b67b334 Daniel Han committed 1 month ago
Bump install.sh / install.ps1 pin to unsloth>=2026.9.12, unsloth-zoo>=2026.9.8 (#12209)
a7c66df Daniel Han committed 1 month ago
GKD: right-align left-padded rows before the student forward (#12200)
05d47bd Daniel Han committed 1 month ago
Keep xFormers attention causal when the decoder is called without a mask (#12199)
825c1f2 Daniel Han committed 1 month ago
Unpack the mirrored uv wheel with python3 -m zipfile instead of an inline one-liner (#12194)
78e634c Daniel Han committed 1 month ago
Studio: vendor laya 0.3.5 and hold its weights in float16 (#12202)
4ab5c77 Daniel Han committed 1 month ago