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: raise the startup bundle budget by #13989's growth (#13995)
1133b82 Daniel Han committed 1 month ago
Tests: gate the new Q-GaLore offload CUDA case on a real CUDA device (#13994)
0669098 Daniel Han committed 1 month ago
head_dim 512 attention: sm80+ only, no 256 MB autotune buffer, OOM falls back to SDPA, no per-call mask sync (#13977)
1574001 Daniel Han committed 1 month ago
Studio: Chat with files X hides the panel, app tooltips, and higher-contrast traces (#13989)
b0e74e7 Michael Han committed 1 month ago
Studio: raise the startup bundle budget by the growth since 17901061d1 (#13991)
6583ae4 Daniel Han committed 1 month ago
GRPO: latch the trainer's autocast precision on the efficient logprob path (#13968)
02d7f8e Daniel Han committed 1 month ago
Add `unsloth doctor --gpus` multi-GPU preflight and UNSLOTH_DEBUG_DISTRIBUTED (#13884)
e61d529 Daniel Han committed 1 month ago
FSDP2: save_pretrained, merged saves and 4-bit checkpoints for sharded models (#13895)
81e51f6 Daniel Han committed 1 month ago
FSDP2: shard QLoRA 4-bit weights with fused LoRA kernels kept on (#13880)
086db55 Daniel Han committed 1 month ago
GRPO no-grad packing: check the OOM length cap before flattening ids (#13980)
6106de9 Daniel Han committed 1 month ago
Studio: hide safetensors models that exceed VRAM when the fit filter is on (#13945)
6c555fa oobabooga committed 1 month ago
RMSNorm save output: skip inference tensor weights, keep the default path free of opt-in checks (#13981)
adebbfa Daniel Han committed 1 month ago
GRPO: keep TRL vLLM weight sync on TRL < 0.28 unless fast_inference shares weights (#13911)
36878d6 Daniel Han committed 1 month ago
Studio: keep the tool results of the turn in progress when compaction resets mid-turn (#13951)
08fb7bc oobabooga committed 1 month ago
Sync full finetune weights into colocated vLLM in a few load_weights calls (#13901)
6c229ce Daniel Han committed 1 month ago
Studio: even 17px spacing in the Hub filter popover (#13972)
c401482 Michael Han committed 1 month ago
Studio: Models panel part of its card, even form footer, OpenRouter mark under a new file name (#13974)
147b694 Michael Han committed 1 month ago
Studio: draw chat markdown tables inline with hover controls and a centered expand dialog (#13973)
ce83dee Michael Han committed 1 month ago
Studio: rank Latest Unsloth Models by trending weighed against release age (#13969)
7bee02d Michael Han committed 1 month ago
Studio: resizable Model hub split view, smaller row kebab, centred token icon (#13966)
cc0cb59 Michael Han committed 1 month ago
Studio: fix the context meter after tool calls on MLX and transformers (#13913)
8d85be6 Nilay committed 1 month ago
Studio: line the composer up with the chat column in the macOS Desktop app (#13965)
7d33eff Michael Han committed 1 month ago
Studio: train a structured column's conversation as a multi-turn chat (#13940)
f75efb9 Nilay committed 1 month ago
Context parallel: compose with FSDP2 on a (dp_shard, cp) mesh (#13910)
4ec8198 Daniel Han committed 1 month ago
Fix optimizer state offload crash with Q-GaLore weight quantization (#13979)
ee84654 Daniel Han committed 1 month ago
Studio: honour MIG masks and size MIG cards by their slice (#13927)
23bbee0 Daniel Han committed 1 month ago
Studio: let safetensors and MLX vision models read transparent images (#13939)
159084e Nilay committed 1 month ago
Studio: raise the startup bundle budget by the growth since 587079569a (#13983)
1790106 Daniel Han committed 1 month ago
Tests: gate the optimizer state offload CUDA cases on a real CUDA device (#13982)
638578a Daniel Han committed 1 month ago
GRPO: opt-in zero-variance group exclusion and prompt-level loss aggregation (#13891)
fe3370a Daniel Han committed 1 month ago