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
Keep embedding optimizer state 32-bit when embedding_learning_rate is set (#12458)
0c5c476 Vineeth Sai Varikuntla committed 1 month ago
Unsloth Studio: managed accounts keep their model settings and pins across an account switch (#12549)
237e0b6 Leo Borcherding committed 1 month ago
Studio: skip the unused Qwen-Image-2.1 text-encoder lm_head (0.6 to 1.3 GiB less VRAM, bit-identical images) (#12669)
bd58e1f Daniel Han committed 1 month ago
Studio: stop the Windows event loop spinning when its self-pipe is closed (#12665)
6e1683b Etherl committed 1 month ago
Studio: keep managed accounts off owner files and device nodes (#12661)
250abc7 Daniel Han committed 1 month ago
Studio: encode video mp4 with x264 frame threads (export 1.25 to 2.5x faster, same settings) (#12648)
65629d0 Daniel Han committed 1 month ago
Harden npm and yarn installs against supply-chain attacks (#12660)
4d16e94 Daniel Han committed 1 month ago
Studio: gate repo-hosted embedder modules, MLX model_file, and _socket (#12659)
5971d28 Daniel Han committed 1 month ago
Studio: bump PyJWT, urllib3, Pillow and frontend overrides for security advisories (#12656)
f3ed79c Daniel Han committed 1 month ago
Guard against transformers config and chat template CVEs on older versions (#12658)
089f756 Daniel Han committed 1 month ago
Harden torch.export .pt2 loading against CVE-2026-4538 (#12657)
35b90da Daniel Han committed 1 month ago
Follow the multi-model refactor in two guards that went red on main (#12653)
3afc0b5 Daniel Han committed 1 month ago
Studio: prefetch an image load's weights and upload host tensors through a pinned ring (#12590)
124e89a Daniel Han committed 1 month ago
Unsloth Studio: export or convert a GGUF to FastFlowLM Q4NX for the AMD Ryzen AI NPU (#12541)
214f4e2 Leo Borcherding committed 1 month ago
Accept a list train_dataset again on TRL 1.10+ (fixes vision notebooks) (#12628)
e462118 Daniel Han committed 1 month ago
Studio: stop Whisper dropping sentences from clips longer than 30 seconds (#12481)
f7349b6 Nilay committed 1 month ago
Record the Wan fused block's import_module in the risky loader baseline (#12643)
a25b166 Daniel Han committed 1 month ago
Studio: keep the whole int8 Qwen-Image-2.1 denoiser resident at 12 GB, released only while the encoders run (298 to 212 ms per step) (#12536)
80c92a6 Daniel Han committed 1 month ago
studio: let vision models view workspace images in code mode (#12591)
39916b8 Maheswar Kumar committed 1 month ago
Studio: engage the fused int8 GEMM when the placement pins the whole denoiser, and restore torchao weights after an oversized request (#12524)
6a0cd59 Daniel Han committed 1 month ago
Studio: let small models read a long attached file when asked to summarize it (#12580)
dc8ede0 Nilay committed 1 month ago
Keep torch deprecation warnings at the caller through the __getattr__ wrapper (#12635)
dcda825 Daniel Han committed 1 month ago
Studio: fuse MiniMax-H3's ConvRot rotation into the int8 activation quant (#12517)
34b87cb Daniel Han committed 1 month ago
Studio: MiniMax-H3 int8 Linears take the fused-dequant int8 GEMM (A100 9% faster per step, ahead of ComfyUI) (#12507)
bbe2467 Daniel Han committed 1 month ago
Studio: MiniMax-H3 attention fast path and fused q/k norm + RoPE (5-7% faster per step) (#12459)
366d907 Daniel Han committed 1 month ago
Studio: hold cudnn.benchmark off for Qwen-Image, HunyuanVideo-1.5, FLUX.1, Z-Image and SDXL (#12568)
45dd665 Daniel Han committed 1 month ago
Studio: hold cudnn.benchmark off for Wan so renders are reproducible across servers (#12565)
efc6d60 Daniel Han committed 1 month ago
Fix text-only VLM decoder saves reloading with random weights (#12561)
f934297 Stefan Wang committed 1 month ago