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 diffusion: skip the offload cost probe's pinned buffer on an integrated GPU (#15178)
9f499ff Daniel Han committed 27 days ago
Pipeline parallelism v1: apply_pipeline_parallel for dense decoders (GPipe, 1F1B) (#15175)
f6e690e Daniel Han committed 27 days ago
Bump dompurify from 3.4.16 to 3.4.17 in /docker/jupyter/unsloth_labext in the npm_and_yarn group across 1 directory (#14932)
ab2a618 dependabot[bot] committed 27 days ago
Find the Run settings footer by its rule, not by div count (#15181)
3bac26d Daniel Han committed 27 days ago
Studio: Sampling footer waits for the saved-sampling read before Remember and Save (#15177)
984d6e5 Michael Han committed 27 days ago
Bump vite, @modelcontextprotocol/sdk and dompurify in studio/frontend (#14442)
f80130e dependabot[bot] committed 27 days ago
Bump the cargo-tauri group across 1 directory with 2 updates (#14503)
09f9fb6 dependabot[bot] committed 27 days ago
Bump setuptools from 84.0.0 to 84.1.0 in the python-security group across 1 directory (#15051)
6bc9165 dependabot[bot] committed 27 days ago
Studio video: stream the Wan encode under model / sequential offload too (#15173)
43b83d5 Daniel Han committed 27 days ago
Studio: API page fonts, Sampling footer in chat settings (#15111)
87a7605 Michael Han committed 27 days ago
Studio: restyle Library and System storage sections as a usage overview (#15083)
46d43ef Michael Han committed 27 days ago
Studio: pinned Run settings footer, model picker on the radius scale (#15104)
b143b37 Michael Han committed 27 days ago
Studio: one audio icon everywhere, ai-speech for Text to Speech (#15112)
cd44b60 Michael Han committed 27 days ago
Studio video: cache Wan text encodes per text so an unchanged negative prompt is encoded once (#15166)
a218851 Daniel Han committed 27 days ago
Decode: write K/V straight into the cache slot, and opt-in CUDA-graph replay of the decode step (#15171)
aae9f54 Daniel Han committed 27 days ago
Block compile: per-region recompile limit with isolate_recompiles on torch 2.13+ (#15157)
5b7953b Daniel Han committed 27 days ago
Studio diffusion: choose whole-module vs streamed offload per request from its step count (#15163)
90ccf12 Daniel Han committed 27 days ago
Lift for_inference's tensor parallel helpers in the return-logits test (#15162)
38f52c6 Daniel Han committed 27 days ago
Clear the compile kill switch in the block CUDA graphs tests (#15167)
007a846 Daniel Han committed 27 days ago
Keep test_studio_root_resilience on the backend's utils and loggers (#15165)
1128468 Daniel Han committed 27 days ago
FSDP2: free-memory fit check, MFU without recompute (HFU separately), nccl2 warning on small GPUs (#15158)
deed5b9 Daniel Han committed 27 days ago
DDP: find unused params for a fully finetuned VLM on text-only batches; native FSDP2 names expert parallelism (#15156)
0ef5f16 Daniel Han committed 27 days ago
Studio video: cache the offloaded native int8 DiT per expert (Wan2.2-A14B warm load 210 s to 21 s) (#15152)
f9c7416 Daniel Han committed 27 days ago
Opt-in CUDA graphs for compiled blocks under gradient checkpointing (UNSLOTH_BLOCK_CUDAGRAPHS=1) (#15154)
8feaae2 Daniel Han committed 27 days ago
Padding-free: call flash varlen through a plain autograd.Function in eager (#15160)
71f705b Daniel Han committed 27 days ago
Studio diffusion: seed the dense bf16 placement from FLUX.1's measured peak (#15149)
68c1c15 Daniel Han committed 27 days ago
Studio video: keep Wan2.1-T2V-1.3B in fp16 on fp16-only GPUs (#15147)
47b4c3e Daniel Han committed 27 days ago
Tensor parallelism v1: apply_tensor_parallel for dense decoders (LoRA / QLoRA) (#15137)
d2ca651 Daniel Han committed 27 days ago
Studio diffusion: move the deferred compile off the render for offloaded and Qwen-Image-2.1 denoisers (#15150)
c2c7d34 Daniel Han committed 27 days ago