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: route untuned archs (sm86 / 90 / 100 / 121) onto the factored ConvRot kernel behind a probe and latency check (#14988)
1d9c522 Daniel Han committed 29 days ago
MLX backend tests: stub zoo's new fused_drafter_ops (#15020)
5ec3ede Daniel Han committed 29 days ago
Name the encoding on the trainer.py read in test_pretokenized_dataset (#15018)
245c2b1 Daniel Han committed 29 days ago
Studio sd.cpp: pass --conv-direct off when Studio's policy keeps direct convolution off (#14859)
5dd5154 Daniel Han committed 29 days ago
Check the GGUF load marker on load_model through the AST (#15017)
8373f09 Daniel Han committed 29 days ago
Studio MLX images: a LoRA load keeps sd.cpp or diffusers instead of routing into the MLX refusal (#14874)
8ef8dfe Daniel Han committed 29 days ago
Studio: run the Qwen-Image-2.1 CFG branch on the replica with its own prefix KV cache (#14954)
036bf6a Daniel Han committed 29 days ago
Studio sd.cpp: route all-Q8_0 image GGUFs to BF16 cuBLAS, keep k-quants on MMQ (#14983)
0a34710 Daniel Han committed 29 days ago
Studio diffusion: learned peaks per render phase, so later loads keep text encoders resident (#14979)
93fc9c7 Daniel Han committed 29 days ago
Agent guides CI: resolve the newest Node 24 for the openclaw legs (#15013)
3ab2c1e Daniel Han committed 29 days ago
Pin CUDA availability in the step graph capture-stream release test (#15015)
f2ecfe1 Daniel Han committed 29 days ago
Studio diffusion: cut the int8 ConvRot overhead (one rotation per shared input, SwiGLU fused into the down quant, factored rotation kernel) (#14982)
6143055 Daniel Han committed 29 days ago
Studio diffusion: block-offload re-pins land at the addresses the step graph recorded (#14765)
0291ac9 Daniel Han committed 29 days ago
Studio diffusion: cap the prefetcher's pinned staging so guided Z-Image-Turbo renders on a 31 GB host (Kaggle T4x2) (#14975)
8dff63c Daniel Han committed 29 days ago
Studio: import and export skills as a ZIP from the Skills dialog (#14181)
d6604ac Samih Qureshi committed 29 days ago
Studio: truncate completion_mask / assistant_masks with pre-tokenized input_ids (#14884)
93c5b32 Daniel Han committed 29 days ago
Studio video: MiniMax-H3 fused VAE decoder follows the mixed-precision load, and encode overlaps decode (#14986)
7f32241 Daniel Han committed 29 days ago
Studio diffusion: fp32 VAE upcasts keep the offloaded VAE's pinned host weights (SDXL-class force_upcast VAEs) (#14946)
30e7688 Daniel Han committed 29 days ago
Studio diffusion: fixed 2 GiB discrete-card reserve (ComfyUI-style fixed reserve plus activation estimate) (#14951)
abe78a0 Daniel Han committed 29 days ago
get_peft_model: pass a scalar layers_to_transform to PEFT as a list (#14910)
4de2f1e Daniel Han committed 29 days ago
FP8 block dequant on ROCm: round fp32 to fp16 with integer ops (#14992)
3922fa9 Daniel Han committed 29 days ago
Studio video: keep per-request residency fitting after the stream step graph is dropped (Wan2.2-TI2V-5B at 8 / 12 GB) (#14937)
598ea6a Daniel Han committed 29 days ago
Studio diffusion: measured activation rows for MOSS-Image and FLUX.3 Klein base; layer split prices Z-Image from its measured row (#14950)
bef946c Daniel Han committed 29 days ago
Studio: use the MPS SDPA prefill kernel per call on torch 2.13+ (#14941)
60c7715 Long Yixing committed 29 days ago
Fix test_every_resolver_probe_forwards_the_trust_decision on main (#14990)
042a717 Daniel Han committed 29 days ago
MTP: keep the trained head when saving a named PEFT adapter (#14912)
e49be2a Daniel Han committed 29 days ago