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
Frontend CI: back off for minutes before giving up on the Bun download (#17221)
a72288a Daniel Han committed 11 days ago
Studio MiniMax-H3 Sol VAE: free expired Q/K/V slots before building the next, at most two live (#17085)
bb83800 Daniel Han committed 11 days ago
Studio diffusion: read the XPU's float64 support from the device (Arc A-series LTX RoPE) (#17070)
93f1913 Daniel Han committed 11 days ago
Studio MiniMax-H3: run the text stream once per render, pack the sequence in place, look up the streamed conditioner's vocabulary rows on the host (#17203)
cf47cf5 Daniel Han committed 11 days ago
Backend CI: back off for about five minutes before giving up on the ruff-unsloth wheel (#17211)
1714b16 Daniel Han committed 11 days ago
Studio: read the hosted LTX-2.5 int8 checkpoint when the load's own plan offloads the denoiser (#17143)
5cafbf2 Daniel Han committed 11 days ago
Backend CI: split the g-r pytest shard into g-l and m-r (#17187)
20cbf1b Daniel Han committed 11 days ago
Studio: no header strips on macOS glass or over a plain background image (#17154)
3b7fe25 Michael Han committed 11 days ago
Studio diffusion: reload a released text encoder from its own safetensors files, no snapshot write (#17184)
f53feca Daniel Han committed 11 days ago
Context parallel: patch DPO / GRPO even when context_parallel is imported before the models (#17185)
dcb1eb1 Daniel Han committed 11 days ago
Studio tests: the two-card compiled W8A8 test judges each card on its own arch (follow-up to #17078) (#17195)
edc48a6 Daniel Han committed 11 days ago
Studio tests: make the compiled-wrapper onload test hold its own weakref (follow-up to #17141) (#17189)
2dc6977 Daniel Han committed 11 days ago
Studio video: log an automatic text-encoder pick after the load retries its hosted encoder (#17093)
e56e1e4 Daniel Han committed 11 days ago
Context parallel: Ulysses evidence for gpt-oss (attention sinks, MoE, aux off) (#17183)
400288d Daniel Han committed 11 days ago
Studio llama.cpp: the token-count memo renders again on a new day (#17090)
f54a471 Daniel Han committed 11 days ago
Studio: install FlashInfer's prebuilt kernels into the vLLM engine env (#17063)
bdebdc8 Daniel Han committed 11 days ago
Context parallel: Ulysses for Gemma 4 text models (#15883)
3329d6d Immanuel Sun committed 11 days ago
Studio: fp8 denoisers on sm_89 no longer hang on the first render (CUDA graph replay under torch < 2.12) (#17166)
2092c03 Daniel Han committed 11 days ago
Studio Wan fused blocks: one Triton binary per kernel for every sequence length (#17101)
29b1bed Daniel Han committed 11 days ago
Studio Qwen-Image-2.1: encode prompts without the text encoder's unused K/V cache (#17118)
bc8d970 Daniel Han committed 11 days ago
Studio: run INT8 on T4 and other Turing GPUs instead of refusing it (#17167)
3f5d012 Daniel Han committed 11 days ago
FSDP2 save: compose the writer across equal meshes, not only identical ones (#17158)
897c742 Daniel Han committed 11 days ago
Studio FLUX.3: keep bf16 weights as stored on fp16-only GPUs and compute in float32 per layer (#17173)
f81dad0 Daniel Han committed 11 days ago
Studio int8 GEMM: the unlisted-card race switch acts between loads (#17084)
8b11dab Daniel Han committed 11 days ago
Studio int8 GEMV: wide scale groups tile inside shared memory, a grouped miss no longer demotes per-row Linears (#17106)
9d8cd39 Daniel Han committed 11 days ago
Studio Sol-H3 VAE: decide window masks from the geometry, build each RoPE table once per decode (#17100)
ac8be67 Daniel Han committed 11 days ago
Studio MiniMax-H3 VAE: stitch batched decoder tiles into the frame as they arrive (#17097)
162990f Daniel Han committed 11 days ago
Studio GGUF: cast K-quant and legacy weights to bf16 in their own buffer (#17080)
46c0c8f Daniel Han committed 11 days ago
Studio upscaler: normalise seams and finalise uint8 in place on the host result (#17083)
0c6edee Daniel Han committed 11 days ago
Studio NVFP4: the protected branch compiles once from a cold decode table and accepts fp16 inputs (#17081)
461ba86 Daniel Han committed 11 days ago