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
CPU backend: context parallelism engages (or refuses) instead of silently running DDP; CP modules import (#16802)
a215ec7 Daniel Han committed 14 days ago
FP8 LoRA test: expect the kernel route when unsloth_zoo drops HF output flags (#16855)
d36781d Daniel Han committed 14 days ago
Studio tests: check family picker labels in backend CI, where families are added (#16856)
4037c03 Daniel Han committed 15 days ago
Parallel capabilities: follow the measured composition evidence (packed CP, Unsloth checkpoint, tiling, block swap) (#16773)
30949ea Daniel Han committed 15 days ago
Studio: label the Chroma and Anima families in the family picker (#16809)
d5745d0 Daniel Han committed 15 days ago
Studio video: decode the VAE untiled on CPU hosts when it fits in RAM (Wan2.2 VAE decode 1.5x) (#16766)
73984c4 Daniel Han committed 15 days ago
Studio video: opt-in Sol-H3-VAE decoder for MiniMax-H3 (#16747)
9541cd2 Daniel Han committed 15 days ago
Studio: install flash-attn's Triton AMD backend on ROCm RDNA3 / RDNA4 (10-15% faster diffusion steps on gfx1151) (#16767)
f0b5800 Daniel Han committed 15 days ago
Studio diffusion: compiled int8 W8A8 linear on RDNA3 / RDNA4 ROCm (Qwen-Image-2.1 int8 21% faster per step on gfx1151, now ahead of bf16) (#16825)
22233cd Daniel Han committed 15 days ago
Studio: int8 activations on T4 for Qwen-Image-Edit (2.03x per step) (#16813)
7b7ebff Daniel Han committed 15 days ago
Studio CPU diffusion: one thread per logical CPU on small SMT hosts, for sd.cpp and torch (1.03x to 1.26x) (#16774)
4db7518 Daniel Han committed 15 days ago
Studio sd.cpp: --mmap for CPU renders on builds whose --mmap keeps the CPU repack (#16758)
c014c8b Daniel Han committed 15 days ago
Studio diffusion: fuse the SwiGLU product into the down projection's act quant on untuned sm80+ archs (#16785)
268f414 Daniel Han committed 15 days ago
Studio: apply ComfyUI LoRA and LoKR files to Anima, Krea 2 and MOSS-Image (#16711)
083ba2f oobabooga committed 15 days ago
Diffusion autoopt parity: add tf32 to the Anima and Chroma default-tier rows (#16838)
482a764 Daniel Han committed 15 days ago
Studio Images: pin the parked SAM 3 copy with registered host memory by default (unpark 0.36 -> 0.14 s) (#16791)
0ab3f07 Daniel Han committed 15 days ago
Studio video: keep a streamed LTX-2 denoiser partly resident per request (21% faster per step) (#16787)
c6f4f78 Daniel Han committed 15 days ago
Studio multi-GPU: split an even Lumina 2 batch across two cards (1.7-2.1x, fp-order only) (#16786)
1e50add Daniel Han committed 15 days ago
Studio video: MiniMax-H3 int8 video decoder on by default (30-33% faster decode, LPIPS 0.0004-0.0013) (#16764)
ace7f0b Daniel Han committed 15 days ago
Studio diffusion: TF32 matmuls on the default speed tier, not only max (#16762)
7d0c095 Daniel Han committed 15 days ago
Studio diffusion: fused FLUX AdaLN on by default on NVIDIA Linux (2.4% per eager step, fp32 error unchanged) (#16760)
8033ef3 Daniel Han committed 15 days ago
Studio: move MLX to mlx-vlm 0.7.12 and serve mlx-vlm-only text families (#16765)
bfdd8fa Long Yixing committed 15 days ago
Avoid duplicate successful sandbox probes during refresh (#16735)
280de8a Nathan Tran committed 15 days ago
Studio video: denormalize decoded frames slice by slice in the uint8 export (#16748)
3fc6b69 Daniel Han committed 15 days ago
Studio diffusion: stream the LTX-2.5 encoder, cache and trim FLUX.3 prompt encodes (#16750)
89db0db Daniel Han committed 15 days ago
Studio: reuse an identical llama-server prompt count within a tool round (#16756)
732c95b Daniel Han committed 15 days ago
Fused LoRA backward: skip the gradient GEMMs of a frozen A or B (#16752)
16060ab Daniel Han committed 15 days ago
Studio: fit Auto context to system RAM on CPU-only machines (#16689)
80cf144 Nilay committed 15 days ago