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: save sampling defaults per model and use them for API requests (#14515)
3ea0140 Daniel Han committed 29 days ago
Studio: model picker hides empty uploads and fades models this device cannot run (#14935)
9f59a73 Michael Han committed 29 days ago
FP8 block dequant on gfx11 ROCm: ragged / rectangular e4m3fn blocks take the byte decoder (#14936)
ca4de9f Daniel Han committed 29 days ago
Studio: gate saved Connection keys on a stamped key purpose, not the key name (#14824)
19ca9c3 Daniel Han committed 29 days ago
Keep the checkpoint_skip_layers="auto" planner out of torch.compile regions (#14928)
969bbfc Daniel Han committed 29 days ago
Native FSDP2: keep accelerate FSDP2 for custom Accelerator loops, support unsharded generation on torch 2.6 / 2.7 (#14882)
482058c Daniel Han committed 29 days ago
GRPO: sync LoRA weights to TRL's own vLLM engine on TRL < 0.28, refuse quantized bases (#14831)
4dad471 Daniel Han committed 29 days ago
Large head attention: keep CUDA graph capture off the mask scan and bound int32 row offsets (#14790)
e79ad3e Daniel Han committed 29 days ago
Studio: remove the New badge from the Browser settings tab (#14939)
0ee5cfc Michael Han committed 29 days ago
Studio: list imatrix-weighted quants named <model>-<QUANT>-imatrix.gguf (#14931)
43eca72 Daniel Han committed 29 days ago
Studio Images: keep v-prediction detection on cache-only SDXL loads by remembering the online markers per snapshot (#14873)
d436a92 Daniel Han committed 29 days ago
Studio: tell websites the macOS browser panel is Safari (#14797)
96ccc9a Nilay committed 29 days ago
Installers: pin uv 0.13.5 and replace uv older than 0.13.4 (emSign root) (#14894)
d4f1ceb Daniel Han committed 29 days ago
Format four files merged without the repo formatter in #14913 and #14823 (#14930)
cf9438c Daniel Han committed 29 days ago
Studio: steadier run settings controls, grey sliders, 18px thumbs (#14926)
af623ab Michael Han committed 29 days ago
FSDP2 QLoRA merged saves: stage the packed 4-bit base on the CPU and merge one layer at a time (#14823)
bb166e3 Daniel Han committed 29 days ago
Studio Images: Compare grid counts Hires CFG when sending the negative prompt and records Hires settings (#14875)
63ac122 Daniel Han committed 29 days ago
Studio diffusion: seed FLUX.2 klein residency from its measured peak on the first load (#14866)
a4e25b4 Daniel Han committed 29 days ago
Studio video: fit per-request residency on the streaming tier and for HunyuanVideo-1.5 (#14865)
9a07082 Daniel Han committed 29 days ago
FP8 per-tensor / rowwise dequant under torch.compile below sm89: one opaque byte-decode op (#14913)
da109fd Daniel Han committed 29 days ago
Fetch get_statistics repos concurrently so from_pretrained does not wait on four HF round trips (#14476)
2eb94bc Daniel Han committed 29 days ago
Studio ROCm: keep stored flash-attention verdicts from recompiling per bucket, and gate the CFG replica too (#14869)
1f55767 Daniel Han committed 29 days ago
Format two files merged without the repo formatter in #14769 (#14908)
08a1e39 Daniel Han committed 29 days ago
Security audit: baseline the pandas 3.1.0 find_stack_level test (#14870)
e5201c3 Daniel Han committed 29 days ago
Tests: rewrite the remote_images import in the no-torch format_conversion harness (#14903)
4188371 Daniel Han committed 29 days ago
Studio Images: run ESRGAN upscalers through the spandrel descriptor (#14876)
0d28035 Daniel Han committed 29 days ago
Studio Images: Hires fix latent upscale interpolates the base latents directly, as ComfyUI's LatentUpscale (#14863)
4afc4f0 Daniel Han committed 29 days ago