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
Finish a 16bit load of a static per-tensor fp8 checkpoint (Mistral-Small-4) (#11531)
bda4023 Daniel Han committed 1 month ago
Load and train remote-code multimodal wrappers: Phi-4-reasoning-vision in 4-bit and Nemotron-3-Nano-Omni (#11526)
b53a6d7 Daniel Han committed 1 month ago
Studio: tile the VAE instead of refusing oversized upscales, and in-app remedies for image refusals (#11798)
ac1c4da Daniel Han committed 1 month ago
Studio: keep bf16 for auto precision on image families that cannot compile (#11819)
e5082af Daniel Han committed 1 month ago
Studio: apply remembered settings for MLX and safetensors models to API loads (#11904)
54fbe52 Long Yixing committed 1 month ago
Studio: run an explicit int8 under offload on NVIDIA as torchao-free W8A8 (#11712)
de7c4fd Daniel Han committed 1 month ago
Studio: count the hosted text encoder and keep the GGUF denoiser resident when only the encoder does not fit (#11802)
ea11ce2 Daniel Han committed 1 month ago
Unsloth Studio (AMD): allow INT8 / FP8 image and video precision without torchao (#11631)
f0b73b2 Daniel Han committed 1 month ago
fix(studio): classify MLX requests from the applied chat template override (#11903)
4cfeec9 Long Yixing committed 1 month ago
Studio: cache text-encoder outputs so repeat prompts skip the encoder (#11844)
0df4178 Daniel Han committed 1 month ago
Studio: auto step cache (FBCache) only on the max speed tier (#11791)
8a24eb0 Daniel Han committed 1 month ago
fix(grpo): use the evaluated model's output head for log-probs (#11877)
c0f0c19 cherry77-cloud committed 1 month ago
Studio: faster MiniMax-H3 video VAE encode and decode (#11801)
14354db Daniel Han committed 1 month ago
Studio: release VRAM when an image model is unloaded mid-render (#11790)
676a801 Daniel Han committed 1 month ago
Studio: HunyuanVideo-1.5 padded-text trim on the max speed tier (#11824)
e874b93 Daniel Han committed 1 month ago
Point -bf16 4-bit requests at quantization_config, and warn when a bitsandbytes load quantized nothing (#11586)
999e964 Daniel Han committed 1 month ago
Studio: count rendered GGUF prompts before shared-KV admission (#10794)
3480890 Lee Jackson committed 1 month ago
Studio: pin the kept offload weights on first onload when host RAM allows (#11766)
b0af0ea Daniel Han committed 1 month ago
Studio: whole-model offload points weights back at their host tensors instead of copying them (#11764)
28d5e76 Daniel Han committed 1 month ago
Studio: compile the diffusion denoiser on fp16 GPUs when a speed tier is picked (T4 and other pre-Ampere cards) (#11899)
ae65563 Daniel Han committed 1 month ago
Load 4.x remote code and config-only remote code on transformers 5 (Trinity-Large, MiniMax-M3) (#11658)
5e39eea Daniel Han committed 1 month ago
Pin the ARM64 Arrow overlay to a commit before its wheels are signed (#11911)
4a16bff Daniel Han committed 1 month ago
fix(studio): show compaction notices for tool-loop checkpoints (#11702)
44ac9b5 Lee Jackson committed 1 month ago
Studio: video status reports CUDA graphs off when they never engage (#11886)
4088130 Daniel Han committed 1 month ago
Dequantize FP8 weights left raw by text_only and offloaded loads (#11841)
a20dcb2 Daniel Han committed 1 month ago
Studio: stop a compiled Qwen-Image-2.1 render holding 2 GiB of prefix K/V it does not need (#11882)
fbf50af Daniel Han committed 1 month ago
Studio: carve the auto eval split for vision datasets held as lists (#11851)
f9d7561 Nilay committed 1 month ago
Walk every sub-config when deciding whether a config carries remote code (#11555)
42a23e8 Daniel Han committed 1 month ago