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: batch the raw text blank row filter and EOS append for map-style datasets (#15978)
5ac8384 Daniel Han committed 22 days ago
Studio: stop skipping scheduled runs right after Studio starts (#16037)
c8f38af Nilay committed 22 days ago
Studio chat: publish held stream text on the next frame instead of waiting for the next chunk (#15998)
35310d5 Daniel Han committed 22 days ago
Studio: strip only harmony protocol tokens from gpt-oss answers (#16004)
21904ce Daniel Han committed 22 days ago
Studio: include Schema transform output in recipe dataset downloads (#16039)
a90de04 Nilay committed 22 days ago
Studio: add Spanish (Latin America) as a display language (#16089)
ee6e7b7 Michael Han committed 22 days ago
Desktop: run the AppImage on the host's WebKitGTK 2.54+ when available (#15503)
05ede7c Etherl committed 22 days ago
Studio: let conversation search return every earlier turn that fits (#16036)
af3712b Nilay committed 22 days ago
Studio: no text selection from the clicks that follow a dropdown dismissal (#16077)
e07e636 Michael Han committed 22 days ago
Studio: fill locale gaps and translate the sign-in, consent and Desktop startup screens (#16069)
8cb9120 Michael Han committed 22 days ago
Tests: stub @/lib/folder-icons in the Hub GGUF card Metal budget test (#16082)
cab10d4 Daniel Han committed 22 days ago
Studio: align the Images and Video preset dropdown with the title's top (#16081)
c6056ff Michael Han committed 22 days ago
Studio: name a too-short context length when completion masking drops rows (#16000)
bebfa2b Daniel Han committed 22 days ago
fix(studio): list approved custom-code models in /v1/models (#15945)
d0c962d Long Yixing committed 22 days ago
Studio: scale the token count timeout so long chats can compact (#15900)
60dc72a oobabooga committed 22 days ago
float16: agree on GradScaler skips across pipeline stages, scale the CP start, probe the float16 FP8 GEMM (#16070)
aa3dba5 Daniel Han committed 22 days ago
Studio: stop the chat composer draft mirroring into other open tabs (#15795)
c99ccea Debojyoti Banerjee committed 22 days ago
Studio: keep the composer still when switching chats in WebKit (#16076)
0d5d40c Michael Han committed 22 days ago
Context parallel: allow text-only Qwen3.5 checkpoints that carry a vision_config (#16072)
c41ed6a Daniel Han committed 22 days ago
Studio video: batch the CFG pair into one call on a compiled resident Wan denoiser (#16067)
856799b Daniel Han committed 22 days ago
Context parallelism: refuse it beside tensor parallelism instead of training the wrong loss (#16068)
4089c57 Daniel Han committed 22 days ago
FP8 scaled_mm: keep native fp16 output on torch >= 2.14 (#16063)
9a1ea09 Daniel Han committed 22 days ago
MXFP4 torch decode fallback: run ldexp on the operands' GPU (#16062)
eb6f104 Daniel Han committed 22 days ago
gpt-oss: accept an absolute transformers_weights inside the local model dir (#16061)
d4ebf72 Daniel Han committed 22 days ago
Scope WavLM's 4-bit attention skip to WavLM's own layers (#16060)
f842c04 Daniel Han committed 22 days ago
Cross entropy: trace under torch.compile(dynamic=True) (#16059)
bdea0c0 Daniel Han committed 22 days ago
Packed mRoPE flash attention: cover a padded tail, same decision when compiled (#16058)
85176e5 Daniel Han committed 22 days ago
GRPO: keep masking completions cut by the vLLM context limit (#16057)
0f289fd Daniel Han committed 22 days ago
Desktop: Add GPU rendering setting with a Linux launch-crash fallback (#15482)
7af6ba3 Etherl committed 22 days ago