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
Route GKD distillation through the chunked generalized JSD (#12136)
41b2ff6 Daniel Han committed 1 month ago
Repair chat templates that always append the generation prompt, also on FastModel loads (#12139)
e417798 Daniel Han committed 1 month ago
Narrow Kimi-K3's zero-padded KDA A_log to num_heads so the fla backward runs (#12132)
4e6f0a4 Daniel Han committed 1 month ago
Load speech-to-text models (Voxtral, Qwen2-Audio) through FastModel (#12127)
e3e0a21 Daniel Han committed 1 month ago
Keep Linear layers an FP8 checkpoint stores in bf16 unconverted (#12124)
b779aca Daniel Han committed 1 month ago
Keep flash attention off towers the Auto classes do not register (#12129)
4780f63 Daniel Han committed 1 month ago
Rebuild CohereTokenizer from tokenizer.json when transformers v5 changes its ids (#12130)
12b6dbf Daniel Han committed 1 month ago
Dequantize block-FP8 weights with a ragged last block on 16-bit loads (GLM-5.3) (#12133)
a9ed05a Daniel Han committed 1 month ago
Add a CI gate for code shapes that heuristic antivirus scanners quarantine (#12178)
b91172e Daniel Han committed 1 month ago
Studio: never ask for approval to run search_conversation (#12175)
269b67b Daniel Han committed 1 month ago
Studio: keep conversation recall working across repeated compactions (#12174)
6beb057 Daniel Han committed 1 month ago
Turn off the MoE aux loss for dense models that carry a router config (#12076)
8af84f1 Daniel Han committed 1 month ago
Studio: list at most the 12 most recently active projects and sections in Move to (#12155)
3bf6002 Michael Han committed 1 month ago
Refuse K-EXAONE 2.0 on a transformers that ignores its config (#12128)
0309133 Daniel Han committed 1 month ago
Build the native image processor at defaults when a VLM repo has no preprocessor_config.json (#12125)
28c117a Daniel Han committed 1 month ago
Let a loaded Kimi K2.5 / K2.7 processor take processor(text=..., images=...) (#12126)
c94ae3c Daniel Han committed 1 month ago
Studio: LTX-2.3 about 4.5x faster per clip (unguided distilled sampling, compile fixes, hosted FP8) (#12067)
3d45be3 Daniel Han committed 1 month ago
Phi-4-reasoning-vision: give remote multimodal prep an indexable cache view on transformers 5 (#12123)
3040eca Daniel Han committed 1 month ago
Keep Nemotron-H mixer.out_proj unquantized under a caller's BitsAndBytesConfig (#12131)
876faf0 Daniel Han committed 1 month ago
Attention resolver: respect a declared _supports_sdpa = False, and skip flash_attention_2 when a class's compatible flash kernels exclude it (MiMo-V2-Flash) (#12147)
c86708f Daniel Han committed 1 month ago
SAC probe: verify Studio identity before sending a password (#12176)
741ea5c Daniel Han committed 1 month ago
unsloth start opencode: size the output limit to the context and add --max-tokens (#12111)
07f004d Michael Han committed 1 month ago
Studio: round hover for the Settings close button (#12172)
3a3cca5 Michael Han committed 1 month ago
Tests: stop Windows tests tripping Bitdefender and the 16-bit application dialog on real machines (#12167)
be9e6af Daniel Han committed 1 month ago
Studio: ignore an SSLKEYLOGFILE the process cannot write instead of failing every HTTPS client (#12166)
f9150d2 Daniel Han committed 1 month ago
Studio: Select all only picks the models the search shows (#11485)
87ff26a Nilay committed 1 month ago
Studio: reset the download progress bar when a retry restarts the file (#11593)
8fb2fa3 Nilay committed 1 month ago
Fix training with accelerate 1.15 on torch without a distributed backend (AMD Windows ROCm) (#12162)
eaab494 Daniel Han committed 1 month ago