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
Honour YaRN passed as rope_parameters, and rope_scaling on FastModel (#15214)
801ba6d Daniel Han committed 26 days ago
Studio diffusion: read hosted files we already ship (LTX-2 fp8 DiT, Qwen-Image-2.1 bf16 VAE), and name a missing int8 text encoder file (#15318)
b75d92c Daniel Han committed 26 days ago
Studio: keep a system column when the advisor mapping carries metadata (#14999)
ecf0054 Vineeth Sai Varikuntla committed 26 days ago
Studio: reject a training random_seed outside the numpy range up front (#15176)
33114da Vineeth Sai Varikuntla committed 26 days ago
TP / PP: every causal LM architecture trains like one process or is refused up front (#15230)
4f18eb4 Daniel Han committed 26 days ago
Studio diffusion: cast cache for fp32-shard components at bf16 too, and on the image path (Z-Image-Turbo reads 11.5 GiB, not 22.9) (#15321)
cad94d3 Daniel Han committed 26 days ago
Studio Images: one denoiser stream-out per prompt encode, and the Z-Image CFG negative encoded once (#15310)
709cdc3 Daniel Han committed 26 days ago
Studio: bring back the scheduled task model picker and model loading (#15255)
136074f Daniel Han committed 26 days ago
GRPO kept-set refusals: stop saying UNSLOTH_GRPO_TRUNCATED_SUPPORT=1 was set (#15307)
fbc905f Daniel Han committed 26 days ago
Studio: reject a NaN weight_decay in a training request (#14721)
01dcfef Vineeth Sai Varikuntla committed 26 days ago
Studio: answer 400 to out-of-range sampling on /v1/messages with a Connection model (#15282)
631e562 Vineeth Sai Varikuntla committed 26 days ago
Keep Gemma embedding normalizer out of place for external inputs_embeds (#15308)
553c7e4 Daniel Han committed 26 days ago
Read CSV text cells larger than 128 KB in RawTextDataLoader (#14380)
1e1f211 Vineeth Sai Varikuntla committed 26 days ago
Fix gradients through external inputs_embeds when applying attention masks (#15234)
5efeeb7 cherry77-cloud committed 26 days ago
Model picker contracts: read the multi-quant parent row's gutter through its aliases (#15319)
df02bf1 Daniel Han committed 26 days ago
Fix non-contiguous position indices in fused and unfused RoPE (#15235)
6c3dcdd cherry77-cloud committed 26 days ago
Studio: show settings and the row menu on the loaded model in Recommended (#15249)
c9e5d4a Michael Han committed 26 days ago
Studio: one run settings dropdown per model, and the last setting used is what loads (#15247)
c9ddc5a Michael Han committed 26 days ago
Studio: cache fp32-shard video components cast to float16 for later loads (#15303)
d4be247 Daniel Han committed 26 days ago
Studio: show newly downloaded quants of a loaded GGUF in the model picker (#15233)
2563513 Michael Han committed 26 days ago
Studio: control heights, composer fades, API page layout, settings polish (#15232)
61cfbaa Michael Han committed 26 days ago
Native FSDP2: keep gathered weights allocated so block CUDA graphs replay (#15215)
1185f84 Daniel Han committed 26 days ago
Studio GGUF: charge images in the fit and tool-result sizing counts (#14968) (#15014)
8b3f6fe Debojyoti Banerjee committed 26 days ago
Fused preference loss tests: build the CPU float64 model from transformers' own Llama (#15301)
7132f8a Daniel Han committed 26 days ago
Studio video: pin MiniMax-H3's lazily streamed blocks with the background pinner (#15297)
afcf0fc Daniel Han committed 26 days ago
Studio: let FLUX.3 Klein edit through the Edit workflow (#15263)
277bd03 Daniel Han committed 26 days ago