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
Gate XPU bf16 on native support, hint CUDA_DEVICE_ORDER on mixed NVIDIA hosts, read the Metal GPU family for MLX bf16 (#14371)
e172530 Daniel Han committed 1 month ago
llama.cpp prebuilt: accept upstream win-rocm bundles, run native arm64 under Rosetta, source-build on Jetson (#14373)
3594ed5 Daniel Han committed 1 month ago
Studio/CLI: report Rosetta, skip audio.cpp CUDA on pre-Turing GPUs, pass doctor on MPS/XPU, hide XPU from token preview (#14374)
0c817e5 Daniel Han committed 1 month ago
Studio: Load anyway for the unified-memory diffusion refusal, and a capacity fix for its pre-download check (#14379)
34c5203 Daniel Han committed 1 month ago
Studio diffusion: record a prompt length's CUDA graph on the first render only (#14387)
c6d77f2 Daniel Han committed 1 month ago
NF4 decode on T4: use the bitsandbytes 4-bit GEMM above the GEMV row cap (#14386)
9731b3a Daniel Han committed 1 month ago
Studio: pin the FLUX.3 prompt upsampler to the image model's card under HIP masks and on Vulkan builds (#14356)
f55a9d3 Daniel Han committed 1 month ago
Studio: run Ada FP8 precise accumulate through a unit-scale GEMM plus a fused scale epilogue (#13793)
edb8e67 Daniel Han committed 1 month ago
Fix phone-deployment QAT for Gemma3 and LoRA, and write the matching ExecuTorch export config (#14376)
7c0e7d2 Daniel Han committed 1 month ago
fix: release the device cache through the active accelerator on the vLLM resume/teardown paths (#13438)
2371163 li-lizhe committed 1 month ago
Tests: keep gloo FSDP2 test ranks on the CPU on GPU hosts (#14383)
17b4b59 Daniel Han committed 1 month ago
Stop fix_executorch from hiding an installed torchtune (#14375)
4ff83b5 Daniel Han committed 1 month ago
Installers: do not install generic aarch64 CUDA torch on Jetson before CUDA 13 (#14363)
f1f283a Daniel Han committed 1 month ago
Studio diffusion: run FLUX.3 base conditioning blocks once per render (#14367)
4e0fe5b Daniel Han committed 1 month ago
MoE grouped GEMM: require sm90 on every visible card for TMA and launch on the operands' device (#14358)
9b5e9b9 Daniel Han committed 1 month ago
Studio sd.cpp: explicit VRAM budget for MiniMax-H3 and a CUDA capability fallback (#13714)
9322afb Daniel Han committed 1 month ago
save_pretrained_gguf: stop rewriting a reused local checkpoint's tokenizer with a BOS-stripped template (#14095)
5aa5f7b Leo Borcherding committed 1 month ago
check_tokenizer: repair out-of-vocab added tokens on transformers 5.x, and raise when one can't be repaired (#14096)
7341003 Leo Borcherding committed 1 month ago
load_in_fp8=True quantizes a model with no FP8 repo instead of silently loading it in bf16 (#14093)
96f065b Leo Borcherding committed 1 month ago
Studio diffusion: cache runtime-quantized denoisers as local pre-quant checkpoints (#13715)
a6c1e06 Daniel Han committed 1 month ago
fix: preserve sentence encoder adapters and export dense 16bit weights (#14246)
1917996 cherry77-cloud committed 1 month ago
Studio compile cache: keep every graph across re-saves, and save under torch's compile lock (#14341)
1a6397c Daniel Han committed 1 month ago
Unsloth Studio (AMD/Windows): show an iGPU's own memory on Settings > About, not the system RAM it can borrow (#14317)
f7b147d Leo Borcherding committed 1 month ago
Keep SFT kwargs passed beside a plain TrainingArguments instead of falling back to trl's defaults (#14094)
ff9c958 Leo Borcherding committed 1 month ago
Studio: Mamba-2 hybrids use the extension-free mamba-ssm wheel when unsloth-zoo runs their conv on Triton (#14160)
0d7e6a4 Daniel Han committed 1 month ago
Studio diffusion: single-card CFG batching for a DiT streamed from host memory (#13787)
d47cddd Daniel Han committed 1 month ago
Studio: prune dead startup records and preserve process identity (#14301)
abfa6c7 oobabooga committed 1 month ago
Studio: size sd.cpp's thread pool from the Apple fast cores (#13349)
429b2bb Daniel Han committed 1 month ago
Studio: stop recipe dialogs opening a hint tooltip on open (#13668)
72a45f7 Keyi Yu committed 1 month ago