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
Context parallel: no graph break in compiled Ulysses attention on packed rows (#16734)
4accec0 Daniel Han committed 15 days ago
Studio: answer Refresh and Edit in a Compare pane with its own model (#16698)
ff41710 Nilay committed 15 days ago
Multi-account Playwright: an aborted /login must not use up the stalled-load retry (#16768)
66c239a Daniel Han committed 15 days ago
Studio: run each Compare model with its own recommended sampling (#16686)
f7e499a Nilay committed 15 days ago
Keep reasoning when formatting chats with the gemma-4-thinking template (#16701)
846dac6 Nilay committed 15 days ago
Studio: show the chat background in conversations on Windows at 100% scaling (#16702)
157305f Leo Borcherding committed 15 days ago
Studio: keep the typed message when picking a saved prompt (#16695)
507937d Nilay committed 15 days ago
Studio: stop silently dropping earlier replies sent as item references (#16694)
5e702b1 Nilay committed 15 days ago
Studio: load NVFP4 diffusion checkpoints from .safetensors only (#16739)
fbf5da0 Daniel Han committed 15 days ago
Context parallel: compiled layers follow each packed row under Ulysses (#16733)
e464a30 Daniel Han committed 15 days ago
CPU backend: install the every_n_layers gradient checkpointing router (#16730)
f450da3 Daniel Han committed 15 days ago
Notebooks CI: refresh the Colab oracle snapshot to the torch 2.14.1 GPU runtime (#16738)
0efd019 Daniel Han committed 15 days ago
CPU backend: honour unsloth_tiled_mlp and UNSLOTH_TILED_MLP in FastLanguageModel.from_pretrained (#16731)
4ee2f91 Daniel Han committed 15 days ago
Studio video: report the dense bf16 text encoder when the automatic hosted fp8 encoder cannot be fetched (#16724)
ff7b85d Daniel Han committed 15 days ago
Kaggle Studio GPU CI: request hardware detection after sign-in (#16728)
7ec4b20 Daniel Han committed 15 days ago
Scan baseline: re-review the five unsloth-zoo sites 2026.11.2 moved (#16732)
cb92a16 Daniel Han committed 15 days ago
Pipeline parallel: weight the MoE router aux loss per accumulation step as the model forward does (#16708)
590119d Daniel Han committed 15 days ago
Studio diffusion: Automatic prefers one card plus a second-card copy over a layer split for FLUX when the primary holds the whole denoiser (#16687)
09fd0d3 Daniel Han committed 15 days ago
Studio diffusion: Automatic partner card, FLUX batch split, LoRA scale mirror, CFG batching fallback (#16682)
4a86327 Daniel Han committed 15 days ago
Studio diffusion: CFG-parallel for a group-offloaded denoiser (#16681)
4ddd527 Daniel Han committed 15 days ago
Studio diffusion: Automatic arms CFG-parallel on an idle card of the same model (#16680)
3ea5429 Daniel Han committed 15 days ago
Studio diffusion: CFG-parallel for U-Net families (SDXL) by batch split (#16679)
eb805d3 Daniel Han committed 15 days ago
Decode with CUDA graphs by default in plain generate(), with memory release and an eager fallback (#16691)
db82c05 Daniel Han committed 15 days ago
Studio: state the audio memory estimate in placement refusals (#16628)
781b72f Debojyoti Banerjee committed 15 days ago
README: say UNSLOTH_LLAMA_CPP_BACKEND picks the GGUF chat build, not PyTorch (#14256)
59631d2 Leo Borcherding committed 16 days ago
Capability registry: refuse tensor parallel where tp_native cannot shard, and add the zoo MoE families (#16709)
fd50f8f Daniel Han committed 16 days ago
Studio: preserve refusal messages in imported chats (#16599)
91be109 DimitriChrysafis committed 16 days ago
Studio: decode escaped text in spreadsheet previews (#16592)
a17bfe9 DimitriChrysafis committed 16 days ago
Native quant tests: make peft see torchao as absent when the test blocks it (#16704)
a1e340b Daniel Han committed 16 days ago