unslothaiunsloth
Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, Kimi K3, MiniMax-H3, Gemma 4, FLUX and more.
TL;DR · 30-second scan
unsloth (Python) — Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, Kimi K3, MiniMax-H3, Gemma 4, FLUX and more.
You are fine-tuning Llama, Mistral, Qwen or Gemma on one consumer GPU and hitting memory limits.
AI & ML
Fine-tunes open models roughly twice as fast and with substantially less VRAM than a standard Hugging Face setup, through hand-written Triton kernels and a reworked backprop path. The practical consequence is what matters: models that needed a rented A100 become trainable on a single consumer GPU or a free Colab tier. For anyone in India experimenting on a budget, that is the difference between doing this and not. Apache-2.0, Python, and it tracks new model releases quickly. The tradeoff is scope — it targets single-GPU work, multi-GPU support has lagged the single-GPU path, and the optimisations bind it to specific model architectures, so the newest release is not always supported on day one.
You are fine-tuning Llama, Mistral, Qwen or Gemma on one consumer GPU and hitting memory limits.
You train across a multi-GPU cluster, or you need architectures outside its supported list.
Add this badge to your README to show your project is curated on StackPicks. Free, lightweight (180×28 SVG), and gives your visitors a one-click way to see honest take + alternatives.
[](https://stackpicks.dev/repo/unslothai-unsloth)
<a href="https://stackpicks.dev/repo/unslothai-unsloth"><img src="https://stackpicks.dev/api/badge/unslothai-unsloth" alt="Featured on StackPicks" width="180" height="28" /></a>
Are you the maintainer of unslothai/unsloth? Add the badge and we'll feature your project in the weekly curator newsletter.