DeepYard
U

Unsloth

3

Fine-tune LLMs 2-5x faster with 80% less memory

Open Sourceopen-source Trending

About

Unsloth is an open-source library that dramatically accelerates LLM fine-tuning by optimizing the training loop with custom Triton kernels. It achieves 2-5x faster training with up to 80% less memory usage, allowing users to fine-tune larger models on consumer GPUs. Unsloth supports popular models and is compatible with Hugging Face's ecosystem.

Details

Typefine-tuning-optimizer
LanguagesPython
Supported ModelsLlama, Mistral, Phi, Gemma, Qwen, CodeLlama, TinyLlama

Tags

fine-tuningoptimizationmemory-efficientfast-trainingtritonconsumer-gpulora