/
#18
Unsloth AI
Open-source LLM fine-tuning library using hand-written Triton kernels and manually derived backpropagation to accelerate LoRA and QLoRA training. Supports Llama, Qwen, Mistral, Gemma, DeepSeek, and Phi model families. Delivers 2x faster training with up to 70% less VRAM versus standard Hugging Face Transformers, enabling fine-tuning on consumer GPUs locally.
Topics
Subtopics
MIXTURE OF EXPERTSSMALL MODELSVISION LANGUAGE
Links