Large Language Model Fine-tuning & Quantization

9 repos

This cluster focuses on fine-tuned and quantized versions of popular open-source LLMs (Llama, Mistral, Zephyr) optimized for inference efficiency and conversational tasks. The repositories center on techniques like supervised fine-tuning (SFT), reinforcement learning from human feedback alignment (RefAlign), and 8-bit quantization via bitsandbytes to make models smaller and faster while maintaining quality. Researchers and practitioners exploring this cluster will find model checkpoints, evaluation metrics, and optimization approaches for deploying language models in resource-constrained environments.

Related clusters