LLM Compression: Enhancing AWQ

Dec 1, 2024
Code

Graduation project focused on improving AWQ (Activation-aware Weight Quantization) with extra scaling.

  • Obtained lower perplexity for INT3-quantized OPT and Llama 2 models.
Machine Learning LLM Compression
Junhyun Kim
Junhyun Kim
M.S. Robotics Student

Robotics student interested in robot manipulation and vision-language-action models.