Jiameng Li

C-PTQ: Fisher-weighted Channel-wise Sensitivity for Post-training Quantization of MLLMs

Jiameng Li, Han Zhou, Matthew B. Blaschko

Multimodal large language models (MLLMs) require huge memory and computational costs, which limits their practical deployment. Post-training quantization (PTQ) techniques offer an efficient solution for model compression and inference acceleration. Yet, the quantized model faces…Read more