Yejoon Lee

Unified Static-Dynamic Pruning for Efficient LLM Inference

Jinhyeok Kim, Yejoon Lee, Jaeyoung Do

The increasing deployment of large language models (LLMs) has magnified the computational and memory bottlenecks of autoregressive decoding, where low compute intensity and bandwidth-bound kernels dominate inference cost. Weight pruning offers a promising remedy, but existing met…Read more