SwiftKV: Fast Prefill-Optimized Inference with Knowledge-Preserving Model Transformation.

Aurick Qiao, Zhewei Yao, Samyam Rajbhandari, Yuxiong He

Browse the full EMNLP paper archive.