CacheFormer: High Attention-Based Segment Caching

Sushant Singh, Ausif Mahmood

Efficiently handling long contexts in transformer-based language models with low perplexity is an active area of research. Numerous recent approaches like Linformer, Longformer, Pe…