Dodo: Dynamic Contextual Compression for Decoder-only LMs

Autor:	Qin, Guanghui, Rosset, Corby, Chau, Ethan C., Rao, Nikhil, Van Durme, Benjamin
Rok vydání:	2023
Předmět:	Computer Science - Computation and Language Computer Science - Artificial Intelligence Computer Science - Machine Learning I.2.7 I.2.6
Druh dokumentu:	Working Paper
Popis:	Transformer-based language models (LMs) are inefficient in long contexts. We propose Dodo, a solution for context compression. Instead of one vector per token in a standard transformer model, Dodo represents text with a dynamic number of hidden states at each layer, reducing the cost of self-attention to a fraction of typical time and space. Moreover, off-the-shelf models such as LLaMA can be adapted to Dodo by efficient parameter tuning methods such as LoRA. In use, Dodo can act as either an autoregressive LM or a context compressor for downstream tasks. We demonstrate through experiments in language modeling, question answering, and summarization that Dodo retains capabilities in these tasks, while drastically reducing the overhead during decoding. For example, in the autoencoding task, Dodo shrinks context at a 20x compression ratio with a BLEU score of 98% for reconstruction, achieving nearly lossless encoding. Comment: ACL 2024 camera-ready. 15 pages and 7 figures
Databáze:	arXiv
Externí odkaz:	http://arxiv.org/abs/2310.02409 Zobrazit plný text záznamu View this record from Arxiv