Dodo: Dynamic Contextual Compression for Decoder-only LMs
This addresses the problem of high computational overhead in long-context language models for users in NLP applications, but it is incremental as it builds on existing transformer and compression techniques.
The paper tackles the inefficiency of transformer-based language models in long contexts by proposing Dodo, a dynamic contextual compression method that reduces self-attention cost to a fraction, achieving a 20x compression ratio with 98% BLEU score for reconstruction in autoencoding tasks.
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.