ClusterGen: Token Generation in Sublinear Time and Memory with Clustering KV Cache
Track: long paper (up to 4 pages)
Keywords: clustering, kv cache, streaming algorithm, reservoir sampling
Abstract: Despite the significant success of large language models (LLMs), their extensive memory requirements pose challenges for deploying them in long-context token generation. The substantial memory footprint of LLM decoders arises from the necessity to store all previous tokens in the attention module, a requirement imposed by key-value (KV) caching. In this work, our focus is on developing an efficient compression technique for the KV cache. Empirical evidence indicates a significant clustering tendency within key embeddings in the attention module. Building on this key insight, we have devised a novel caching method with sublinear complexity, employing online clustering on key tokens and online l2 sampling on values. The result is a provably accurate and efficient attention decoding algorithm, termed ClusterGen. Not only does this algorithm ensure a sublinear memory footprint and sublinear time complexity, but we also establish a tight error bound for our approach.
Empirical evaluations on long-context question-answering tasks demonstrate that ClusterGen significantly outperforms existing and state-of-the-art KV cache compression methods in terms of performance and efficiency.
Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.
Submission Number: 88
Loading