Efficient Beam Search for Large Language Models Using Trie-Based Decoding
This addresses memory constraints for deploying large language models in resource-limited environments, though it appears incremental as it optimizes an existing decoding approach.
This paper tackles the memory inefficiency of batch-based beam search in large language models by introducing a trie-based parallel decoding method that shares a single KV cache across beams with common prefixes, resulting in 4-8× memory savings and up to 2.4× faster decoding without quality loss.
This work presents a novel trie (prefix-tree)-based parallel decoding method that addresses the memory inefficiency of batch-based beam search. By sharing a single KV cache across beams with common prefixes, our approach dramatically reduces memory usage and enables efficient decoding. We evaluated our method across three attention architectures, Multi-Head Attention (Phi-3.5-mini-instruct), Grouped Query Attention (Llama-3.1-8B-Instruct), and Sliding Window Attention (Mistral-Small-24B-Instruct-2501), using CNN/DailyMail for abstractive summarization and HumanEval for code generation. Our experiments demonstrate substantial memory savings (4--8$\times$) and up to 2.4$\times$ faster decoding, without compromising generation quality. These results highlight our method's suitability for memory-constrained environments and large-scale deployments.