Speculative Streaming: Fast LLM Inference without Auxiliary Models

Nikhil Bhendawade; Irina Belousova; Qichen Fu; Henry Mason; Mohammad Rastegari; Mahyar Najibi

返回文献探索

Paper · arXiv 2402.11131

Speculative Streaming: Fast LLM Inference without Auxiliary Models

Nikhil Bhendawade, Irina Belousova, Qichen Fu, Henry Mason, Mohammad Rastegari, Mahyar Najibi

42 upvotesFebruary 16, 2024arXiv 预印本

AI 摘要

Speculative Streaming is a method that incorporates drafting into a target language model, enhancing decoding speed without quality loss while using fewer parameters.

speculative decodinglarge target language modelauxiliary draft modelfine-tuningnext token predictionfuture n-gram predictionspeculative streamingdecoding speedSummarizationStructured QueriesMeaning Representationparameter-efficientMedusa-style architectures

Abstract

Speculative decoding is a prominent technique to speed up the inference of a large target language model based on predictions of an auxiliary draft model. While effective, in application-specific settings, it often involves fine-tuning both draft and target models to achieve high acceptance rates. As the number of downstream tasks grows, these draft models add significant complexity to inference systems. We propose Speculative Streaming, a single-model speculative decoding method that fuses drafting into the target model by changing the fine-tuning objective from next token prediction to future n-gram prediction. Speculative Streaming speeds up decoding by 1.8 - 3.1X in a diverse set of tasks, such as Summarization, Structured Queries, and Meaning Representation, without sacrificing generation quality. Additionally, Speculative Streaming is parameter-efficient. It achieves on-par/higher speed-ups than Medusa-style architectures while using ~10000X fewer extra parameters, making it well-suited for resource-constrained devices.