Speculative Streaming: Fast LLM Inference without Auxiliary Models
This addresses the problem of high inference system complexity and resource usage for developers and users in application-specific LLM settings, offering a parameter-efficient solution.
The paper tackles the complexity and resource demands of using auxiliary draft models in speculative decoding for LLM inference by introducing Speculative Streaming, a single-model method that changes the fine-tuning objective to future n-gram prediction, achieving speed-ups of 1.8-3.1X across tasks without quality loss.
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.