Scaling LLM Speculative Decoding: Non-Autoregressive Forecasting in Large-Batch Scenarios
This work addresses a bottleneck in efficient LLM inference for real-world applications, offering a scalable solution that reduces resource demands compared to prior speculative decoding methods.
The paper tackles the problem of scaling speculative decoding for LLM inference in large-batch scenarios, where existing methods rely on complex draft trees and high computational resources. It proposes SpecFormer, a novel architecture that integrates unidirectional and bidirectional attention to enable parallel generation, achieving consistent acceleration with lower training and computational costs across various model scales.
Speculative decoding accelerates LLM inference by utilizing otherwise idle computational resources during memory-to-chip data transfer. Current speculative decoding methods typically assume a considerable amount of available computing power, then generate a complex and massive draft tree using a small autoregressive language model to improve overall prediction accuracy. However, methods like batching have been widely applied in mainstream model inference systems as a superior alternative to speculative decoding, as they compress the available idle computing power. Therefore, performing speculative decoding with low verification resources and low scheduling costs has become an important research problem. We believe that more capable models that allow for parallel generation on draft sequences are what we truly need. Recognizing the fundamental nature of draft models to only generate sequences of limited length, we propose SpecFormer, a novel architecture that integrates unidirectional and bidirectional attention mechanisms. SpecFormer combines the autoregressive model's ability to extract information from the entire input sequence with the parallel generation benefits of non-autoregressive models. This design eliminates the reliance on large prefix trees and achieves consistent acceleration, even in large-batch scenarios. Through lossless speculative decoding experiments across models of various scales, we demonstrate that SpecFormer sets a new standard for scaling LLM inference with lower training demands and reduced computational costs.