Make Some Noise: Unlocking Language Model Parallel Inference Capability through Noisy Training
This addresses the need for more efficient inference acceleration in language models, particularly for reducing migration costs and device memory demands, though it is incremental as it builds on existing speculative decoding approaches.
The paper tackles the problem of costly and memory-intensive speculative decoding methods for large language models by proposing the Make Some Noise (MSN) training framework, which enhances parallel decoding capability through noisy training and achieves a 2.3-2.7x inference speed improvement without performance loss.
Existing speculative decoding methods typically require additional model structure and training processes to assist the model for draft token generation. This makes the migration of acceleration methods to the new model more costly and more demanding on device memory. To address this problem, we propose the Make Some Noise (MSN) training framework as a replacement for the supervised fine-tuning stage of the large language model. The training method simply introduces some noise at the input for the model to learn the denoising task. It significantly enhances the parallel decoding capability of the model without affecting the original task capability. In addition, we propose a tree-based retrieval-augmented Jacobi (TR-Jacobi) decoding strategy to further improve the inference speed of MSN models. Experiments in both the general and code domains have shown that MSN can improve inference speed by 2.3-2.7x times without compromising model performance. The MSN model also achieves comparable acceleration ratios to the SOTA model with additional model structure on Spec-Bench.