Generalized Parallel Scaling with Interdependent Generations
This addresses a bottleneck in parallel LLM inference for AI practitioners by enabling more efficient and higher-quality response generation, though it is an incremental improvement over existing methods.
The paper tackles the problem of parallel LLM inference where independent responses waste compute resources by not sharing information, proposing Bridge to generate interdependent responses in parallel. Bridge improves relative mean accuracy gains from reinforcement learning by up to 50% and boosts consistency of correct responses with only 2.8%-5.1% new parameters.
Parallel LLM inference scaling involves sampling a set of $N>1$ responses for a single input prompt. However, these $N$ parallel responses tend to be generated independently from each other, partitioning compute resources and leaving potentially useful information in one generation untapped by others. This is in contrast to response length scaling where past computation is used in all future steps. For higher quality responses and response sets, we propose Bridge to generate interdependent responses in parallel by rethinking batched LLM hidden states as holistic tensors rather than independent slices. With only a small amount (2.8%-5.1%) of new parameters, Bridge improves the relative mean accuracy gains from reinforcement learning with verifiable rewards by up to 50% and boosts consistency of correct responses. Trained once, Bridge scales to any generation width, all with greater performance than independent generations, unlocking a more general mode of parallel scaling that effectively leverages information between sequences, compatible with any post-generation aggregation technique.