Expanding Computation Spaces of LLMs at Inference Time
This addresses the problem of enhancing inference-time computational capacity for language model users, though it is incremental as it builds on prior work on filler tokens.
The study investigated whether language models can use artificially inserted filler tokens at inference to expand computational space, finding that smaller models benefit most with up to 12.372 percentage point gains in tasks like open-domain QA and math.
Chain-of-thought (CoT) rationale enables language models to use additional task-related text for problem-solving, benefiting not only from detailed reasoning steps but also from the expanded computational space of longer inputs. Prior work has trained filler or special tokens to serve as additional computation spaces. In this study, we investigate whether language models can leverage artificially inserted sequences of filler tokens solely at inference. We first identify effective token types, numbers, and insertion locations, then examine at what stage of training models begin to exploit the expanded computation space, and finally analyze dynamics within these spaces via attention maps. Experiments on models ranging from 1.7B to 32B across open-domain QA and math tasks show that appropriate token types and counts vary, but placing filler tokens directly before the final 'Answer:' token is most effective. Smaller models benefit most, up to 12.372 percentage points in SmolLM2-1.7B-Instruct, indicating that these spaces act as additional computational capacity rather than redundant input. Attention maps reveal that expanded spaces often continue the original attention mechanism and sometimes focus on questions or answer options, suggesting meaningful computation for problem-solving.