10 Oct 2025
Researchers developed the Sample Set Aggregator (SSA), a compact language model designed to synthesize multiple candidate solutions generated by a larger base LLM into a single, refined answer. This approach improves over naive majority voting by 8% pass@5 on the MATH dataset, demonstrating robust performance across various mathematical reasoning benchmarks while maintaining computational efficiency comparable to much larger models.
There are no more papers matching your filters at the moment.
