Priority Sampling of Large Language Models for Compilers
This addresses the need for more reliable and controllable sampling in code generation for compilers, though it appears incremental as it builds on existing sampling paradigms.
The paper tackles the problem of sampling from large language models for code generation, where existing methods like Nucleus Sampling suffer from repetition or incoherence and require task-specific temperature tuning. The result is Priority Sampling, a deterministic technique that produces unique samples ordered by confidence, achieving a performance improvement from 2.87% to 5% over -Oz and outperforming an autotuner in just 30 samples.
Large language models show great potential in generating and optimizing code. Widely used sampling methods such as Nucleus Sampling increase the diversity of generation but often produce repeated samples for low temperatures and incoherent samples for high temperatures. Furthermore, the temperature coefficient has to be tuned for each task, limiting its usability. We present Priority Sampling, a simple and deterministic sampling technique that produces unique samples ordered by the model's confidence. Each new sample expands the unexpanded token with the highest probability in the augmented search tree. Additionally, Priority Sampling supports generation based on regular expression that provides a controllable and structured exploration process. Priority Sampling outperforms Nucleus Sampling for any number of samples, boosting the performance of the original model from 2.87% to 5% improvement over -Oz. Moreover, it outperforms the autotuner used for the generation of labels for the training of the original model in just 30 samples.