[AAAI 2026] Speculative Sampling with Reinforcement Learning

This repository contains the official code for the paper "Speculative Sampling with Reinforcement Learning" (PDF).

Directory Structure

This repository is organized into two main directories (or branches):

For Training and Testing

cd ReSpS_train_test/

Contains the code for training and evaluating the RL policy. Refer to the README in that directory for detailed setup and usage instructions.

For Statistical Analysis

cd ReSps_stats/

Contains the code for statistical analysis of the RL policy results. Refer to the README in that directory for analysis scripts and instructions.

Getting Started

1. Choose the appropriate directory based on your needs:

   • Training/Testing: Navigate to ReSpS_train_test/
   • Statistical Analysis: Navigate to ReSps_stats/

2. Follow the installation and usage instructions in the respective README files.

Acknowledgments

We would like to thank the authors of the original EAGLE for their contributions and insights. Our work builds upon their foundational research and methodologies.

License

This project is licensed under the MIT License. See the LICENSE file for details in each directory.