Entry points

Training Script

Main training script for MLIR RL using PPO.

This module implements the primary training loop for the reinforcement learning system. It initializes the models, loads benchmark data, and iterates through PPO training steps including trajectory collection, policy updates, and periodic evaluation.

Execution Command

CONFIG_FILE_PATH=path/to/config.json poetry run train

Evaluation Script

Evaluation script for trained MLIR RL models.

This module evaluates all trained model checkpoints found in the models directory on benchmark datasets, measuring optimization quality through speedup factors and execution times.

Execution Command

CONFIG_FILE_PATH=path/to/config.json poetry run evaluate

Baseline

Baseline execution time measurement for MLIR benchmarks.

This script measures the execution time of MLIR code without any transformations. It executes the evaluation codes and updates the execution times (in nanoseconds) in the current eval json file.

Execution Command

CONFIG_FILE_PATH=path/to/config.json poetry run baseline