deepreinforce-ai/CUDA-L1

CUDA-L1是一个基于对比强化学习的自动化CUDA优化框架，它通过KernelBench数据集中的250个PyTorch工作负载来评估和优化GPU内核的生成。CUDA-L1 is an automated reinforcement learning framework for CUDA optimization based on contrastive reinforcement learning, which evaluates and optimizes the generation of GPU kernels using 250 PyTorch workloads from the KernelBench dataset.