ParallelKernelBench: Can LLMs write fast multi-GPU kernels?

**Brief Summary (2‑3 sentences)** ParallelKernelBench is an open‑source framework that explores whether large language models can generate high‑performance mult