KernelBench: Can LLMs Write GPU Kernels? - Benchmark with Torch -> CUDA problems - View it on GitHub