NvFuser with torch.compile

Is there any torch dynamo backend that generates CUDA code (.cu) and compiles it using nvcc (similar to how inductor uses g++ to compile fused c++) ? AFAIK, triton generates ptx code directly