ChatRWKV: 0.5.0 operators.cu fails to compile on compute 6.x

[...]/rwkv/cuda/operators.cu(123): error: no instance of overloaded function "atomicAdd" matches the argument list
            argument types are: (__half *, __half)
          atomicAdd(&y[k], __float2half(y_local));
          ^

This is likely because my GPU (a 1060) only supports compute 6.1 while atomicAdd support for __half requires compute 7.0 per https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#atomicadd

It seems like

#ifdef __CUDA_ARCH__ <= 600
/* magic stuff here */
#endif

would be needed to support lower compute versions. I don’t know enough about this to contribute anything more helpful, unfortunately.