Activation functions.


void nvte_gelu(const NVTETensor input, NVTETensor output, const NVTETensor scale, NVTETensor amax, NVTETensor scale_inv, cudaStream_t stream)

Compute GELU activation of the input.

  • input[in] Input tensor for GELU activation.

  • output[out] Output tensor.

  • scale[in] Scaling factor of the output tensor.

  • amax[inout] AMAX value of the output tensor.

  • scale_inv[out] Inverse of the output’s scaling factor.

  • stream[in] CUDA stream used for the operation.