Public
  1. Public

onnxruntime

Public

Commits

Hector LiGitHub
38963d81eb4
Hector Li authored and GitHub committed bbdd1d658b7
use cudnnRNNForwardInferenceEx for unpacked (padded) layout case (#899)

1. Use cudnnRNNForwardInferenceEx for unpacked (padded) layout case, place the sequence_lens data on CPU
2. Fix hard code device ID issue. In cuda kernel, it should get the device id from provider.