DeepSpeed
e721cb69 - Supporting different hidden dimensions for transformer kernels-v2 (#934)

Commit
4 years ago
Supporting different hidden dimensions for transformer kernels-v2 (#934) Co-authored-by: Jeff Rasley <jerasley@microsoft.com>
Parents
  • csrc/transformer
    • File
      ds_transformer_cuda.cpp
    • File
      gelu_kernels.cu
  • tests/unit
    • File
      test_cuda_backward.py
    • File
      test_cuda_forward.py