Skip to content

[Not_for_merge][AMD][Workaround] Enable bf16 for transform_inference extension #1

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
k-artem wants to merge 1 commit into
base: master
Choose a base branch
from
Open

[Not_for_merge][AMD][Workaround] Enable bf16 for transform_inference extension #1

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions ...transformer/inference/csrc/pt_binding.cpp → .../transformer/inference/csrc/pt_binding.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,6 +11,10 @@
#include "inference_cublas_wrappers.h"
#include "inference_cuda_layers.h"

#ifdef BF16_AVAILABLE
#include <cuda_bf16.h>
#endif

std::array<int, 3> gemm_algos = std::array<int, 3>({99, 99, 99});

// NOTE: This activation function type enum should be always in sync
Expand Down
6 changes: 3 additions & 3 deletions op_builder/transformer_inference.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,7 +55,7 @@ def filter_ccs(self, ccs):

def sources(self):
return [
'csrc/transformer/inference/csrc/pt_binding.cpp',
'csrc/transformer/inference/csrc/pt_binding.cu',
'csrc/transformer/inference/csrc/gelu.cu',
'csrc/transformer/inference/csrc/relu.cu',
'csrc/transformer/inference/csrc/layer_norm.cu',
Expand Down Expand Up @@ -83,6 +83,6 @@ def nvcc_args(self):
This cannot be avoided via forward declarations for this transformer_inference extension,
since `pt_binding.cpp` code explicitly requires the BF16 header, so disable it for now.
"""
if self.is_rocm_pytorch():
self.enable_bf16 = False
#if self.is_rocm_pytorch():
# self.enable_bf16 = False
return args
Loading