nn.LSTM, nn.Embedding, nn.Tanh C++ Implementations

umutsuluhan · March 16, 2023, 2:52am

Hi,

I need C++ backend implementations of nn.LSTM, nn.Embedding and nn.Tanh for a project that I must match exact or close behaviour of PyTorch. I was able to find code but couldn’t be sure. Could you please pin these implementations?

Thanks

ptrblck · March 16, 2023, 7:12am

Could you post the code locations you have already found so that we could double check them, please?

umutsuluhan · March 16, 2023, 2:44pm

Yeah, sure my bad. I had to include it. This is LSTM implementation I found:

github.com

pytorch/pytorch/blob/6faa6c68e8b76fb68f3a2b2783685102d0e87c00/aten/src/ATen/native/vulkan/ops/Lstm.cpp

#include <ATen/TensorOperators.h>
#include <ATen/native/vulkan/ops/Lstm.h>
#include <ATen/native/vulkan/ops/Mm.h>
#include <torch/library.h>

#ifndef AT_PER_OPERATOR_HEADERS
#include <ATen/Functions.h>
#else
#include <ATen/ops/addmm.h>
#include <ATen/ops/cat.h>
#include <ATen/ops/sigmoid.h>
#include <ATen/ops/slice.h>
#include <ATen/ops/tanh.h>
#endif

namespace at {
namespace native {
namespace vulkan {
namespace ops {
namespace {

This file has been truncated. show original

Please let me know if this is correct. I haven’t had a chance to look for Embedding and Tanh but those should be pretty straightforward.

Thanks