InfiniTrain/infini_train/src/kernels/cpu/embedding.cc at d23535ce99d78126882759575834bec99cc428d3 · InfiniTensor/InfiniTrain · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
#include <memory>

#include "glog/logging.h"

#include "infini_train/include/dispatcher.h"
#include "infini_train/include/tensor.h"

namespace infini_train::kernels::cpu {
std::shared_ptr<Tensor> EmbeddingForward(const std::shared_ptr<Tensor> &input, const std::shared_ptr<Tensor> &weight) {
    /*
        x: [*]
        -> Embedding (weight: [num_embeddings, embedding_dim])
        -> o: [*, embedding_dim]
    */
    CHECK(input->Dtype() == DataType::kINT64);
    const auto &input_dims = input->Dims();
    CHECK_EQ(weight->Dims().size(), 2);
    const int embedding_dim = weight->Dims()[1];
    auto output_dims = input_dims;
    output_dims.push_back(embedding_dim);
    auto output = std::make_shared<Tensor>(output_dims, DataType::kFLOAT32);

    for (int i = 0; i < input->NumElements(); ++i) {
        int idx = static_cast<int>(static_cast<const int64_t *>(input->DataPtr())[i]);
        for (int j = 0; j < embedding_dim; ++j) {
            static_cast<float *>(output->DataPtr())[i * embedding_dim + j]
                = static_cast<float *>(weight->DataPtr())[idx * embedding_dim + j];
        }
    }

    return output;
}

std::shared_ptr<Tensor> EmbeddingBackward(const std::shared_ptr<Tensor> &input, const std::vector<int64_t> &weight_dims,
                                          const std::shared_ptr<Tensor> &grad_output) {
    CHECK(input->Dtype() == DataType::kINT64);
    CHECK_EQ(weight_dims.size(), 2);
    const int embedding_dim = weight_dims[1];
    CHECK_EQ(input->Dims().size() + 1, grad_output->Dims().size());
    for (int idx = 0; idx < input->Dims().size(); ++idx) { CHECK_EQ(input->Dims()[idx], grad_output->Dims()[idx]); }
    CHECK_EQ(*grad_output->Dims().rbegin(), embedding_dim);

    auto grad_weight = std::make_shared<Tensor>(weight_dims, DataType::kFLOAT32);
    grad_weight->Fill(0.0);

    for (int i = 0; i < input->NumElements(); ++i) {
        int idx = static_cast<int>(static_cast<const int64_t *>(input->DataPtr())[i]);
        for (int j = 0; j < embedding_dim; ++j) {
            static_cast<float *>(grad_weight->DataPtr())[idx * embedding_dim + j]
                += static_cast<const float *>(grad_output->DataPtr())[i * embedding_dim + j];
        }
    }
    return grad_weight;
}

} // namespace infini_train::kernels::cpu

#define REGISTER_CPU_EMBEDDING_KERNEL(kernel_name)                                                                     \
    REGISTER_KERNEL(infini_train::Device::DeviceType::kCPU, kernel_name, infini_train::kernels::cpu::kernel_name)

REGISTER_CPU_EMBEDDING_KERNEL(EmbeddingForward)
REGISTER_CPU_EMBEDDING_KERNEL(EmbeddingBackward)

#undef REGISTER_CPU_EMBEDDING_KERNEL