train/scripts/convert_sbert_from_huggingface_to_tencentpretrain.py

import argparse
import collections
import torch


def convert_sbert_transformer_encoder_from_huggingface_to_tencentpretrain(input_model, output_model, layers_num):
    for i in range(layers_num):
        for j in range(2):
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".self_attn.linear_layers.0.weight"] = \
                input_model["encoder.layer." + str(i) + ".attention.self.query.weight"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".self_attn.linear_layers.0.bias"] = \
                input_model["encoder.layer." + str(i) + ".attention.self.query.bias"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".self_attn.linear_layers.1.weight"] = \
                input_model["encoder.layer." + str(i) + ".attention.self.key.weight"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".self_attn.linear_layers.1.bias"] = \
                input_model["encoder.layer." + str(i) + ".attention.self.key.bias"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".self_attn.linear_layers.2.weight"] = \
                input_model["encoder.layer." + str(i) + ".attention.self.value.weight"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".self_attn.linear_layers.2.bias"] = \
                input_model["encoder.layer." + str(i) + ".attention.self.value.bias"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".self_attn.final_linear.weight"] = \
                input_model["encoder.layer." + str(i) + ".attention.output.dense.weight"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".self_attn.final_linear.bias"] = \
                input_model["encoder.layer." + str(i) + ".attention.output.dense.bias"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".layer_norm_1.gamma"] = \
                input_model["encoder.layer." + str(i) + ".attention.output.LayerNorm.weight"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".layer_norm_1.beta"] = \
                input_model["encoder.layer." + str(i) + ".attention.output.LayerNorm.bias"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".feed_forward.linear_1.weight"] = \
                input_model["encoder.layer." + str(i) + ".intermediate.dense.weight"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".feed_forward.linear_1.bias"] = \
                input_model["encoder.layer." + str(i) + ".intermediate.dense.bias"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".feed_forward.linear_2.weight"] = \
                input_model["encoder.layer." + str(i) + ".output.dense.weight"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".feed_forward.linear_2.bias"] = \
                input_model["encoder.layer." + str(i) + ".output.dense.bias"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".layer_norm_2.gamma"] = \
                input_model["encoder.layer." + str(i) + ".output.LayerNorm.weight"]
            output_model["encoder.encoder_" + str(j) + ".transformer." + str(i) + ".layer_norm_2.beta"] = \
                input_model["encoder.layer." + str(i) + ".output.LayerNorm.bias"]


def main():
    parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
    parser.add_argument("--input_model_path", type=str, default="models/input_model.bin",
                        help=".")
    parser.add_argument("--output_model_path", type=str, default="models/output_model.bin",
                        help=".")
    parser.add_argument("--layers_num", type=int, default=12, help=".")

    args = parser.parse_args()

    input_model = torch.load(args.input_model_path, map_location='cpu')

    output_model = collections.OrderedDict()

    for i in range(2):
        output_model["embedding.embedding_" + str(i) + ".word.embedding.weight"] = \
            input_model["embeddings.word_embeddings.weight"]
        output_model["embedding.embedding_" + str(i) + ".pos.embedding.weight"] = \
            input_model["embeddings.position_embeddings.weight"]
        output_model["embedding.embedding_" + str(i) + ".seg.embedding.weight"] = \
            torch.cat((torch.Tensor([[0]*input_model["embeddings.token_type_embeddings.weight"].size()[1]]),
                       input_model["embeddings.token_type_embeddings.weight"]), dim=0)
        output_model["embedding.embedding_" + str(i) + ".layer_norm.gamma"] = \
            input_model["embeddings.LayerNorm.weight"]
        output_model["embedding.embedding_" + str(i) + ".layer_norm.beta"] = \
            input_model["embeddings.LayerNorm.bias"]

    convert_sbert_transformer_encoder_from_huggingface_to_tencentpretrain(input_model, output_model, args.layers_num)
    torch.save(output_model, args.output_model_path)

if __name__ == "__main__":
    main()