code/transformercode/config.py

import torch

# TRAIN_SAMPLE_PATH = BASE_PATH+"/news_data/train.json"
# VAL_SAMPLE_PATH = BASE_PATH+"/news_data/val.json"

# ZH_VOCAB_PATH = BASE_PATH+"/news_data/zh.txt"
# EN_VOCAB_PATH = BASE_PATH+"/news_data/en.txt"

# TRAIN_SAMPLE_PATH = "news_data/data/json/train_pretty.json"
# VAL_SAMPLE_PATH = "news_data/data/json/test_pretty.json"

# ZH_VOCAB_PATH = "news_data/data/zh.txt"
# EN_VOCAB_PATH = "news_data/data/en.txt"

# TRAIN_SAMPLE_PATH = "news_data/data/json/train_clean.json"
# VAL_SAMPLE_PATH = "news_data/data/json/dev_clean.json"

# ZH_VOCAB_PATH = "news_data/data/zh_augmented.txt"
# EN_VOCAB_PATH = "news_data/data/en_augmented.txt"

# TRAIN_SAMPLE_PATH = "news_data/data/json/train_clean_aligned.json"
# VAL_SAMPLE_PATH = "news_data/data/json/dev_clean_aligned.json"


# ZH_VOCAB_PATH = "news_data/data/zh_aligned.txt"
# EN_VOCAB_PATH = "news_data/data/en_aligned.txt"


# 最新更改
BASE_PATH = "D:/gitcode/transformer_code/"
TRAIN_SAMPLE_PATH = BASE_PATH + "news_data/data/json/train_clean_aligned.json"
VAL_SAMPLE_PATH = BASE_PATH + "news_data/data/json/dev_clean_aligned.json"

ZH_VOCAB_PATH = BASE_PATH + "news_data/data/zh_aligned.txt"
EN_VOCAB_PATH = BASE_PATH + "news_data/data/en_aligned.txt"



PAD_ID = 0
UNK_ID = 1
SOS_ID = 2
EOS_ID = 3

#超参数
D_MODEL = 512
N_HEAD = 8
D_FF = 2048
N = 6
# N = 4
DROPOUT = 0.1

BATCH_SIZE = 32
LR = 1e-5
EPOCH = 50


MAX_LEN = 128 #句子最大长度

device = "cuda" if torch.cuda.is_available() else "cpu"

#标签平滑系数
LABEL_SMOOTHING = 0.1