Chen YiJia
commited on
Commit
•
091d980
1
Parent(s):
5cedadf
reorganise
Browse files- .DS_Store +0 -0
- first_epoch/config.json → config.json +0 -0
- first_epoch/eval_results.txt → eval_results.txt +0 -0
- first_epoch/generation_config.json → generation_config.json +0 -0
- first_epoch/model_args.json → model_args.json +0 -0
- first_epoch/optimizer.pt → optimizer.pt +0 -0
- first_epoch/pytorch_model.bin → pytorch_model.bin +0 -0
- first_epoch/scheduler.pt → scheduler.pt +0 -0
- first_epoch/special_tokens_map.json → special_tokens_map.json +0 -0
- first_epoch/spiece.model → spiece.model +0 -0
- test.py +0 -33
- first_epoch/tokenizer_config.json → tokenizer_config.json +0 -0
- train.py +0 -49
- first_epoch/training_args.bin → training_args.bin +0 -0
.DS_Store
ADDED
Binary file (6.15 kB). View file
|
|
first_epoch/config.json → config.json
RENAMED
File without changes
|
first_epoch/eval_results.txt → eval_results.txt
RENAMED
File without changes
|
first_epoch/generation_config.json → generation_config.json
RENAMED
File without changes
|
first_epoch/model_args.json → model_args.json
RENAMED
File without changes
|
first_epoch/optimizer.pt → optimizer.pt
RENAMED
File without changes
|
first_epoch/pytorch_model.bin → pytorch_model.bin
RENAMED
File without changes
|
first_epoch/scheduler.pt → scheduler.pt
RENAMED
File without changes
|
first_epoch/special_tokens_map.json → special_tokens_map.json
RENAMED
File without changes
|
first_epoch/spiece.model → spiece.model
RENAMED
File without changes
|
test.py
DELETED
@@ -1,33 +0,0 @@
|
|
1 |
-
from datasets.load import load_dataset
|
2 |
-
import logging
|
3 |
-
import sacrebleu
|
4 |
-
import pandas as pd
|
5 |
-
from simpletransformers.t5 import T5Model, T5Args
|
6 |
-
|
7 |
-
raw_datasets = load_dataset('iwslt2017', 'iwslt2017-zh-en')
|
8 |
-
|
9 |
-
logging.basicConfig(level=logging.INFO)
|
10 |
-
transformers_logger = logging.getLogger("transformers")
|
11 |
-
transformers_logger.setLevel(logging.WARNING)
|
12 |
-
|
13 |
-
|
14 |
-
model_args = T5Args()
|
15 |
-
model_args.max_length = 512
|
16 |
-
model_args.length_penalty = 1
|
17 |
-
model_args.num_beams = 10
|
18 |
-
|
19 |
-
model = T5Model("mt5", "outputs", args=model_args)
|
20 |
-
|
21 |
-
en_zh_test = pd.DataFrame(raw_datasets['test']['translation'])
|
22 |
-
zh_truth = en_zh_test['zh'].tolist()
|
23 |
-
en_input = en_zh_test['en'].tolist()
|
24 |
-
|
25 |
-
zh_preds = model.predict(en_input)
|
26 |
-
en_zh_bleu = sacrebleu.corpus_bleu(zh_preds, zh_truth)
|
27 |
-
print("----------------------------------------------")
|
28 |
-
print("English to Chinese: ", en_zh_bleu.score)
|
29 |
-
|
30 |
-
en_preds = model.predict(zh_truth)
|
31 |
-
zh_en_bleu = sacrebleu.corpus_bleu(en_preds, en_input)
|
32 |
-
print("----------------------------------------------")
|
33 |
-
print("Chinese to English: ", zh_en_bleu.score)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
first_epoch/tokenizer_config.json → tokenizer_config.json
RENAMED
File without changes
|
train.py
DELETED
@@ -1,49 +0,0 @@
|
|
1 |
-
from datasets.load import load_dataset
|
2 |
-
import pandas as pd
|
3 |
-
import logging
|
4 |
-
from simpletransformers.t5 import T5Args, T5Model
|
5 |
-
|
6 |
-
logging.basicConfig(level=logging.INFO)
|
7 |
-
transformers_logger = logging.getLogger("transformers")
|
8 |
-
transformers_logger.setLevel(logging.WARNING)
|
9 |
-
|
10 |
-
raw_datasets = load_dataset('iwslt2017', 'iwslt2017-zh-en')
|
11 |
-
|
12 |
-
train_df = pd.DataFrame(raw_datasets['train']['translation'])
|
13 |
-
train_df.columns = ['input_text', 'target_text']
|
14 |
-
reverse_df = train_df.copy()
|
15 |
-
reverse_df.columns = ['target_text', 'input_text']
|
16 |
-
train_df['prefix'] = 'translate english to chinese'
|
17 |
-
reverse_df['prefix'] = 'translate chinese to english'
|
18 |
-
train_df = pd.concat([train_df, reverse_df])
|
19 |
-
|
20 |
-
eval_df = pd.DataFrame(raw_datasets['validation']['translation'])
|
21 |
-
eval_df.columns = ['input_text', 'target_text']
|
22 |
-
reverse_df = eval_df.copy()
|
23 |
-
reverse_df.columns = ['target_text', 'input_text']
|
24 |
-
eval_df['prefix'] = 'translate english to chinese'
|
25 |
-
reverse_df['prefix'] = 'translate chinese to english'
|
26 |
-
eval_df = pd.concat([eval_df, reverse_df])
|
27 |
-
|
28 |
-
model_args = T5Args()
|
29 |
-
model_args.max_seq_length = 96
|
30 |
-
model_args.train_batch_size = 20
|
31 |
-
model_args.eval_batch_size = 20
|
32 |
-
model_args.num_train_epochs = 4
|
33 |
-
model_args.evaluate_during_training = True
|
34 |
-
model_args.evaluate_during_training_steps = 5000
|
35 |
-
model_args.use_multiprocessing = False
|
36 |
-
model_args.fp16 = False
|
37 |
-
model_args.save_steps = -1
|
38 |
-
model_args.save_model_every_epoch = True
|
39 |
-
model_args.save_eval_checkpoints = False
|
40 |
-
model_args.no_cache = True
|
41 |
-
model_args.reprocess_input_data = True
|
42 |
-
model_args.overwrite_output_dir = False
|
43 |
-
model_args.preprocess_inputs = False
|
44 |
-
model_args.num_return_sequences = 1
|
45 |
-
model_args.wandb_project = "MT5 English-Chinese Translation"
|
46 |
-
|
47 |
-
model = T5Model("mt5", "outputs", args=model_args)
|
48 |
-
|
49 |
-
model.train_model(train_df, eval_data=eval_df, output_dir='mt5_more_epochs')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
first_epoch/training_args.bin → training_args.bin
RENAMED
File without changes
|