-
Notifications
You must be signed in to change notification settings - Fork 0
/
14_finetune_gpt2.py
63 lines (51 loc) · 1.98 KB
/
14_finetune_gpt2.py
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
import os
import json
from pathlib import Path
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1,2,3"
from transformers import GPT2LMHeadModel, Trainer, TrainingArguments, GPT2TokenizerFast
from datasets import Dataset
script_directory = Path(__file__).parent
data_directory = script_directory / "data"
web_qsp__directory = data_directory / f"benchmarks/WebQSP"
artifacts_finetune_directory = data_directory / f"artifacts/WebQSPFineTuneGPT2"
training_strings = []
with open(web_qsp__directory / "webqsp.examples.train.json", "r") as f:
data = json.load(f)
for datapoint in data:
for answer_str in datapoint['answers_str']:
if answer_str:
utterance = datapoint['utterance']
sentence = f"Question: {utterance}\nAnswer: {answer_str}"
training_strings.append(sentence)
break
# Load the GPT-2 model
model = GPT2LMHeadModel.from_pretrained("gpt2")
tokenizer = GPT2TokenizerFast.from_pretrained("gpt2")
tokenizer.add_special_tokens({'pad_token': tokenizer.eos_token})
def tokenize_function(examples):
return tokenizer(examples['text'], truncation=True, padding='max_length', max_length=512)
dataset = Dataset.from_dict({'text': training_strings})
dataset = dataset.map(tokenize_function, batched=True)
dataset = dataset.map(lambda examples: {'labels': examples['input_ids']}, batched=True)
# Define the training arguments
training_args = TrainingArguments(
output_dir=str(artifacts_finetune_directory),
evaluation_strategy="no",
learning_rate=2e-5,
weight_decay=0.01,
num_train_epochs=4,
per_device_train_batch_size=4,
save_strategy="epoch", # Save a checkpoint at the end of each epoch
save_total_limit=1,
)
# Initialize the Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=dataset,
eval_dataset=None,
)
# Start the training
trainer.train()
model.save_pretrained(str(artifacts_finetune_directory))
tokenizer.save_pretrained(str(artifacts_finetune_directory))