File size: 2,270 Bytes
e0cc050
f2eeb87
 
 
 
 
1ea0d00
f2eeb87
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0f18d6d
 
2eec429
b606f8a
e0cc050
0f18d6d
 
 
 
 
 
 
b459e61
0f18d6d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b459e61
0f18d6d
 
 
 
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
from transformers import BertModel, BertTokenizer, TrainingArguments, Trainer
from datasets import Dataset



# Prepare the dataset (simplified)
def prepare_text_dataset(data, chunk_size):
    # Split the text into smaller chunks (consider logical divisions of the Constitution)
    chunks = [data[i:i+chunk_size] for i in range(0, len(data), chunk_size)]
    # Convert chunks to dictionaries with a single feature "text"
    formatted_data = [{"text": chunk} for chunk in chunks]
    # Create the dataset from the list of dictionaries
    formatted_dataset = Dataset.from_list(formatted_data)
    # Tokenize the text using the MBart tokenizer
    formatted_dataset = formatted_dataset.map(
        lambda x: tokenizer(x["text"], truncation=True, padding="max_length"),
        batched=True
    )

    # Set the format of the dataset to "torch" for compatibility with the model
    formatted_dataset.set_format("torch")
    # Print a message indicating preparation completion (optional)
    print('Prep done')

    return formatted_dataset

def init():
    # Load the model and tokenizer
    model_name = "language-ml-lab/AzerBert"  # Replace with your model name if different
    tokenizer = BertTokenizer.from_pretrained(model_name)
    model = BertModel.from_pretrained(model_name)
    chunk_size = 512

    # Load the plain text (replace with your actual loading logic)
    with open("constitution.txt", "r", encoding="utf-8") as f:
      constitution_text = f.read()
    
    # Prepare the dataset
    train_dataset = prepare_text_dataset(constitution_text, chunk_size)
    
    # Define training arguments
    training_args = TrainingArguments(
        output_dir="./results",  # Adjust output directory
        overwrite_output_dir=True,
        num_train_epochs=3,  # Adjust training epochs
        per_device_train_batch_size=1,  # Adjust batch size based on your GPU memory
        save_steps=500,
        save_total_limit=2,
    )
    
    # Create the Trainer
    trainer = Trainer(
        model=model,
        args=training_args,
        train_dataset=train_dataset,
    )
    
    # Start training
    trainer.train()
    
    # Save the fine-tuned model
    model.save_pretrained("./fine-tuned_model")
    tokenizer.save_pretrained("./fine-tuned_model")