119 lines
3.4 KiB
Python
119 lines
3.4 KiB
Python
|
import json
|
||
|
|
import os
|
||
|
|
import torch
|
||
|
|
from datasets import Dataset
|
||
|
|
from transformers import (
|
||
|
|
AutoModelForCausalLM,
|
||
|
|
AutoTokenizer,
|
||
|
|
TrainingArguments,
|
||
|
|
Trainer,
|
||
|
|
DataCollatorForLanguageModeling
|
||
|
|
)
|
||
|
|
from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training
|
||
|
|
import argparse
|
||
|
|
|
||
|
|
def load_dataset(json_path):
|
||
|
|
"""Load the dataset from a JSON file."""
|
||
|
|
with open(json_path, 'r', encoding='utf-8') as f:
|
||
|
|
data = json.load(f)
|
||
|
|
|
||
|
|
# Convert to the format expected by Hugging Face
|
||
|
|
formatted_data = []
|
||
|
|
for item in data:
|
||
|
|
formatted_data.append({
|
||
|
|
"text": f"Prompt: {item['prompt']}\nCompletion: {item['completion']}\n\n"
|
||
|
|
})
|
||
|
|
|
||
|
|
return Dataset.from_list(formatted_data)
|
||
|
|
|
||
|
|
def tokenize_function(examples, tokenizer):
|
||
|
|
"""Tokenize the examples."""
|
||
|
|
return tokenizer(examples["text"], truncation=True, padding="max_length", max_length=512)
|
||
|
|
|
||
|
|
def main():
|
||
|
|
class Args:
|
||
|
|
def __init__(self):
|
||
|
|
self.dataset_path = "datasets/adriana_finetune_dataset.json"
|
||
|
|
self.model_name = "facebook/opt-350m"
|
||
|
|
self.output_dir = "finetuned_model"
|
||
|
|
self.num_train_epochs = 3
|
||
|
|
self.per_device_train_batch_size = 4
|
||
|
|
self.learning_rate = 5e-5
|
||
|
|
self.use_lora = False
|
||
|
|
|
||
|
|
args = Args()
|
||
|
|
|
||
|
|
# Load dataset
|
||
|
|
print(f"Loading dataset from {args.dataset_path}")
|
||
|
|
dataset = load_dataset(args.dataset_path)
|
||
|
|
|
||
|
|
# Load tokenizer and model
|
||
|
|
print(f"Loading tokenizer and model: {args.model_name}")
|
||
|
|
tokenizer = AutoTokenizer.from_pretrained(args.model_name)
|
||
|
|
tokenizer.pad_token = tokenizer.eos_token
|
||
|
|
|
||
|
|
model = AutoModelForCausalLM.from_pretrained(args.model_name)
|
||
|
|
|
||
|
|
# Apply LoRA if requested
|
||
|
|
if args.use_lora:
|
||
|
|
print("Applying LoRA for efficient finetuning")
|
||
|
|
lora_config = LoraConfig(
|
||
|
|
r=16,
|
||
|
|
lora_alpha=32,
|
||
|
|
target_modules=["c_attn", "c_proj"],
|
||
|
|
lora_dropout=0.05,
|
||
|
|
bias="none",
|
||
|
|
task_type="CAUSAL_LM"
|
||
|
|
)
|
||
|
|
model = prepare_model_for_kbit_training(model)
|
||
|
|
model = get_peft_model(model, lora_config)
|
||
|
|
|
||
|
|
# Tokenize dataset
|
||
|
|
print("Tokenizing dataset")
|
||
|
|
tokenized_dataset = dataset.map(
|
||
|
|
lambda examples: tokenize_function(examples, tokenizer),
|
||
|
|
batched=True,
|
||
|
|
remove_columns=dataset.column_names
|
||
|
|
)
|
||
|
|
|
||
|
|
# Set up training arguments
|
||
|
|
training_args = TrainingArguments(
|
||
|
|
output_dir=args.output_dir,
|
||
|
|
num_train_epochs=args.num_train_epochs,
|
||
|
|
per_device_train_batch_size=args.per_device_train_batch_size,
|
||
|
|
learning_rate=args.learning_rate,
|
||
|
|
weight_decay=0.01,
|
||
|
|
logging_dir=f"{args.output_dir}/logs",
|
||
|
|
logging_steps=10,
|
||
|
|
save_strategy="epoch",
|
||
|
|
fp16=torch.cuda.is_available(),
|
||
|
|
)
|
||
|
|
|
||
|
|
# Set up data collator
|
||
|
|
data_collator = DataCollatorForLanguageModeling(
|
||
|
|
tokenizer=tokenizer,
|
||
|
|
mlm=False
|
||
|
|
)
|
||
|
|
|
||
|
|
# Initialize trainer
|
||
|
|
trainer = Trainer(
|
||
|
|
model=model,
|
||
|
|
args=training_args,
|
||
|
|
train_dataset=tokenized_dataset,
|
||
|
|
data_collator=data_collator,
|
||
|
|
)
|
||
|
|
|
||
|
|
# Train the model
|
||
|
|
print("Starting training")
|
||
|
|
trainer.train()
|
||
|
|
|
||
|
|
# Save the model
|
||
|
|
print(f"Saving model to {args.output_dir}")
|
||
|
|
trainer.save_model(args.output_dir)
|
||
|
|
tokenizer.save_pretrained(args.output_dir)
|
||
|
|
|
||
|
|
print("Finetuning complete!")
|
||
|
|
|
||
|
|
if __name__ == "__main__":
|
||
|
|
main()
|