fine_tune_sakhi.py

from transformers import GPT2LMHeadModel, GPT2Tokenizer, Trainer, TrainingArguments
from datasets import load_dataset

# Load dataset
dataset = load_dataset("json", data_files="fine_tuning_dataset.jsonl")

# Load model and tokenizer
model = GPT2LMHeadModel.from_pretrained("gpt2")
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
tokenizer.pad_token = tokenizer.eos_token

# Tokenize data
def preprocess_data(examples):
    return tokenizer(examples["prompt"], truncation=True, padding="max_length", max_length=50)

tokenized_dataset = dataset.map(preprocess_data, batched=True)

# Training arguments
training_args = TrainingArguments(
    output_dir="./sakhi_finetuned",
    overwrite_output_dir=True,
    num_train_epochs=3,
    per_device_train_batch_size=4,
    save_steps=500,
    save_total_limit=1,
    logging_dir="./logs",
)

# Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=tokenized_dataset["train"],
)

# Train
trainer.train()
model.save_pretrained("./sakhi_finetuned")
tokenizer.save_pretrained("./sakhi_finetuned")