finetune_tinyllama.py

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments, Trainer, DataCollatorForSeq2Seq
from datasets import load_dataset

# Model name
MODEL_NAME = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    torch_dtype=torch.float16,  # Use float16 for better efficiency
    device_map="auto"  # Use GPU if available
)

# Load dataset from JSON file
dataset = load_dataset("json", data_files="processed_dataset.json")

# Tokenization function
def tokenize_function(examples):
    return tokenizer(examples["prompt"], examples["response"], padding="max_length", truncation=True)

# Apply tokenization
dataset = dataset.map(tokenize_function, batched=True)
dataset = dataset.remove_columns(["prompt", "response"])  # Keep only tokenized data

# Data collator (for batching and padding)
data_collator = DataCollatorForSeq2Seq(
    tokenizer=tokenizer,
    model=model,
    padding=True,
    return_tensors="pt"
)

# Training arguments
training_args = TrainingArguments(
    output_dir="./results",
    num_train_epochs=3,
    per_device_train_batch_size=4,
    per_device_eval_batch_size=4,
    save_steps=10_000,
    save_total_limit=2,
    logging_dir="./logs",
    logging_steps=200,
    remove_unused_columns=False,  # Ensure tokenized data isn't removed
    fp16=True,  # Enable mixed precision if using GPU
)

# Trainer setup
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=dataset["train"],
    data_collator=data_collator,
)

# Start training
trainer.train()