Phi3-LoRA-GSM8k

How to use :

!pip install peft accelerate bitsandbytes
from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer

# Function to generate and solve problems using the fine-tuned model
def generate_and_solve_problems(model, tokenizer, num_problems=5):
    """
    Generate and solve math and reasoning problems using the fine-tuned model.

    Parameters:
        model: Fine-tuned language model
        tokenizer: Corresponding tokenizer
        num_problems: Number of problems to generate and solve
    """
    # Prompt template
    test_prompt = """Below is a math problem. Solve the problem step by step and provide a detailed explanation.

### Problem:
{}

### Solution:"""

    # Sample test problems
    test_problems = [
        "A car travels at 40 mph for 2 hours, then at 60 mph for another 3 hours. How far does it travel in total?",
        "If the sum of three consecutive integers is 72, what are the integers?",
        "A train leaves Station A at 10:00 AM traveling at 50 mph. Another train leaves Station A at 12:00 PM traveling at 70 mph on the same track. At what time will the second train catch up to the first?",
        "A rectangle has a length of 12 units and a width of 8 units. If the length is increased by 50% and the width is reduced by 25%, what is the new area of the rectangle?",
        "If a person invests $1000 in a savings account that earns 5% annual interest compounded yearly, how much money will be in the account after 10 years?"
    ]

    # Use only the specified number of problems
    test_problems = test_problems[:num_problems]

    for problem in test_problems:
        # Create the prompt
        prompt = test_prompt.format(problem)

        # Tokenize and generate response
        inputs = tokenizer(prompt, return_tensors="pt", truncation=True, padding=True).to("cuda")
        outputs = model.generate(
            input_ids=inputs["input_ids"],
            attention_mask=inputs["attention_mask"],
            max_length=512,
            temperature=0.7,
            top_p=0.9,
            do_sample=True,
        )
        response = tokenizer.decode(outputs[0], skip_special_tokens=True)

        # Print the problem and the solution
        print(response)
        print("\n" + "="*50 + "\n")

# Example usage with model and tokenizer

base_model_name = "unsloth/phi-3-mini-4k-instruct-bnb-4bit"
lora_model_name = "Vijayendra/Phi3-LoRA-GSM8k"

# Load base model and tokenizer
base_model = AutoModelForCausalLM.from_pretrained(base_model_name, device_map="auto", torch_dtype="auto")
tokenizer = AutoTokenizer.from_pretrained(base_model_name)

# Load the fine-tuned LoRA model
model = PeftModel.from_pretrained(base_model, lora_model_name)
model.eval()

# Call the function to solve problems
generate_and_solve_problems(model, tokenizer)