scripts/test_generation.py

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

# Define your model path
model_path = "./llama3-5b/hf"  # or the path/model_name you have

# Your custom quantization configuration
quantization_config = None

# Load the model and tokenizer
model = AutoModelForCausalLM.from_pretrained(model_path,  
                                             device_map="auto", 
                                             quantization_config=quantization_config, 
                                             output_hidden_states=True)
tokenizer = AutoTokenizer.from_pretrained(model_path)

# Initialize the messages list with a generic short system message
messages = [
    {"role": "system", "content": "You are a helpful assistant."}
]

# Function to generate a response
def generate_response(messages):
    input_ids = tokenizer.apply_chat_template(
        messages,
        add_generation_prompt=True,
        return_tensors="pt"
    ).to(model.device)

    terminators = [
        tokenizer.eos_token_id,
        tokenizer.convert_tokens_to_ids("<|eot_id|>")
    ]

    outputs = model.generate(
        input_ids,
        max_new_tokens=256,
        eos_token_id=terminators,
        do_sample=True,
        temperature=0.6,
        top_p=0.9,
    )
    response = outputs[0][input_ids.shape[-1]:]
    return tokenizer.decode(response, skip_special_tokens=True)

# Interactive loop
while True:
    # Get user input
    user_input = input("User: ")
    
    # Check if the user wants to quit
    if user_input.lower() == 'q':
        break
    
    # Update the messages list with the user input
    messages.append({"role": "user", "content": user_input})
    
    # Generate a response based on the updated messages
    response = generate_response(messages)
    print("Assistant:", response)
    
    # Update the messages list with the generated response
    messages.append({"role": "assistant", "content": response})