test_llama4.py

import torch
from transformers import AutoProcessor, Llama4ForConditionalGeneration
import time
import os
from huggingface_hub import login
import requests
from PIL import Image
from io import BytesIO

# Print versions for debugging
import sys
print(f"Python version: {sys.version}")
print(f"PyTorch version: {torch.__version__}")
import transformers
print(f"Transformers version: {transformers.__version__}")

# Get token from environment
token = os.environ.get("HUGGINGFACE_TOKEN", "")
if token:
    print(f"Token found: {token[:5]}...")
else:
    print("No token found in environment variables!")

# Login to Hugging Face
try:
    login(token=token)
    print("Successfully logged in to Hugging Face Hub")
except Exception as e:
    print(f"Error logging in: {e}")

# Test 1: Simple text generation with Llama 4
def test_text_generation():
    print("\n=== Testing Text Generation ===")
    try:
        from transformers import AutoModelForCausalLM, AutoTokenizer
        
        model_id = "meta-llama/Llama-4-8B-Instruct"  # Using smaller model for faster testing
        
        print(f"Loading tokenizer from {model_id}...")
        tokenizer = AutoTokenizer.from_pretrained(model_id, token=token)
        
        print(f"Loading model from {model_id}...")
        model = AutoModelForCausalLM.from_pretrained(
            model_id,
            token=token,
            torch_dtype=torch.bfloat16,
            device_map="auto"
        )
        
        print("Model and tokenizer loaded successfully!")
        
        # Simple prompt
        prompt = "Write a short poem about artificial intelligence."
        
        print(f"Generating text for prompt: '{prompt}'")
        inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
        
        start_time = time.time()
        outputs = model.generate(**inputs, max_new_tokens=100)
        end_time = time.time()
        
        result = tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        print(f"Generation completed in {end_time - start_time:.2f} seconds")
        print(f"Result: {result}")
        return True
    except Exception as e:
        print(f"Error in text generation test: {e}")
        import traceback
        print(traceback.format_exc())
        return False

# Test 2: Image-text generation with Llama 4 Scout
def test_image_text_generation():
    print("\n=== Testing Image-Text Generation ===")
    try:
        model_id = "meta-llama/Llama-4-Scout-8B-16E-Instruct"  # Using smaller model for faster testing
        
        print(f"Loading processor from {model_id}...")
        processor = AutoProcessor.from_pretrained(model_id, token=token)
        
        print(f"Loading model from {model_id}...")
        model = Llama4ForConditionalGeneration.from_pretrained(
            model_id,
            token=token,
            torch_dtype=torch.bfloat16,
            device_map="auto"
        )
        
        print("Model and processor loaded successfully!")
        
        # Load a test image
        print("Loading test image...")
        response = requests.get("https://huggingface.co/datasets/huggingface/documentation-images/resolve/0052a70beed5bf71b92610a43a52df6d286cd5f3/diffusers/rabbit.jpg")
        img = Image.open(BytesIO(response.content))
        print(f"Image loaded: {img.size}")
        
        # Simple prompt
        prompt = "Describe this image in two sentences."
        
        print(f"Creating messages with prompt: '{prompt}'")
        messages = [
            {
                "role": "user",
                "content": [
                    {"type": "image", "url": "data:image/jpeg;base64," + BytesIO(response.content).getvalue().hex()},
                    {"type": "text", "text": prompt},
                ]
            },
        ]
        
        print("Applying chat template...")
        inputs = processor.apply_chat_template(
            messages,
            add_generation_prompt=True,
            tokenize=True,
            return_dict=True,
            return_tensors="pt",
        ).to(model.device)
        
        print("Generating response...")
        start_time = time.time()
        outputs = model.generate(**inputs, max_new_tokens=100)
        end_time = time.time()
        
        result = processor.batch_decode(outputs[:, inputs["input_ids"].shape[-1]:])[0]
        
        print(f"Generation completed in {end_time - start_time:.2f} seconds")
        print(f"Result: {result}")
        return True
    except Exception as e:
        print(f"Error in image-text generation test: {e}")
        import traceback
        print(traceback.format_exc())
        return False

if __name__ == "__main__":
    print("Starting Llama 4 tests...")
    
    # Run text generation test
    text_success = test_text_generation()
    
    # Run image-text generation test if text test succeeds
    if text_success:
        image_text_success = test_image_text_generation()
    else:
        print("Skipping image-text test due to text test failure")
        image_text_success = False
    
    # Summary
    print("\n=== Test Summary ===")
    print(f"Text Generation Test: {'SUCCESS' if text_success else 'FAILED'}")
    print(f"Image-Text Generation Test: {'SUCCESS' if image_text_success else 'FAILED'}")
    
    if text_success and image_text_success:
        print("\nAll tests passed! Your Llama 4 Scout setup is working correctly.")
    else:
        print("\nSome tests failed. Please check the error messages above.")