backend/scripts/verify_hf.py

import sys
import os
import time
from huggingface_hub import InferenceClient

# Add the project root to sys.path
sys.path.append(os.path.abspath(os.path.join(os.path.dirname(__file__), '..', '..')))

from backend.app.config import config

def test_hf_inference():
    print("Testing Hugging Face Inference...")
    print(f"Model: {config.LLM_MODEL}")
    print(f"API Key Present: {'Yes' if config.HUGGINGFACE_API_KEY else 'No'}")
    
    # Test 1: Intended Model (Llama 3)
    print(f"\n--- Testing {config.LLM_MODEL} ---")
    # Use standard Inference API URL for better compatibility checks
    # model_url = f"https://api-inference.huggingface.co/models/{config.LLM_MODEL}"
    # Sticking to what the app uses to be sure
    model_url = f"https://router.huggingface.co/hf-inference/models/{config.LLM_MODEL}"
    
    client = InferenceClient(model=model_url, token=config.HUGGINGFACE_API_KEY)
    prompt = "Hello!"
    
    try:
        response = client.text_generation(prompt, max_new_tokens=10)
        print("Success!")
    except Exception as e:
        print(f"FAILED with {config.LLM_MODEL}")
        if hasattr(e, 'response'):
            print(f"Status: {e.response.status_code}")
            print(f"Body: {e.response.text}")
        else:
            print(f"Error: {e}")

    # Test 2: Fallback Non-Gated Model (Zephyr 7B Beta) - Standard Endpoint
    fallback_model = "HuggingFaceH4/zephyr-7b-beta"
    print(f"\n--- Testing Fallback {fallback_model} (Standard API) ---")
    model_url_fallback = f"https://api-inference.huggingface.co/models/{fallback_model}"
    
    # Try with Token first
    client_fallback = InferenceClient(model=model_url_fallback, token=config.HUGGINGFACE_API_KEY)
    
    try:
        response = client_fallback.text_generation(prompt, max_new_tokens=10)
        print(f"Success with Token! Generated: {response}")
    except Exception as e:
        print(f"FAILED with Token: {e}")

    # Test 3: Library Managed URL (Zephyr)
    print(f"\n--- Testing Zephyr (Library Managed URL) ---")
    try:
        # Let the library decide the URL
        client_lib = InferenceClient(model="HuggingFaceH4/zephyr-7b-beta", token=config.HUGGINGFACE_API_KEY)
        response = client_lib.text_generation(prompt, max_new_tokens=10)
        print(f"Success (with token)! Generated: {response}")
    except Exception as e:
        print(f"FAILED (with token): {e}")
        
    # Test 5: GPT-2 (Anonymous)
    print(f"\n--- Testing GPT-2 (Anonymous) ---")
    try:
        client_gpt2 = InferenceClient(model="gpt2", token=None)
        response = client_gpt2.text_generation("Hello world", max_new_tokens=10)
        print(f"Success GPT-2! Generated: {response}")
    except Exception as e:
        print(f"FAILED GPT-2: {e}")

if __name__ == "__main__":
    test_hf_inference()