app.py

import spaces
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import os
import json
import requests
from openai import OpenAI

HF_API_KEY = os.getenv("HF_API_KEY")
MODEL_ID = "meta-llama/Llama-Guard-3-1B"

model = AutoModelForCausalLM.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.bfloat16,
    device_map="auto",
    token=HF_API_KEY
).to('cuda')
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, token=HF_API_KEY).to('cuda')


# Model 1: Llama Guard Model
@spaces.GPU
def llama_guard_moderation(input_text):
    conversation = [
        {
            "role": "user",
            "content": [
                {
                    "type": "text", 
                    "text": input_text
                },
            ],
        }
    ]
    input_ids = tokenizer.apply_chat_template(
        conversation, return_tensors="pt"
    ).to(model.device)

    prompt_len = input_ids.shape[1]
    output = model.generate(
        input_ids,
        max_new_tokens=20,
        pad_token_id=0,
    )
    generated_tokens = output[:, prompt_len:]
    return tokenizer.decode(generated_tokens[0])


# Model 2: OpenAI Omni Moderation
def openai_moderation(input_text):
    client = OpenAI()
    response = client.moderations.create(
        model="omni-moderation-latest",
        input=input_text,
    )
    return response.results[0].categories


# Model 3: Sentinel API for LionGuard
def sentinel_moderation(input_text):
    api_key = os.getenv("SENTINEL_API_KEY")
    api_endpoint = os.getenv("SENTINEL_ENDPOINT")
    
    headers = {
        "x-api-key": api_key,
        "Content-Type": "application/json"
    }
    
    payload = {
        "filters": ["lionguard"],
        "text": input_text
    }
    
    response = requests.post(
        url=api_endpoint,
        headers=headers,
        data=json.dumps(payload)
    )
    
    return response.json()["outputs"]["lionguard"]


# Gradio App
def moderate_text(input_text):
    llama_guard_result = llama_guard_moderation(input_text)
    openai_result = openai_moderation(input_text)
    sentinel_result = sentinel_moderation(input_text)

    return llama_guard_result, openai_result, sentinel_result


iface = gr.Interface(
    fn=moderate_text,
    inputs=gr.Textbox(lines=5, label="Enter Text for Moderation"),
    outputs=[
        gr.Textbox(label="Llama Guard Result"),
        gr.Textbox(label="OpenAI Omni Moderation Result"),
        gr.Textbox(label="Sentinel LionGuard Result"),
    ],
    title="Content Moderation Model Comparison",
    description="Compare the performance of 3 content moderation models: Llama Guard, OpenAI Omni Moderation, and Sentinel LionGuard."
)

iface.launch()