import requests
import torch
from PIL import Image
import gradio as gr
from transformers import AutoModelForCausalLM, AutoProcessor

# Load model and processor
model_id_or_path = "rhymes-ai/Aria"
model = AutoModelForCausalLM.from_pretrained(model_id_or_path, device_map="auto", torch_dtype=torch.bfloat16, trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_id_or_path, trust_remote_code=True)

# Function to process the input and generate text
def generate_response(image):
    # Convert the input image to PIL format (if necessary)
    if isinstance(image, str):
        image = Image.open(requests.get(image, stream=True).raw)
    
    # Prepare messages for the model
    messages = [
        {
            "role": "user",
            "content": [
                {"text": None, "type": "image"},
                {"text": "what is the image?", "type": "text"},
            ],
        }
    ]

    text = processor.apply_chat_template(messages, add_generation_prompt=True)
    inputs = processor(text=text, images=image, return_tensors="pt")
    
    # Move pixel values to the correct dtype
    inputs["pixel_values"] = inputs["pixel_values"].to(model.dtype)
    inputs = {k: v.to(model.device) for k, v in inputs.items()}

    # Generate response
    with torch.inference_mode(), torch.cuda.amp.autocast(dtype=torch.bfloat16):
        output = model.generate(
            **inputs,
            max_new_tokens=500,
            stop_strings=["<|im_end|>"],
            tokenizer=processor.tokenizer,
            do_sample=True,
            temperature=0.9,
        )
        output_ids = output[0][inputs["input_ids"].shape[1]:]
        result = processor.decode(output_ids, skip_special_tokens=True)

    return result

# Gradio interface
iface = gr.Interface(
    fn=generate_response,
    inputs=gr.inputs.Image(type="filepath"),
    outputs="text",
    title="Image-to-Text Model",
    description="Upload an image, and the model will describe it.",
)

# Launch the app
iface.launch()