app.py

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

# Define constants
MODEL_NAME = "microsoft/Phi-3.5-vision-instruct"
DESCRIPTION = "# [Phi-3.5-vision Demo](https://huggingface.co/microsoft/Phi-3.5-vision-instruct)"
DEVICE = "cuda"

# Load model and processor
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME, trust_remote_code=True, torch_dtype="auto", _attn_implementation="flash_attention_2").to(DEVICE).eval()
processor = AutoProcessor.from_pretrained(MODEL_NAME, trust_remote_code=True)

def run_example(image, text_input, model_id):
    # Prepare prompt and image for processing
    prompt = f"{text_input}\n"
    image = Image.fromarray(image).convert("RGB")
    
    # Process input
    inputs = processor(prompt, image, return_tensors="pt").to(DEVICE)
    generate_ids = model.generate(**inputs, max_new_tokens=1000, eos_token_id=processor.tokenizer.eos_token_id)
    generate_ids = generate_ids[:, inputs['input_ids'].shape[1]:]
    response = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
    
    return response

css = """
  #output {
    height: 500px; 
    overflow: auto; 
    border: 1px solid #ccc; 
  }
"""

# Set up the Gradio interface
with gr.Blocks(css=css) as demo:
    gr.Markdown(DESCRIPTION)
    with gr.Tab(label="Phi-3.5 Input"):
        with gr.Row():
            with gr.Column():
                input_img = gr.Image(label="Input Picture")
                text_input = gr.Textbox(label="Question")
                submit_btn = gr.Button(value="Submit")
            with gr.Column():
                output_text = gr.Textbox(label="Output Text")
        submit_btn.click(run_example, inputs=[input_img, text_input, MODEL_NAME], outputs=output_text)

demo.launch(debug=True)