app.py

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
import gradio as gr

# Load the model and tokenizer
model = AutoModelForCausalLM.from_pretrained(
    "qresearch/llama-3.1-8B-vision-378",
    trust_remote_code=True,
    torch_dtype=torch.float16,
).to("cuda")

tokenizer = AutoTokenizer.from_pretrained("qresearch/llama-3.1-8B-vision-378", use_fast=True)

# Define the function to process the image and instruction
def describe_image(image, instruction):
    description = model.answer_question(
        image, instruction, tokenizer, max_new_tokens=1000, do_sample=True, temperature=0.3
    )
    return description

# Create the Gradio interface
interface = gr.Interface(
    fn=describe_image,
    inputs=[
        gr.Image(type="pil"),  # Input for the image
        gr.Textbox(
            placeholder="Enter your instruction here...",
            label="Instruction",
            lines=10,  # Increase number of lines for instruction input
            max_lines=20  # Maximum number of lines for scrolling
        )  
    ],
    outputs=gr.Textbox(
        label="Description",
        lines=10,  # Increase number of lines for output
        max_lines=30  # Maximum number of lines for scrolling
    ),
    title="LLaMA 3.1 with Vision",
    description="Upload an image and enter an instruction to generate a description based on the provided instruction."
)

# Launch the Gradio app
interface.launch()