app.py

from langchain.schema import AIMessage, HumanMessage
from langchain_core.messages import HumanMessage
from langchain_google_genai import ChatGoogleGenerativeAI
import gradio as gr
import os 
from PIL import Image
import time



GOOGLE_AI_STUDIO = 'AIzaSyD9f8u_jY3BB2awtocA6E-XEoyi4hMexRE'

# Setup the language model
llm_img = ChatGoogleGenerativeAI(
    model="gemini-pro-vision",
    temperature=0.0,
    max_output_tokens=2048,
    google_api_key=GOOGLE_AI_STUDIO
)
llm = ChatGoogleGenerativeAI(
    model="gemini-pro",
    temperature=0.0,
    max_output_tokens=2048,
    google_api_key=GOOGLE_AI_STUDIO,
    stream=True
)


def predict(message, history, img):
    history_langchain_format = []
    
    for human, assistant in history:
        history_langchain_format.append(HumanMessage(content=human))
        history_langchain_format.append(AIMessage(content=assistant))
        
    if img is None:
        history_langchain_format.append(HumanMessage(content=message))
        gpt_response = llm.invoke(history_langchain_format)
        ans = ''
        for response in gpt_response.content:
            ans += response
            time.sleep(0.001)
            yield ans
    else:
        msg = HumanMessage(
            content=[
                {"type": "text", "text": message},
                {"type": "image_url", "image_url": img},
            ]
        )
        history_langchain_format.append(msg)
        
        gpt_response = llm_img.invoke([msg])
        ans = ''
        for response in gpt_response.content:
            ans += response
            time.sleep(0.001)
            yield ans
        
interface = gr.ChatInterface(predict, additional_inputs=gr.Image(type='pil', label='Upload your image...'))
interface.launch(interface.queue())