app.py

import torch
from transformers import AutoTokenizer, AutoProcessor, TrainingArguments, LlavaForConditionalGeneration, BitsAndBytesConfig
from trl import SFTTrainer
from peft import LoraConfig, PeftModel
from PIL import Image
import requests
from deep_translator import GoogleTranslator
import gradio as gr
import PIL.Image
import base64
import time
import os


model_id = "HuggingFaceH4/vsft-llava-1.5-7b-hf-trl"
quantization_config = BitsAndBytesConfig(load_in_4bit=True)
base_model = LlavaForConditionalGeneration.from_pretrained(model_id, quantization_config=quantization_config, torch_dtype=torch.float16)

# Load the PEFT Lora adapter
peft_lora_adapter_path = "Praveen0309/llava-1.5-7b-hf-ft-mix-vsft-3"
peft_lora_adapter = PeftModel.from_pretrained(base_model, peft_lora_adapter_path, adapter_name="lora_adapter")
base_model.load_adapter(peft_lora_adapter_path, adapter_name="lora_adapter")

processor = AutoProcessor.from_pretrained("HuggingFaceH4/vsft-llava-1.5-7b-hf-trl")

# Function to translate text from Bengali to English
def deep_translator_bn_en(input_sentence):
    english_translation = GoogleTranslator(source="bn", target="en").translate(input_sentence)
    return english_translation

# Function to translate text from English to Bengali
def deep_translator_en_bn(input_sentence):
    bengali_translation = GoogleTranslator(source="en", target="bn").translate(input_sentence)
    return bengali_translation

def inference(image, image_prompt):
    prompt = f"USER: <image>\n{image_prompt} ASSISTANT:"

    # Assuming your model can handle PIL images
    image = image.convert("RGB")  # Ensure image is RGB mode

    inputs = processor(text=prompt, images=image, return_tensors="pt")
    generate_ids = base_model.generate(**inputs, max_new_tokens=15)
    decoded_response = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
    return decoded_response

def image_to_base64(image_path):
    with open(image_path, 'rb') as img:
        encoded_string = base64.b64encode(img.read())
    return encoded_string.decode('utf-8')

# Function that takes User Inputs and displays it on ChatUI
def query_message(history,txt,img):
    image_prompt = deep_translator_bn_en(txt)
    history += [(image_prompt,None)]
    base64 = image_to_base64(img)
    data_url = f"data:image/jpeg;base64,{base64}"
    history += [(f"{image_prompt} ![]({data_url})", None)]
    return history

# Function that takes User Inputs, generates Response and displays on Chat UI
def llm_response(history,text,img):
        image_prompt = deep_translator_bn_en(text)
        response = inference(img,image_prompt)
        assistant_index = response.find("ASSISTANT:")
        extracted_string = response[assistant_index + len("ASSISTANT:"):].strip()
        output = deep_translator_en_bn(extracted_string)
        history += [(text,output)]
        return history

# Interface Code
with gr.Blocks() as app:
    with gr.Row():
        image_box = gr.Image(type="pil")

        chatbot = gr.Chatbot(
            scale = 2,
            height=500
        )
    text_box = gr.Textbox(
            placeholder="Enter text and press enter, or upload an image",
            container=False,
        )

    btn = gr.Button("Submit")
    clicked = btn.click(query_message,
                        [chatbot,text_box,image_box],
                        chatbot
                        ).then(llm_response,
                                [chatbot,text_box,image_box],
                                chatbot
                                )
app.queue()
app.launch(debug=True,share=True)