app.py

import gradio as gr

from langchain_huggingface import ChatHuggingFace, HuggingFacePipeline
from transformers import BitsAndBytesConfig
import re
from deep_translator import (GoogleTranslator,
                             PonsTranslator,
                             LingueeTranslator,
                             MyMemoryTranslator,
                             YandexTranslator,
                             DeeplTranslator,
                             QcriTranslator,
                             single_detection,
                             batch_detection)
from pyaspeller import YandexSpeller

def error_correct_pyspeller(sample_text):
    """ grammer correction of input text"""
    speller = YandexSpeller()
    fixed = speller.spelled(sample_text)
    return fixed

def postprocerssing(inp_text: str):
    """Post preocessing of the llm response"""
    inp_text = re.sub('<[^>]+>', '', inp_text)
    inp_text = inp_text.split('##', 1)[0]
    inp_text = error_correct_pyspeller(inp_text)
    return inp_text


quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype="float16",
    bnb_4bit_use_double_quant=True,
)

llm = HuggingFacePipeline.from_model_id(
    model_id="Danielrahmai1991/nvlm_adapt_basic_model_16bit",
    task="text-generation",
    pipeline_kwargs=dict(
        max_new_tokens=512,
        do_sample=True,
        repetition_penalty=1.15,
        trust_remote_code= True,
        temperature= 0.75

    ),
  model_kwargs={"quantization_config": quantization_config,
                },
    )

chat_model = ChatHuggingFace(llm=llm)

# history of the messages

def clear_memory(messages):
    messages.clear()
    return "Memory cleaned."
    
def llm_run(prompt, messages):
    print(f"question is {prompt}")
    lang = single_detection(prompt, api_key='4ab77f25578d450f0902fb42c66d5e11')
    if lang == 'en':
        prompt = error_correct_pyspeller(prompt)
    en_translated = GoogleTranslator(source='auto', target='en').translate(prompt)
    messages.append({"role": "user", "content": en_translated})
    ai_msg = chat_model.invoke(messages, skip_prompt = True)
    response_of_llm = postprocerssing(ai_msg.content)
    messages.append({"role": "assistant", "content": response_of_llm})
    response_of_llm = GoogleTranslator(source='auto', target=lang).translate(response_of_llm)
    print(f"out is: {response_of_llm}")
    return response_of_llm

# def greet(prompt, m_type):
#     return "hi"
print("donnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn")

with gr.Blocks(theme=gr.themes.Default(primary_hue=gr.themes.colors.orange, secondary_hue=gr.themes.colors.pink)) as demo:
    stored_message = gr.State([])
    with gr.Row():
        with gr.Column(scale=2):
            text1 = gr.Textbox(lines=7, label="Prompt", scale=2)
            with gr.Row():
                btn1 = gr.Button("Submit", scale=1)
                btn2 = gr.Button("Clear",  scale=1)
                btn3 = gr.Button("Clean Memory",  scale=2)
        with gr.Column(scale=2):
            out_text = gr.Text(lines=15, label="Output", scale=2)
    btn1.click(fn=llm_run, inputs=[text1, stored_message], outputs=out_text)
    btn2.click(lambda: [None, None], outputs=[text1, out_text])
    btn3.click(fn=clear_memory, inputs=[stored_message], outputs=[out_text])
    
# demo = gr.Interface(fn=llm_run, inputs=["text"], outputs="text")
demo.launch(debug=True, share=True)

# import gradio as gr

# def greet(inp, messages):
#     messages.append(inp)
#     print(messages)
#     return "Hello " + inp + "!"
    
# def clear_memory(messages):
#     messages.clear()
#     return "Memory cleaned."
    
# with gr.Blocks(theme=gr.themes.Default(primary_hue=gr.themes.colors.orange, secondary_hue=gr.themes.colors.pink)) as demo:
#     stored_message = gr.State([])
#     with gr.Row():
#         with gr.Column(scale=2):
#             text1 = gr.Textbox(lines=7, label="Prompt", scale=2)
#             with gr.Row():
#                 btn1 = gr.Button("Submit", scale=1)
#                 btn2 = gr.Button("Clear",  scale=1)
#                 btn3 = gr.Button("Clean Memory",  scale=2)
#         with gr.Column(scale=2):
#             out_text = gr.Text(lines=15, label="Output", scale=2)
#     btn1.click(fn=greet, inputs=[text1, stored_message], outputs=out_text)
#     btn2.click(lambda: [None, None], outputs=[text1, out_text])
#     btn3.click(fn=clear_memory, inputs=[stored_message], outputs=[out_text])
# demo.launch()