app.py

import os
import streamlit as st
from langchain.llms import HuggingFaceHub
from models import return_sum_models

class LLM_Langchain():
    def __init__(self):
        st.header('🦜 Code summarization')
        st.warning("Warning: input function needs cleaning and may take long to be processed at first time")
        st.warning("Note: you should not copy the whole function from IDE, the \"\\n\" character needs typing by hand")
        st.info("Reference: [CodeT5](https://arxiv.org/abs/2109.00859), [The Vault](https://arxiv.org/abs/2305.06156), [CodeXGLUE](https://arxiv.org/abs/2102.04664)")
        st.info("About me: namnh113")
        

        self.API_KEY = st.sidebar.text_input(
            'API key',
            type='password',
            help="Type in your HuggingFace API key to use this app")

            
        self.model_parent = st.sidebar.selectbox(
            label = "Choose language",
            options = ["python", "java", "javascript", "php", "ruby", "go", "cpp"],
            help="Choose languages",
            )

        if self.model_parent is None:
            model_name_visibility = True
        else:
            model_name_visibility = False
            
        model_name = return_sum_models(self.model_parent)
        list_model = [model_name]
        if self.model_parent in ["python", "java"]:
            list_model += [model_name+"_v2"]
        if self.model_parent != "cpp":
            list_model += ["Salesforce/codet5-base-multi-sum", f"Salesforce/codet5-base-codexglue-sum-{self.model_parent}"]
        
        self.checkpoint = st.sidebar.selectbox(
            label = "Choose model (namnh113/... is my model)",
            options = list_model,
            help="Model used to predict",
            disabled=model_name_visibility
            )

        self.max_new_tokens = st.sidebar.slider(
            label="Token Length",
            min_value=32,
            max_value=248,
            step=4,
            value=128,
            help="Set the max tokens to get accurate results"
            )

        self.num_beams = st.sidebar.slider(
            label="num beams",
            min_value=1,
            max_value=10,
            step=1,
            value=2,
            help="Set num beam"
            )
        
        self.top_k = st.sidebar.slider(
            label="top k",
            min_value=1,
            max_value=50,
            step=1,
            value=30,
            help="Set the top_k"
            )
        
        self.top_p = st.sidebar.slider(
            label="top p",
            min_value=0.1,
            max_value=1.0,
            step=0.05,
            value=0.95,
            help="Set the top_p"
            )


        self.model_kwargs = {
            "max_new_tokens": self.max_new_tokens,
            "top_k": self.top_k,
            "top_p": self.top_p,
            "num_beams": self.num_beams
        }

        os.environ['HUGGINGFACEHUB_API_TOKEN'] = self.API_KEY


    def generate_response(self, input_text):
        
        input_text = "Summarize " + self.model_parent.capitalize() + ": " + input_text
        llm = HuggingFaceHub(
            repo_id = self.checkpoint,
            model_kwargs = self.model_kwargs
        )

        return llm(input_text)
    
    

    def form_data(self):
        # with st.form('my_form'):
            try:
                if not self.API_KEY.startswith('hf_'):
                    st.warning('Please enter your API key!', icon='⚠')
                
                
                if "messages" not in st.session_state:
                    st.session_state.messages = []

                st.write(f"You are using {self.checkpoint} model")

                for message in st.session_state.messages:
                    with st.chat_message(message.get('role')):
                        st.write(message.get("content"))
                text = st.chat_input(disabled=False)
                
                if text:
                    st.session_state.messages.append(
                        {
                            "role":"user",
                            "content": text
                        }
                    )
                    with st.chat_message("user"):
                        st.write(text)
                    
                    if text.lower() == "clear":
                        del st.session_state.messages
                        return
                        
                    result = self.generate_response(text)
                    result = result.replace(' * ', '\n* ')
                    st.session_state.messages.append(
                        {
                            "role": "assistant",
                            "content": result
                        }
                    )
                    with st.chat_message('assistant'):
                        st.markdown(result)
                
            except Exception as e:
                st.error(e, icon="🚨")

model = LLM_Langchain()
model.form_data()