app.py

import requests
import gradio as gr
from dotenv import load_dotenv
import os
#from openai import OpenAI
from langchain_openai import OpenAI
import spacy
#from langchain.chat_models import ChatOpenAI
from langchain_openai import ChatOpenAI
from langchain.schema import AIMessage, HumanMessage
import pandas as pd
import uuid
import json

# Load environment variables from .env file
load_dotenv()

# Access the env
HF_TOKEN = os.getenv('HUGGING_FACE_TOKEN')

GITHUB_TOKEN = "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" # move to env

# openai setup
# client = OpenAI(
#   api_key=os.getenv('OPENAI_API_KEY')
# )

# hugging face setup
#model_name = "mmnga/ELYZA-japanese-Llama-2-7b-instruct-gguf"
API_URL = f"https://api-inference.huggingface.co/models/"
#API_URL = f"https://api-inference.huggingface.co/models/{model_name}"
headers = {"Authorization": f"Bearer {HF_TOKEN}"}

# Global variable to control debug printing
DEBUG_MODE = True


def share_to_gist(content, public=False):
    url = "https://api.github.com/gists"
    headers = {
        "Authorization": f"token {os.getenv(GITHUB_TOKEN)}",
        "Accept": "application/vnd.github.v3+json",
    }
    data = {
        "public": public,
        "description": "Chat history",
        "files": {
            "chat.txt": {
                "content": content
            }
        }
    }
    response = requests.post(url, headers=headers, data=json.dumps(data))
    gist_url = response.json().get('html_url', '')
    return gist_url

def generate_unique_id():
    return str(uuid.uuid4())

def debug_print(*args, **kwargs):
    if DEBUG_MODE:
        print(*args, **kwargs)

def split_sentences_ginza(input_text):
    nlp = spacy.load("ja_core_news_sm")
    doc = nlp(input_text)
    sentences = [sent.text for sent in doc.sents]
    return sentences


file_path = 'anki_japanese_english_pairs.csv'

def load_csv(file_path):
    # Load the CSV file into a DataFrame
    df = pd.read_csv(file_path)
    
    return df

def get_sentence_pair(df):

    # Get a random row from the DataFrame
    random_row = df.sample(1)
    #debug_print("### random_row:", random_row)

    #print(random_row.shape)

    japanese_sentence = str(random_row.iloc[0, 0])
    english_sentence = str(random_row.iloc[0,  1])

    debug_print("### Japanese sentence:", japanese_sentence)
    debug_print("### English sentence:", english_sentence)


    return japanese_sentence, english_sentence


japanese_sentence, english_sentence = get_sentence_pair(load_csv(file_path))


llm = ChatOpenAI(temperature=0.7, model='gpt-3.5-turbo')

def predict(message, history):

    # Define your initial setup prompt here
    initial_setup = f'''
    Japanese students are learning to translate Japanese text to English text. They will be given a Japanese sentence to translate, and will provide an English translation attempt. 
    Based on the feedback you provide, they will revise their translation. This process will continue until their translation is accurate. 

    Encourage the student by specifying the strengths of their writing.
    DO NOT PROVIDE THE CORRECT ENGLISH TRANSLATION until the student gets the correct translation. Let the student work it out.
    Provide your feedback as a list in the format: a, b, c etc.
    Do not respond in Japanese - always respond in English even if the student uses Japanese with you.
    
    Execute the following tasks step by step:
    1. Ask the student to translate the following sentence from Japanese to English: {japanese_sentence}. Here is the English translation for reference: {english_sentence}
    2. Suggest only mechanical corrections (i.e., spelling, grammar, and punctuation) for the student. Ask for another translation attempt.

    Start by asking the student to translate the Japanese sentence.
    '''

    # removed from prompt
    # The student's translation need not match the provided English translation exactly, but it should be accurate to the Japanese text.

    # Start your history with a SystemMessage containing the setup prompt
    history_langchain_format = [AIMessage(content=initial_setup)]
    #history_langchain_format.append(HumanMessage(content="Let's start."))


    for human, ai in history:
        if human is not None:
            history_langchain_format.append(HumanMessage(content=human)) # convert to str to avoid error; not compatible with multimodal
        if ai is not None:
            history_langchain_format.append(AIMessage(content=ai))

    history_langchain_format.append(HumanMessage(content=message))

    #debug_print("### Full history: ", history_langchain_format)
    gpt_response = llm(history_langchain_format)
    return  gpt_response.content

welcome_message = "Hi! 👋. Are you ready to practise translation?"

# with gr.Blocks() as app:
#     chatbot = gr.Chatbot()
#     message = gr.Textbox()
#     clear = gr.ClearButton([message, chatbot])
#     message.submit(predict, [message, chatbot], [message, chatbot])

app = gr.ChatInterface(fn=predict, title="Translation Chatbot", chatbot=gr.Chatbot(value=[(None, welcome_message)],),)#, multimodal=True) # chatbot=gr.Chatbot(value=[["Welcome 👋. I am an assistant",]])



app.launch()