chatbot-GPT/main.py

import io
import os
import json

import requests
from openai import OpenAI
from flask import Flask, request, jsonify, send_from_directory, url_for

from convert import alpaca_to_chatgpt, csv_to_jsonl

app = Flask(__name__)
ssl = None
# ssl =('/etc/ssl/sample.crt', '/etc/ssl/sample.pem')

app.openai_key = os.environ.get("OPENAI_KEY", "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")
app.openai_client = OpenAI(api_key=app.openai_key)

app.chat_messages = [
    {"role": "system",
     "content": "Please respond professionally and in a friendly manner, using the same language as the original request."}
]
app.translate_messages = [
    {"role": "system",
     "content": "Please translate using the requested language."}
]
app.suggest_messages = [
    {"role": "system",
     "content": "Please suggest reply messages based on the previous conversations and the user's request."}
]
app.recommend_messages = [
    {"role": "system",
     "content": "Give normalized total weight of each category in json based on headlines"
     }
]
app.summary_messages = [
    {"role": "system",
     "content": "Please summarize an article."
     }
]
UPLOAD_FOLDER = 'files'
app.config['UPLOAD_FOLDER'] = UPLOAD_FOLDER


@app.route('/files/<name>')
def download_file(name):
    return send_from_directory(app.config["UPLOAD_FOLDER"], name)


@app.route('/', methods=['GET', 'POST'])
def test():
    return jsonify({"status": "0"})


def recommend(headlines, category):
    chat_messages = app.recommend_messages.copy()
    try:
        json_payload = {
            "role": "user",
            "content": f"""{headlines}
                                    Berikan nilai berat masing-masing kategori, jumlahkan dan normalisasikan:
                                    {category}
                                    Berikan dalam bentuk json
                                    """
        }
        chat_messages.append(json_payload)
        print(chat_messages)
        json_response = app.openai_client.chat.completions.create(model="gpt-3.5-turbo-1106",
                                                                  messages=chat_messages,
                                                                  response_format={"type": "json_object"}
                                                                  )
        print(json_response.choices[0].message.content)
        return json.loads(json_response.choices[0].message.content)
    except Exception as error_print:
        app.logger.error(error_print)
        result = {}, 405


def vision(message, image_url=None, image_b64=None):
    chat_messages = app.chat_messages.copy()
    url = ""
    if image_url:
        url = f"{image_url}"
    elif image_b64:
        url = f"data:image/jpeg;base64,{image_b64}"
    try:
        json_payload = {
            "role": "user",
            "content": [
                {"type": "text", "text": message},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": url,
                    },
                },
            ],
        }
        chat_messages.append(json_payload)
        print(chat_messages)
        json_response = app.openai_client.chat.completions.create(
            model="gpt-4o",
            messages=chat_messages,
            max_tokens=500
        )
        return {"role": "assistant", "content": json_response.choices[0].message.content}
    except Exception as error_print:
        app.logger.error(error_print)
        result = {}, 405


@app.route('/gpt', methods=['POST'])
def gpt():
    chat_messages = app.chat_messages.copy()
    chat_model = "gpt-3.5-turbo"
    use_video = False
    suggest = False
    summarize = False
    predict_q = 0
    max_char_msg = 500
    max_resp_token = 600
    category = []
    headlines = []
    image_url = ""
    num_choices = 1
    json_payload = request.get_json()
    if not json_payload:
        json_payload = []
    has_named_params = False
    if isinstance(json_payload, dict):
        has_named_params = 'payload' in json_payload
        if 'payload' in json_payload:
            if 'predict_q' in json_payload:
                predict_q = 5 if json_payload['predict_q'] > 5 else 0 if json_payload['predict_q'] < 0 else json_payload['predict_q']
            if 'num_choices' in json_payload:
                num_choices = 5 if json_payload['num_choices'] > 5 else 1 if json_payload['num_choices'] < 1 else json_payload['num_choices']
            if 'use_video' in json_payload:
                use_video = json_payload['use_video'] == "1"
            if 'chat_model' in json_payload:
                chat_model = json_payload['chat_model']
                max_resp_token = 2048
            if 'translate' in json_payload:
                chat_messages = app.translate_messages.copy()
                json_payload['payload'][-1]['content'] = json_payload['payload'][-1][
                                                             'content'] + f" (Translate to {json_payload['translate']})"
            elif 'suggest' in json_payload:
                suggest = json_payload['suggest'] == "1"
                if suggest:
                    chat_messages = app.suggest_messages.copy()
                else:
                    chat_messages = app.chat_messages.copy()
                json_payload['payload'][-1]['content'] = json_payload['payload'][-1][
                                                             'content'] + f" What can I say to him/her?"
            elif 'summarize' in json_payload:
                summarize = json_payload['summarize'] == "1"
                if summarize:
                    chat_messages = app.summary_messages.copy()
                    max_char_msg = 2000
                    max_resp_token = 1000
                else:
                    chat_messages = app.chat_messages.copy()
                json_payload['payload'][-1]['content'] = f"Please summarize this article:\n" + \
                                                         json_payload['payload'][-1]['content']
            else:
                chat_messages = app.chat_messages.copy()
            json_payload = json_payload['payload']
            if isinstance(json_payload, dict):
                json_payload = [json_payload]
        elif 'greeting' in json_payload:
            chat_messages = app.chat_messages.copy()
            company_name = json_payload['greeting']['company_name']
            timestamp = json_payload['greeting']['timestamp']
            islamic_message = f"Apakah Nama '{company_name}' terdapat unsur islami? Jawab dengan 'Ya' atau 'Tidak'"
            islam_messages = app.chat_messages.copy()
            islam_messages.append({
                "role": "user",
                "content": islamic_message
            })
            islamic_response = app.openai_client.chat.completions.create(model="gpt-3.5-turbo",  # GPT-3.5 Turbo engine
                                                                         messages=islam_messages,
                                                                         max_tokens=2, temperature=0.5)
            if 'Ya' in islamic_response.choices[0].message.content:
                greeting_message = f"Buatkan respons chatbot berupa greeting dari chat perusahaan bernama {company_name} pada jam {timestamp}, tidak perlu mention waktu, dan jawab dengan 'Assalamu'alaikum...' terlebih dahulu"
            else:
                greeting_message = f"Buatkan respons chatbot berupa greeting dari chat perusahaan bernama {company_name} pada jam {timestamp}, tidak perlu mention waktu"
            json_payload = [
                {
                    "role": "user",
                    "content": greeting_message
                }
            ]
        elif 'recommend' in json_payload:
            headlines = json_payload['recommend']['headlines']
            category = json_payload['recommend']['category']
            return recommend(headlines, category)
        elif 'image_url' in json_payload:
            image = json_payload['image_url']
            message = json_payload["message"] if 'message' in json_payload else "Ini gambar apa?"
            return vision(message, image_url=image)
        elif 'image_b64' in json_payload:
            image = json_payload['image_b64']
            message = json_payload["message"] if 'message' in json_payload else "Ini gambar apa?"
            return vision(message, image_b64=image_url)
        else:
            chat_messages = app.chat_messages.copy()
            json_payload = [json_payload]
    json_payload = json_payload[-5:]
    for message in json_payload:
        if message['role'] == 'user':
            content = message['content'].lower()
        else:
            content = message['content']
        content_arr = content.split(" ")
        new_content_arr = content[:max_char_msg].split(" ")
        new_content_len = len(new_content_arr)
        arr = []
        for i in range(new_content_len):
            arr.append(content_arr[i])
        message['content'] = " ".join(arr)
        chat_messages.append(message)
    app.logger.info(chat_messages)
    result = {}
    try:
        n = num_choices
        json_response = app.openai_client.chat.completions.create(model=chat_model,
                                                                  messages=chat_messages,
                                                                  max_tokens=max_resp_token, temperature=0.7, n=n)
        app.logger.info(json_response.choices[0].message)
        if has_named_params:
            if suggest:
                choices = json_response.choices
                messages = [i.message for i in choices]
                json_formatted = []
                for message in messages:
                    json_formatted.append({"role": "assistant", "content": message.content})
                result = {"url": "", "message": json_formatted}
            else:
                if use_video:
                    # TODO: to be implemented
                    result = {"url": url_for('download_file', name="test.mp4", _external=True),
                              "message": {"role": "assistant", "content": json_response.choices[0].message.content}}
                else:
                    result = {"role": "assistant", "content": json_response.choices[0].message.content}
                if predict_q:
                    query_q = {
                        "role": "user",
                        "content": f"Berikan {predict_q} pertanyaan lain yang akan saya ajukan berdasarkan percakapan kali ini dalam bentuk json array"
                    }
                    chat_messages.append(result)
                    chat_messages.append(query_q)
                    json_response_q = app.openai_client.chat.completions.create(model=chat_model,
                                                                              messages=chat_messages,
                                                                              max_tokens=max_resp_token,
                                                                              temperature=0.2, response_format={"type": "json_object"})
                    json_response_dict = json.loads(json_response_q.choices[0].message.content)
                    print(json_response_dict)
                    if json_response_dict is not None:
                        if isinstance(json_response_dict, dict):
                            first_key = next(iter(json_response_dict))
                            json_response_dict = json_response_dict[first_key]
                        elif isinstance(json_response_dict, str):
                            json_response_dict = [json_response_dict]
                        result["predict_q"] = json_response_dict
        else:
            result = {"role": "assistant", "content": json_response.choices[0].message.content}
    except Exception as error_print:
        app.logger.error(error_print)
        result = {}, 405
    return result


@app.route('/train', methods=['POST'])
def train():
    prev_model = "gpt-3.5-turbo"
    if 'job_id' in request.form:
        return train_with_id(job_id=request.form['job_id'])
    elif 'train_file' in request.files:
        train_file = request.files['train_file']
        app.logger.info({"filename": train_file.filename})
        openai_file = None
        if train_file.filename.split('.')[1] == 'jsonl':
            openai_file = train_file.stream.read()
        elif train_file.filename.split('.')[1] == 'csv':
            openai_file = csv_to_jsonl(train_file.stream.read())
        elif train_file.filename.split('.')[1] == 'json':
            openai_file = alpaca_to_chatgpt(train_file)
        if 'prev_model' in request.form:
            prev_model = request.form['prev_model']
        app.logger.info(f"Previous model: {prev_model}")
        if 'mock' not in request.form:
            f = app.openai_client.files.create(
                file=openai_file,
                purpose="fine-tune"
            )
            job = app.openai_client.fine_tuning.jobs.create(
                training_file=f.id,
                model=prev_model,
                hyperparameters={
                    "n_epochs": 5
                }
            )
            app.logger.info({"mock": "no", "status": job.status, "job_id": job.id})
            return {"status": job.status, "job_id": job.id}
        else:
            app.logger.info({"mock": "yes", "status": "ok"})
            return {"status": "ok"}
    else:
        app.logger.error({"status": "error", "message": "Training file not found"})
        return {"status": "error", "message": "Training file not found"}

def train_with_id(job_id):
    try:
        job = app.openai_client.fine_tuning.jobs.retrieve(job_id)
        if job.fine_tuned_model is None:
            app.logger.info({"job_id": job_id, "status": job.status})
            return {"status": job.status}
        else:
            app.logger.info({"job_id": job_id, "status": job.status, "model_name": job.fine_tuned_model})
            return {"status": job.status, "model_name": job.fine_tuned_model}
    except Exception as error_print:
        app.logger.error(error_print)
        return {"status": "Could not find job from id"}

@app.route('/llama', methods=['POST'])
def llama():
    max_char_msg = 500
    max_resp_token = 600
    json_payload = request.get_json()
    if not json_payload:
        json_payload = []
    has_named_params = False
    if isinstance(json_payload, dict):
        has_named_params = 'payload' in json_payload
        if 'payload' in json_payload:
            json_payload = json_payload['payload']
            if isinstance(json_payload, dict):
                json_payload = [json_payload]
        else:
            json_payload = [json_payload]
    message = json_payload[-1]
    content = message['content']
    content_arr = content.split(" ")
    new_content_arr = content[:max_char_msg].split(" ")
    new_content_len = len(new_content_arr)
    arr = []
    for i in range(new_content_len):
        arr.append(content_arr[i])
    content = " ".join(arr)
    content = content + " Jawab dengan Bahasa Indonesia"
    try:
        json_request = {
            "model": "llama3",
            "prompt": content,
            "stream": False
        }
        r = requests.post("http://localhost:11434/api/generate", json=json_request)
        if r.status_code == 200:
            result = {
                "role": "assistant",
                "content": r.json()["response"]
            }
        else:
            result = {}, r.status_code
    except Exception as error_print:
        app.logger.error(error_print)
        result = {}, 405
    return result



# Press the green button in the gutter to run the script.
if __name__ == '__main__':
    app.run(host='0.0.0.0', port=8348, debug=True, ssl_context=ssl)

# See PyCharm help at https://www.jetbrains.com/help/pycharm/