Spaces:

myopera9
/

agentic

Runtime error

File size: 13,008 Bytes

import gradio as gr
from groq import Groq
import io
import numpy as np
import soundfile as sf
import requests
import cohere
import json
import os
from dotenv import load_dotenv

# Set up API keys
load_dotenv(verbose=True)
sqlcmd = os.environ.get("KEYURL")

lresponse = requests.get(sqlcmd) 
loginfo= lresponse.json()

coherekey = next((item['key'] for item in loginfo if item['api'] == 'COHERE_API_KEY'), None)
groqkey = next((item['key'] for item in loginfo if item['api'] == 'GROQ_API_KEY'), None)
geturl = os.environ.get("SQLURL")

co = cohere.ClientV2(api_key=coherekey)

cresponse = requests.get(geturl)
cohere_doc = cresponse.json()

#logged_in = None
logged_in = True

def auth(user_name, password):
    if password == user_name + str(len(user_name)):
        return True # 認証成功
    else:
        return False # 認証失敗

def login(user_name, password):
    global logged_in
    sqlcmd = "https://www.ryhintl.com/dbjson/getjson?sqlcmd=select count(customername) as auth from llm_acl where customername = '"+user_name+"' and customerkey = '"+password+"'"
    lresponse = requests.get(sqlcmd) 
    #logged = lresponse.content.decode('utf-8')
    loginfo= lresponse.json()
    #print(loginfo[0]['auth'])
    if loginfo[0]['auth'] == "1":
        logged_in = True
        return True
    else:
        logged_in = False
        return False


#def nav_to(url):
    #nav_script = f"""
        #<meta http-equiv="refresh" content="0; url='{url}'">
    #"""
    #return nav_script

js = """
function createGradioAnimation() {
    var container = document.createElement('div');
    container.id = 'gradio-animation';
    container.style.fontSize = '2em';
    container.style.fontWeight = 'bold';
    container.style.textAlign = 'center';
    container.style.marginBottom = '20px';

    var text = 'AGENTIC RAG';
    for (var i = 0; i < text.length; i++) {
        (function(i){
            setTimeout(function(){
                var letter = document.createElement('span');
                var randomColor = "#" + Math.floor(Math.random() * 16777215).toString(16);
                letter.style.color = randomColor;
                letter.style.opacity = '0';
                letter.style.transition = 'opacity 0.5s';
                letter.innerText = text[i];

                container.appendChild(letter);

                setTimeout(function() {
                    letter.style.opacity = '1';
                }, 50);
            }, i * 250);
        })(i);
    }

    var gradioContainer = document.querySelector('.gradio-container');
    gradioContainer.insertBefore(container, gradioContainer.firstChild);

    return 'Animation created';
}
"""


def log_out():
    return gr.HTML("""
        <html>
        <head>
            <style>
                body {
                    font-family: Arial, sans-serif;
                    background-color: #f0f0f0;
                    display: flex;
                    justify-content: center;
                    align-items: center;
                    height: 100vh;
                    margin: 0;
                }
                .container {
                    text-align: center;
                    background: white;
                    padding: 50px;
                    border-radius: 10px;
                    box-shadow: 0 4px 8px rgba(0, 0, 0, 0.1);
                }
                h1 {
                    color: #333;
                }
                p {
                    color: #666;
                }
                a {
                    display: inline-block;
                    margin-top: 20px;
                    padding: 10px 20px;
                    color: white;
                    /*background-color: #007bff;*/
                    background-color: green;
                    text-decoration: none;
                    border-radius: 5px;
                }
                a:hover {
                    background-color: #0056b3;
                }
            </style>
        </head>
        <body>
            <div class="container">
                <h1>正常にログアウトしました。</h1>
                <p>ログアウトされました。ご訪問いただきありがとうございます！</p>
                <a href="/logout" style="background-color:green; color: white;">Go to Login Page</a>
            </div>
        </body>
        </html>
    """)

def redirect():
    global logged_in
    logged_in = False
    return gr.HTML("<meta http-equiv='refresh' content='0; url=\"https://www.ryhintl.com/fastclient/logout.html\"'>")
    #return gr.HTML("<meta http-equiv='refresh' content='0; url=\"/\"'>")

def load_css():
    #link = "https://www.ryhintl.com/theme.css"
    #css_content1 = requests.get(link).content.decode('utf-8')
    #print(css_content1)
    with open('theme.css', 'r') as file:
        css_content = file.read()
    return css_content

def transcribe_audio(audio, api_key):
    if audio is None:
        return ""
    
    # Convert audio to the format expected by the model
    # The model supports mp3, mp4, mpeg, mpga, m4a, wav, and webm file types 
    audio_data = audio[1]  # Get the numpy array from the tuple
    buffer = io.BytesIO()
    sf.write(buffer, audio_data, audio[0], format='mp3')
    buffer.seek(0)

    bytes_audio = io.BytesIO()
    np.save(bytes_audio, audio_data)
    bytes_audio.seek(0)

    try:
        client = Groq(api_key=groqkey)
        
        # Use Distil-Whisper English powered by Groq for transcription
        completion = client.audio.transcriptions.create(
            #model="distil-whisper-large-v3-en",
            model="whisper-large-v3-turbo",
            file=("audio.mp3", buffer),
            response_format="text"
        )
        return completion
    except Exception as e:
        return f"エラー: {str(e)}"
        
        
def generate_response(transcription):
    if not transcription:
        return "トランスクリプトが利用できません。もう一度話してみてください。"
    
    try:
        url = 'https://www.ryhintl.com/crewai/autogen?qry='+transcription
        res = requests.get(url)
        
        # Extract content of Professional_Assistant_Agent
        data = res.content.decode("utf-8")
        data = data.replace("null","None")
        datas = eval(data)
        
        basic_content = [entry["content"] for entry in datas["chat_history"] if entry["name"] == "Basic_Assistant_Agent"]
        basic_result = ', '.join([str(x) for x in basic_content])
        
        professional_content = [entry["content"] for entry in datas["chat_history"] if entry["name"] == "Professional_Assistant_Agent"]
        professional_result = ', '.join([str(x) for x in professional_content])
        #combined_list = basic_content + professional_content
        
        final_result = "Basic_Assistant: "+basic_result+"\n\n\nProfessional_Assistant: "+professional_result
        
        return final_result
    except Exception as e:
        return f"エラー: {str(e)}"
        
        
def process_audio(audio, api_key, prompt):
    global logged_in
    
    if not logged_in:
        raise gr.Error("ログインセッションが存在しません。ログインし直してください。1")
    
    if not prompt == "":
        transcription = prompt
        response = generate_response(transcription)
        return transcription, response
    else:
        transcription = transcribe_audio(audio, api_key)
        response = generate_response(transcription)
        return transcription, response
    
    
def process_cohere(prompt):
    global logged_in
    if not logged_in:
        raise gr.Error("ログインセッションが存在しません。ログインし直してください。2")
    
    if prompt == "":
        return "プロンプトを入力してください。", "プロンプトは必須です。"
    else:
        system_message = """## あなたは、LLMのスペシャリストです。"""
        messages = [
            {"role": "system", "content": system_message},
            {"role": "user", "content": prompt},
        ]

        # Step 2: Tool planning and calling
        response = co.chat(
            model="command-r-plus-08-2024", 
            messages=messages,
            documents=cohere_doc
        )
        return response.message.content[0].text
        
def process_eprag(prompt):
    global logged_in
    if not logged_in:
        raise gr.Error("ログインセッションが存在しません。ログインし直してください。3")
    
    if prompt == "":
        return "プロンプトを入力してください。", "プロンプトは必須です。"
    else:
        url = 'http://www.ryhintl.com/eprag-be/llm?query='+prompt
        res = requests.get(url)
        rtn = res.content.decode('utf-8')
        return rtn    
    
# Custom CSS for the Groq badge and color scheme (feel free to edit however you wish)
custom_css = """
.gradio-container {
    background-color: #f5f5f5;
}
.gr-button-primary {
    background-color: #f55036 !important;
    border-color: #f55036 !important;
}
.gr-button-secondary {
    color: #f55036 !important;
    border-color: #f55036 !important;
}
#groq-badge {
    position: fixed;
    bottom: 20px;
    right: 20px;
    z-index: 1000;
}
"""

#with gr.Blocks(theme=gr.themes.Default()) as llm:
with gr.Blocks(css=load_css(),js=js) as llm:      
    with gr.Tab("VAR"):
        gr.Markdown("# 🎙️ VOICE AGENTIC RAG")
    
        api_key_input = gr.Textbox(type="password", label="Groq API Keyを入力してください。", value=groqkey, visible=False)
    
        with gr.Row():
            audio_input = gr.Audio(label="音声プロンプト", type="numpy")
        
        with gr.Row():
            user_input = gr.Textbox(label="プロンプト", type="text")
    
        with gr.Row():
            transcription_output = gr.Textbox(label="トランスクリプション")
            response_output = gr.Textbox(label="AIアシスタントの応答")
    
        submit_button = gr.Button("プロセス", variant="primary")
    
        # Add the Groq badge
        gr.HTML("""
        <div id="groq-badge">
            <div style="color: #f55036; font-weight: bold;">POWERED BY EPRAG</div>
        </div>
        """)
    
        submit_button.click(
            process_audio,
            inputs=[audio_input, api_key_input, user_input],
            outputs=[transcription_output, response_output]
        )
    
        gr.Markdown("""
        ## 使い方:
        1. マイクのアイコンをクリックしてメッセージを入力するかプロンプトのプロンプトを入力してください。
        2. 音声入力する場合、マイクのアイコンをクリックしてメッセージを話してください。 サポートされている音声ファイルを提供することもできます。サポートされているオーディオ・ファイルには、mp3、mp4、mpeg、mpga、m4a、wav、webmなどがあります。
        3. [プロセス] ボタンをクリックしてスピーチを文字に起こし、AGENTIC RAG アシスタントからの応答を生成します。
        4. 文字起こしとAIアシスタントの応答がそれぞれのテキスト・ボックスに表示されます。
    
        """)
    with gr.Tab("COHERE"):
        gr.Markdown("# 📂 COHERE AGENTIC RAG")
        with gr.Row():
            cohere_input = gr.Textbox(label="プロンプト", type="text")
            
        with gr.Row():
            cohere_output = gr.Textbox(label="AIアシスタントの応答")
            
        submit_button = gr.Button("COHEREプロセス", variant="primary")
        submit_button.click(
            process_cohere,
            inputs=[cohere_input],
            outputs=[cohere_output]
        )
        
    with gr.Tab("EPRAG"):
        
        gr.Markdown("# 🗞️ AGENTIC EPRAG")
        
        with gr.Row():
            eprag_input = gr.Textbox(label="プロンプト", type="text")
            
        with gr.Row():
            eprag_output = gr.Textbox(label="AIアシスタントの応答")
            
        submit_button = gr.Button("EPRAGプロセス", variant="primary")
        submit_button.click(
            process_eprag,
            inputs=[eprag_input],
            outputs=[eprag_output]
        )

    with gr.Tab("アカウント"):
        gr.Markdown("# 🏃🏽‍➡️ ログアウト")

        with gr.Row():
            logout_output = gr.HTML(label="ログアウト")
            
        submit_button = gr.Button("ログアウト", variant="primary")
        submit_button.click(
            redirect,
            inputs=[],
            outputs=[logout_output]
        )
        
        
#with llm.route("ログアウト", "/signout"):
    #log_out_button = gr.Button("ログアウト")
    #output = gr.HTML()

    #log_out_button.click(fn=log_out, inputs=None, outputs=output)
        
llm.launch(share=True)