app.py

import gradio as gr
import os 
import json 
import requests


#Streaming endpoint 
API_URL = "https://api.openai.com/v1/chat/completions" #os.getenv("API_URL") + "/generate_stream"

#Testing with my Open AI Key 
#OPENAI_API_KEY = os.getenv("OPENAI_API_KEY") 

def predict(inputs, top_p, temperature, openai_api_key, chat_counter, chatbot=[], history=[]):  #repetition_penalty, top_k

    payload = {
    "model": "gpt-4-1106-preview",
    "messages": [{"role": "user", "content": f"{inputs}"}],
    "temperature" : 1.0,
    "top_p":1.0,
    "n" : 1,
    "stream": True,
    "presence_penalty":0,
    "frequency_penalty":0,
    }

    headers = {
    "Content-Type": "application/json",
    "Authorization": f"Bearer {openai_api_key}"
    }

    print(f"chat_counter - {chat_counter}")
    if chat_counter != 0 :
        messages=[]
        for data in chatbot:
          temp1 = {}
          temp1["role"] = "user" 
          temp1["content"] = data[0] 
          temp2 = {}
          temp2["role"] = "assistant" 
          temp2["content"] = data[1]
          messages.append(temp1)
          messages.append(temp2)
        temp3 = {}
        temp3["role"] = "user" 
        temp3["content"] = inputs
        messages.append(temp3)
        #messages
        payload = {
        "model": "gpt-4-1106-preview",
        "messages": messages, #[{"role": "user", "content": f"{inputs}"}],
        "temperature" : temperature, #1.0,
        "top_p": top_p, #1.0,
        "n" : 1,
        "stream": True,
        "presence_penalty":0,
        "frequency_penalty":0,
        }

    chat_counter+=1

    history.append(inputs)
    print(f"payload is - {payload}")
    # make a POST request to the API endpoint using the requests.post method, passing in stream=True
    response = requests.post(API_URL, headers=headers, json=payload, stream=True)
    #response = requests.post(API_URL, headers=headers, json=payload, stream=True)
    token_counter = 0 
    partial_words = "" 

    counter=0
    for chunk in response.iter_lines():
        #Skipping first chunk
        if counter == 0:
          counter+=1
          continue
        #counter+=1
        # check whether each line is non-empty
        if chunk.decode() :
          chunk = chunk.decode()
          # decode each line as response data is in bytes
          if len(chunk) > 12 and "content" in json.loads(chunk[6:])['choices'][0]['delta']:
              #if len(json.loads(chunk.decode()[6:])['choices'][0]["delta"]) == 0:
              #  break
              partial_words = partial_words + json.loads(chunk[6:])['choices'][0]["delta"]["content"]
              if token_counter == 0:
                history.append(" " + partial_words)
              else:
                history[-1] = partial_words
              chat = [(history[i], history[i + 1]) for i in range(0, len(history) - 1, 2) ]  # convert to tuples of list
              token_counter+=1
              yield chat, history, chat_counter  # resembles {chatbot: chat, state: history}  
                   

def reset_textbox():
    return gr.update(value='')

title = """<h1 align="center">🔥ChatGPT-4 Turbo API 🚀Streaming🚀</h1>"""
description = """Language models can be conditioned to act like dialogue agents through a conversational prompt that typically takes the form:
```
User: <utterance>
Assistant: <utterance>
User: <utterance>
Assistant: <utterance>
...
```
In this app, you can explore the outputs of a gpt-3.5-turbo LLM.
"""

css = """
body { background-color: #333; color: #fff; }
#col_container {width: 1000px; margin-left: auto; margin-right: auto;}
#chatbot {height: 520px; overflow: auto; background-color: #222; color: #fff;}
.gradio-toolbar { background-color: #222; }
.text_input { background-color: #222; color: #fff; }
.label { color: #fff; }
.slider { background-color: #222; }
.button { background-color: #444; color: #fff; }
"""

with gr.Blocks(css=css) as demo:
    gr.HTML(title)
    with gr.Column(elem_id="col_container"):
        openai_api_key = gr.Textbox(type='password', label="Insira sua chave de API OpenAI aqui")
        chatbot = gr.Chatbot(elem_id="chatbot")
        inputs = gr.Textbox(placeholder="Olá!", label="Digite uma entrada e pressione Enter", lines=3)
        state = gr.State([]) 
        b1 = gr.Button(value="Executar", variant="primary")

        # Parâmetros
        with gr.Accordion("Parameters", open=False):
            top_p = gr.Slider(minimum=0, maximum=1.0, value=1.0, step=0.05, label="Top-p")
            temperature = gr.Slider(minimum=0, maximum=5.0, value=1.0, step=0.1, label="Temperature")
            chat_counter = gr.Number(value=0, visible=False)

        # Associando a função 'predict' ao campo de entrada 'inputs' e ao botão 'b1'
        inputs.submit(predict, inputs=[inputs, top_p, temperature, openai_api_key, chat_counter, chatbot, state], outputs=[chatbot, state, chat_counter])
        b1.click(predict, inputs=[inputs, top_p, temperature, openai_api_key, chat_counter, chatbot, state], outputs=[chatbot, state, chat_counter])

        # Botão para limpar o campo de texto
        b1.click(reset_textbox, inputs=[], outputs=[inputs])

    demo.queue().launch(debug=True)