hyejascript / app.py
seawolf2357's picture
Update app.py
16b1415 verified
raw
history blame
6.7 kB
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("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")
def predict(inputs, top_p, temperature, openai_api_key, chat_counter, chatbot=[], history=[]):
# ์‚ฌ์šฉ์ž์˜ ์ž…๋ ฅ์„ ๋‚˜๋ ˆ์ด์…˜ ์Šคํƒ€์ผ์˜ ํ”„๋กฌํ”„ํŠธ๋กœ ๋ณ€ํ™˜
narration_prompt = f"๋™์˜์ƒ์— ์‚ฌ์šฉํ•  ์ „๋ฌธ์ ์ธ ๋‚˜๋ ˆ์ด์…˜์„ ์ž‘์„ฑํ•˜๋ผ. ๋ฐ˜๋“œ์‹œ ํ•œ๊ธ€๋กœ ์ž‘์„ฑํ• ๊ฒƒ. ์ผ์ฒด์˜ ์ง€๋ฌธ์ด๋‚˜ ์ง€์‹œ, ๋ฐฐ๊ฒฝ ์„ค๋ช… ๋“ฑ์„ ๋…ธ์ถœ ํ•˜๊ฑฐ๋‚˜ ์ถœ๋ ฅํ•˜์ง€ ๋ง๊ณ  ์ˆœ์ˆ˜ํ•œ ๋‚˜๋ ˆ์ด์…˜๋งŒ 2์ค„์”ฉ ๋ฌถ์–ด์„œ ์ตœ๋Œ€ 8์ค„ ์ด๋‚ด๋กœ ์ถœ๋ ฅ๋ ฅ. ์ž…๋ ฅ: '{inputs}'"
payload = {
"model": "gpt-4-1106-preview",
"messages": [{"role": "system", "content": narration_prompt}],
"temperature": temperature,
"top_p": top_p,
"n": 1,
"stream": True,
"presence_penalty": 0,
"frequency_penalty": 0,
"max_tokens": 1000 # ๋‚˜๋ ˆ์ด์…˜์˜ ๊ธธ์ด๋ฅผ ์ œํ•œ
}
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
# ๋‚˜๋ ˆ์ด์…˜์˜ ๊ธธ์ด๋ฅผ 8์ค„๋กœ ์ œํ•œ
if len(history) >= 8:
break
#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}
return chat, history, chat_counter
def reset_textbox():
return gr.update(value='')
title = """<h1 align="center">ํ˜œ์ž ์Šคํฌ๋ฆฝํŠธ</h1>"""
description = """์˜์ƒ ์ƒ์„ฑ์„ ์œ„ํ•œ ์Šคํฌ๋ฆฝํŠธ๋ฅผ AI๊ฐ€ ์ž๋™์œผ๋กœ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. ์ฃผ์ œ ํ‚ค์›Œ๋“œ๋‚˜ ๋ชฉ์  ๋“ฑ ํ•„์š”ํ•œ ๋‚ด์šฉ๋งŒ ๊ฐ„๋‹จํžˆ ์ž…๋ ฅํ•˜์„ธ์š”. :
```
User: <utterance>
Assistant: <utterance>
User: <utterance>
Assistant: <utterance>
...
```
In this app, you can explore the outputs of a gpt-3.5-turbo LLM.
"""
with gr.Blocks(css = """#col_container {width: 1000px; margin-left: auto; margin-right: auto;}
#chatbot {height: 520px; overflow: auto;}""") as demo:
gr.HTML(title)
with gr.Column(elem_id = "col_container"):
openai_api_key = gr.Textbox(type='password', label="Enter your OpenAI API key here")
# ์ถœ๋ ฅํผ (chatbot)์„ ์‚ฌ์šฉ์ž ์ž…๋ ฅํผ (inputs) ์•ž์— ๋ฐฐ์น˜
chatbot = gr.Chatbot(elem_id='chatbot') # c
inputs = gr.Textbox(placeholder="์—ฌ๊ธฐ์— ์ž…๋ ฅํ•˜์„ธ์š”.", label="๋‚˜๋ ˆ์ด์…˜ ์Šคํฌ๋ฆฝํŠธ๋ฅผ ์ƒ์„ฑํ•˜๊ณ  ์‹ถ์€ ์ฃผ์ œ์–ด๋‚˜ ๋ฌธ์žฅ์„ ์ž…๋ ฅํ•˜์„ธ์š”.") # t
state = gr.State([]) # s
b1 = gr.Button()
#inputs, top_p, temperature, top_k, repetition_penalty
with gr.Accordion("Parameters", open=False):
top_p = gr.Slider( minimum=-0, maximum=1.0, value=1.0, step=0.05, interactive=True, label="Top-p (nucleus sampling)",)
temperature = gr.Slider( minimum=-0, maximum=5.0, value=1.0, step=0.1, interactive=True, label="Temperature",)
#top_k = gr.Slider( minimum=1, maximum=50, value=4, step=1, interactive=True, label="Top-k",)
#repetition_penalty = gr.Slider( minimum=0.1, maximum=3.0, value=1.03, step=0.01, interactive=True, label="Repetition Penalty", )
chat_counter = gr.Number(value=0, visible=False, precision=0)
examples = gr.Examples(examples=[
["์ƒํ’ˆ ์„ค๋ช…:์ƒˆ๋กœ ์ถœ์‹œ๋œ 'ํ† ๋ฆฌ' ๋ฆฝ๋ฐค์€ FDA ์Šน์ธ, ์ตœ๊ณ ์˜ ๋ณด์Šต๋ ฅ, ๊ตฌ๋งค์ง€์ˆ˜ 1์œ„ "],
["๋ธŒ๋žœ๋”ฉ: 'ํ† ๋ฆฌ'๋ฆฝ๋ฐค์€ 20๋Œ€ ์—ฌ์„ฑ์—๊ฒŒ ์–ดํ•„ํ•  ๋ธŒ๋žœ๋”ฉ์ด ํ•„์š”ํ•ด"],
["๊ด‘๊ณ : ์„ค๋‚  ๋ถ€๋ชจ๋‹˜๊ณผ ์นœ์ง€ ์„ ๋ฌผ์€ ๋ฒ•์„ฑํฌ ๋ณด๋ฆฌ๊ตด๋น„๊ฐ€ ์ตœ๊ณ ๋ž๋‹ˆ๋‹ค."],
["์ •๋ณด ๊ณต์œ : ๋น„ํƒ€๋ฏผC ๊ณผ๋‹ค ๋ณต์šฉ์€ ๊ฑด๊ฐ•์— ์˜คํžˆ๋ ค ํ•ด๋กญ๋‹ค."],
["ํ™๋ณด: 'OpenAI'๋Š” '์ฑ—GPT'์˜ ๋งž์ถค GPT '์Šคํ† ์–ด'๋ฅผ ์˜คํ”ˆํ•˜์˜€๋‹ค."],
["์ธ์‚ฌ: '์• ํ”Œ ๋ฒ•์ธ'์˜ ๊ณ ๊ฐ๊ณผ ์ž„์ง์›์„ ์œ„ํ•œ ์ง„์ทจ์ ์ธ 2024๋…„ ์‹ ๋…„ ์ธ์‚ฌ"]
], inputs=[inputs], fn=predict, outputs=[chatbot, state, chat_counter])
inputs.submit( predict, [inputs, top_p, temperature, openai_api_key, chat_counter, chatbot, state], [chatbot, state, chat_counter],)
b1.click( predict, [inputs, top_p, temperature, openai_api_key, chat_counter, chatbot, state], [chatbot, state, chat_counter],)
b1.click(reset_textbox, [], [inputs])
inputs.submit(reset_textbox, [], [inputs])
#gr.Markdown(description)
demo.queue().launch(debug=True)