seawolf2357's picture
Update app.py
83746e4 verified
raw
history blame
2.88 kB
import gradio as gr
import requests
import os
import json
from collections import deque
# ν™˜κ²½ λ³€μˆ˜μ—μ„œ API 토큰 κ°€μ Έμ˜€κΈ°
TOKEN = os.getenv("HUGGINGFACE_API_TOKEN")
# API 토큰이 μ„€μ •λ˜μ–΄ μžˆλŠ”μ§€ 확인
if not TOKEN:
raise ValueError("API token is not set. Please set the HUGGINGFACE_API_TOKEN environment variable.")
# λŒ€ν™” 기둝을 κ΄€λ¦¬ν•˜λŠ” 큐 (μ΅œλŒ€ 10개의 λŒ€ν™” 기둝을 μœ μ§€)
memory = deque(maxlen=10)
def respond(
message,
history: list[tuple[str, str]],
system_message="AI Assistant Role",
max_tokens=512,
temperature=0.7,
top_p=0.95,
):
# μ‹œμŠ€ν…œ λ©”μ‹œμ§€μ— 접두사 μΆ”κ°€
system_prefix = "System: μž…λ ₯μ–΄μ˜ μ–Έμ–΄(μ˜μ–΄, ν•œκ΅­μ–΄, 쀑ꡭ어, 일본어 λ“±)에 따라 λ™μΌν•œ μ–Έμ–΄λ‘œ λ‹΅λ³€ν•˜λΌ."
full_system_message = f"{system_prefix}{system_message}"
# ν˜„μž¬ λŒ€ν™” λ‚΄μš©μ„ λ©”λͺ¨λ¦¬μ— μΆ”κ°€
memory.append((message, None))
messages = [{"role": "system", "content": full_system_message}]
# λ©”λͺ¨λ¦¬μ—μ„œ λŒ€ν™” 기둝을 가져와 λ©”μ‹œμ§€ λͺ©λ‘μ— μΆ”κ°€
for val in memory:
if val[0]:
messages.append({"role": "user", "content": val[0]})
if val[1]:
messages.append({"role": "assistant", "content": val[1]})
headers = {
"Authorization": f"Bearer {TOKEN}",
"Content-Type": "application/json"
}
payload = {
"model": "meta-llama/Meta-Llama-3.1-405B-Instruct",
"max_tokens": max_tokens,
"temperature": temperature,
"top_p": top_p,
"messages": messages
}
response = requests.post("https://api-inference.huggingface.co/v1/chat/completions", headers=headers, json=payload, stream=True)
response_text = ""
for chunk in response.iter_content(chunk_size=None):
if chunk:
chunk_data = chunk.decode('utf-8')
response_json = json.loads(chunk_data)
# content μ˜μ—­λ§Œ 좜λ ₯
if "choices" in response_json:
content = response_json["choices"][0]["message"]["content"]
response_text = content
# λ§ˆμ§€λ§‰ λŒ€ν™”μ— λͺ¨λΈμ˜ 응닡을 μΆ”κ°€ν•˜μ—¬ λ©”λͺ¨λ¦¬μ— μ €μž₯
memory[-1] = (message, response_text)
yield content
theme = "Nymbo/Nymbo_Theme"
# Gradio ChatInterface μ„€μ •
demo = gr.ChatInterface(
fn=respond,
theme=theme,
additional_inputs=[
gr.Textbox(value="AI Assistant Role", label="System message"),
gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"),
],
)
if __name__ == "__main__":
demo.queue(max_threads=20).launch()