app.py

import streamlit as st
import requests
import os
import json

def start_server():   
    os.system("uvicorn inference_server:app --port 8080 --host 0.0.0.0 --workers 2")
    st.session_state['server_started'] = True
    
def is_port_in_use(port):
    import socket
    with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
        return s.connect_ex(('0.0.0.0', port)) == 0
 
if 'server_started' not in st.session_state:
    st.session_state['server_started'] = False

if not st.session_state['server_started']:
    start_server()     

st.set_page_config(
    page_title="Incoder Example",
    page_icon="👨‍💻",
)

st.markdown("# 👨‍💻 Incoder Example")

txt = st.text_area('Write code here', '''import os

def remove_file(file):''', height=400)

gen = st.button('Generate')

c = st.code('')

max_length = st.slider('max_length', 1, 2048, 128)
top_k = st.slider('top_k', 0, 100, 50)
top_p = st.slider('top_p', 0.0, 1.0, 0.9)
temperature = st.slider('temperature', 0.0, 1.0, 0.6)
num_beams = st.slider('num_beams', 1, 100, 0)
repetition_penalty = st.slider('repetition_penalty', 1.0, 10.0, 1.0)


if gen:
    c.code('Generating...')
    req = f"http://0.0.0.0:8080?input_text={txt}&top_p={top_p}&top_k={top_k}&temperature={temperature}&num_beams={num_beams}&repetition_penalty={repetition_penalty}&max_length={max_length}"
    res = requests.get(req)
    print('ok')
    c.code(json.loads(res.text))