app.py

"""
@author: idoia lerchundi
"""
import os
import streamlit as st
from huggingface_hub import InferenceClient

# Load the API token from an environment variable
api_key = os.getenv("HF_TOKEN")

# Instantiate the InferenceClient
client = InferenceClient(api_key=api_key)

# Streamlit app title
st.title("Serverless Inference API") 

# Ensure the full_text key is initialized in session state
if "full_text" not in st.session_state:
    st.session_state["full_text"] = ""
    
# Create a text input area for user prompts
with st.form("my_form"):
    text = st.text_area("Enter text:", "Tell me a 4 sentence joke to make me laugh. A short joke, not a long one.")
    submitted = st.form_submit_button("Submit")

# Initialize the full_text variable
full_text = " "
# to get different jokes
top_p_init = 0.7

if submitted:
    top_p_init+=0.2
    messages = [
        {"role": "user", "content": text}
    ]
    
    # Create a new stream for each submission
    stream = client.chat.completions.create(
        model="TinyLlama/TinyLlama-1.1B-Chat-v1.0",
        messages=messages,
        temperature=0.5,
        max_tokens=300,
        top_p=top_p_init,
        stream=True
    )
    top_p_init +=0.1
    # Concatenate chunks to form the full response
    for chunk in stream:
        full_text += chunk.choices[0].delta.content
        
    # Update session state with the full response
    st.session_state["full_text"] = full_text
    
# Display the full response
if st.session_state["full_text"]:
    st.info(st.session_state["full_text"])