Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
|
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import streamlit as st
|
| 2 |
+
from huggingface_hub import InferenceClient
|
| 3 |
+
|
| 4 |
+
# Initialize the InferenceClient
|
| 5 |
+
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
| 6 |
+
|
| 7 |
+
def respond(
|
| 8 |
+
message,
|
| 9 |
+
history: list[tuple[str, str]],
|
| 10 |
+
system_message,
|
| 11 |
+
max_tokens,
|
| 12 |
+
temperature,
|
| 13 |
+
top_p,
|
| 14 |
+
):
|
| 15 |
+
messages = [{"role": "system", "content": system_message}]
|
| 16 |
+
|
| 17 |
+
for val in history:
|
| 18 |
+
if val[0]:
|
| 19 |
+
messages.append({"role": "user", "content": val[0]})
|
| 20 |
+
if val[1]:
|
| 21 |
+
messages.append({"role": "assistant", "content": val[1]})
|
| 22 |
+
|
| 23 |
+
messages.append({"role": "user", "content": message})
|
| 24 |
+
|
| 25 |
+
response = ""
|
| 26 |
+
response_container = st.empty() # Placeholder to update the response text
|
| 27 |
+
|
| 28 |
+
for message in client.chat_completion(
|
| 29 |
+
messages,
|
| 30 |
+
max_tokens=max_tokens,
|
| 31 |
+
stream=True,
|
| 32 |
+
temperature=temperature,
|
| 33 |
+
top_p=top_p,
|
| 34 |
+
):
|
| 35 |
+
token = message.choices[0].delta.content
|
| 36 |
+
|
| 37 |
+
response += token
|
| 38 |
+
response_container.text(response) # Dynamically update the response
|
| 39 |
+
|
| 40 |
+
# Streamlit interface
|
| 41 |
+
st.title("Chatbot Interface")
|
| 42 |
+
|
| 43 |
+
# User inputs
|
| 44 |
+
system_message = st.text_input("System message", value="You are a friendly Chatbot.")
|
| 45 |
+
message = st.text_input("User message", key="user_message")
|
| 46 |
+
max_tokens = st.slider("Max new tokens", 1, 2048, 512)
|
| 47 |
+
temperature = st.slider("Temperature", 0.1, 4.0, 0.7)
|
| 48 |
+
top_p = st.slider("Top-p (nucleus sampling)", 0.1, 1.0, 0.95)
|
| 49 |
+
|
| 50 |
+
# A placeholder for chat history
|
| 51 |
+
history = st.session_state.get('history', [])
|
| 52 |
+
|
| 53 |
+
# Submit button
|
| 54 |
+
if st.button("Send"):
|
| 55 |
+
if message:
|
| 56 |
+
# Append the new message to the history
|
| 57 |
+
history.append((message, "")) # No assistant response yet
|
| 58 |
+
|
| 59 |
+
# Call the respond function and update the history with the assistant's reply
|
| 60 |
+
for response in respond(message, history, system_message, max_tokens, temperature, top_p):
|
| 61 |
+
history[-1] = (message, response)
|
| 62 |
+
|
| 63 |
+
# Save updated history in session state
|
| 64 |
+
st.session_state['history'] = history
|
| 65 |
+
|
| 66 |
+
# Display the chat history
|
| 67 |
+
for user_msg, bot_msg in history:
|
| 68 |
+
st.write(f"**User**: {user_msg}")
|
| 69 |
+
st.write(f"**Assistant**: {bot_msg}")
|