| import time |
| import streamlit as st |
|
|
| COST_PER_1000_TOKENS_USD = 0.139 / 80 |
|
|
|
|
| def stream_handler(session_state, chat_stream, prompt, placeholder): |
| |
| |
| start_time = time.time() |
| full_response = "" |
|
|
| for chunk in chat_stream: |
| if chunk.token.text in ["</s>", "<|im_end|>"]: |
| break; |
| full_response += chunk.token.text |
| placeholder.markdown(full_response + "▌") |
| placeholder.markdown(full_response) |
|
|
| end_time = time.time() |
| elapsed_time = end_time - start_time |
| total_tokens_processed = len(full_response.split()) |
| tokens_per_second = total_tokens_processed // elapsed_time |
| len_response = (len(prompt.split()) + len(full_response.split())) * 1.25 |
| col1, col2, col3 = st.columns(3) |
|
|
| with col1: |
| st.write(f"**{tokens_per_second} tokens/second**") |
|
|
| with col2: |
| st.write(f"**{int(len_response)} tokens generated**") |
|
|
| with col3: |
| st.write( |
| f"**$ {round(len_response * COST_PER_1000_TOKENS_USD / 1000, 5)} cost incurred**" |
| ) |
|
|
| session_state["tps"] = tokens_per_second |
| session_state["tokens_used"] = len_response + session_state["tokens_used"] |
|
|
| return full_response |
|
|
