Usando streamlit

3_streamlit_TikToken_conversation_history_limit/main.py

@@ -0,0 +1,95 @@
+"""
+Data Scientist.: Dr. Eddy Giusepe Chirinos Isidro
+
+
+Executar script
+===============
+
+$ streamlit run main.py 
+"""
+import openai
+import streamlit as st
+import tiktoken
+import tenacity
+
+st.title(":orange[New Streamlit Chat Elements]")
+
+# initialize session state
+if "user_messages" not in st.session_state:
+    st.session_state["user_messages"] = []
+if "assistant_messages" not in st.session_state:
+    st.session_state["assistant_messages"] = []
+
+if "message_history" not in st.session_state:
+    st.session_state["message_history"] = [
+        {"role": "system", "content": "You are a helpful assistant."}
+    ]
+
+
+# Define the function to count tokens
+def count_tokens(text):
+    encoding = tiktoken.encoding_for_model("gpt-3.5-turbo")
+    return len(encoding.encode(text))
+
+# Set the maximum token limit
+MAX_MEMORY_TOKENS = 2000 # ou 5000
+
+# Tenacity tente novamente 5 vezes com 2 segundos de espera cada:
+@tenacity.retry(wait=tenacity.wait_fixed(2), stop=tenacity.stop_after_attempt(5))
+def gpt_call(prompt, placeholder_response): # AQUI FAZEMOS A CHAMADA PARA À OpenAI.
+
+    response = openai.ChatCompletion.create(
+        model="gpt-3.5-turbo-0613",
+        stream = True,
+        request_timeout=30,
+        messages=st.session_state["message_history"],
+    )
+
+    # return response['choices'][0]['message']['content']
+        # Process the response from the API
+    assistant_response = ""
+    for chunk in response:
+        if "content" in chunk["choices"][0]["delta"]:
+            r_text = chunk["choices"][0]["delta"]["content"]
+            assistant_response += r_text
+            placeholder_response.chat_message("assistant").markdown(assistant_response, unsafe_allow_html=True)
+    
+    
+    return assistant_response
+
+
+prompt = st.chat_input("Say something")
+# print(len(st.session_state["user_messages"]))
+
+
+if prompt:
+    st.session_state["user_messages"].append(prompt)
+    st.session_state["message_history"].append({"role": "user", "content": prompt})
+    prompt = st.session_state["user_messages"][-1]
+
+total_user_messages = len(st.session_state["user_messages"])
+
+for i in range(total_user_messages):
+    st.chat_message("user").write(st.session_state["user_messages"][i])
+
+    if i < total_user_messages - 1:
+        st.chat_message("assistant").write(st.session_state["assistant_messages"][i])
+    elif i == total_user_messages-1:
+        placeholder_response = st.empty()
+        st.session_state["assistant_messages"].append(gpt_call(prompt, placeholder_response))
+        st.session_state["message_history"].append({"role": "assistant", "content": st.session_state["assistant_messages"][-1]})
+
+# Calculate the total tokens in the conversation history
+total_tokens = sum(count_tokens(message["content"]) for message in st.session_state.message_history)
+
+# Remove the oldest message from conversation history if total tokens exceed the maximum limit
+while total_tokens > MAX_MEMORY_TOKENS:
+    if len(st.session_state.message_history) > 2:
+        removed_message = st.session_state.message_history.pop(1)
+        total_tokens -= count_tokens(removed_message["content"])
+    else:
+        break
+st.write("total tokens: ", total_tokens)   
+print(st.session_state["message_history"])
+        
+

3_streamlit_TikToken_conversation_history_limit/requirements.txt

@@ -0,0 +1,4 @@
+openai==0.27.8
+streamlit==1.24.0
+tiktoken==0.3.3
+tenacity==8.2.2