fix: Reduce token usage in e2e_tests to avoid Bedrock token throttling

Author: clareliguori

e2e_tests/python/chat_session.py

@@ -68,6 +68,9 @@ async def start(self) -> None:
             messages = []
 
             for i, user_input in enumerate(self.user_utterances):
+                # To reduce token use, clear out past messages for each utterance
+                messages = []
+
                 print(f"\nYou: {user_input}")
 
                 messages.append({"role": "user", "content": [{"text": user_input}]})

e2e_tests/python/main.py

@@ -88,7 +88,7 @@ async def main() -> None:
         "Hello!",
         "What is the current time in Seattle?",
         "Are there any weather alerts right now?",
-        "What documentation sources can you use to answer questions?",
+        "List your available documentation sources.",
         "Tell me a cat fact.",
         "Who is Tom Cruise?",
     ]

e2e_tests/typescript/src/chat_session.ts

@@ -82,6 +82,9 @@ export class ChatSession {
       const messages: Message[] = [];
 
       for (const [i, userInput] of this.userUtterances.entries()) {
+        // To reduce token use, clear out past messages for each utterance
+        messages.length = 0;
+
         console.log(`\You: ${userInput}`);
 
         messages.push({ role: "user", content: [{ text: userInput }] });

e2e_tests/typescript/src/main.ts

@@ -33,7 +33,7 @@ async function main(): Promise<void> {
     "Hello!",
     "What is the current time in Seattle?",
     "Are there any weather alerts right now?",
-    "What documentation sources can you use to answer questions?",
+    "List your available documentation sources.",
     "Tell me a cat fact.",
     "Who is Tom Cruise?",
   ];