Scott Cogan
commited on
Commit
·
ea9d55b
1
Parent(s):
b3920f7
requirements update for llm compat
Browse files
app.py
CHANGED
|
@@ -328,59 +328,65 @@ class BasicAgent:
|
|
| 328 |
}]
|
| 329 |
)
|
| 330 |
|
|
|
|
| 331 |
response = self.primary_llm.invoke(
|
| 332 |
messages_with_system,
|
| 333 |
tools=[genai_tool]
|
| 334 |
)
|
|
|
|
|
|
|
| 335 |
except Exception as e:
|
| 336 |
error_str = str(e)
|
| 337 |
-
|
| 338 |
-
|
| 339 |
-
|
| 340 |
-
|
| 341 |
-
|
| 342 |
-
|
| 343 |
-
|
| 344 |
-
|
| 345 |
-
|
| 346 |
-
|
| 347 |
-
|
| 348 |
-
|
| 349 |
-
|
| 350 |
-
|
| 351 |
-
|
| 352 |
-
|
| 353 |
-
|
| 354 |
-
|
| 355 |
-
|
| 356 |
-
|
| 357 |
-
|
| 358 |
-
"required": ["query"]
|
| 359 |
}
|
| 360 |
-
}
|
| 361 |
-
|
| 362 |
-
|
| 363 |
-
logger.info("Successfully used fallback LLM")
|
| 364 |
-
except Exception as fallback_error:
|
| 365 |
-
logger.error(f"Fallback LLM also failed: {str(fallback_error)}")
|
| 366 |
-
return {
|
| 367 |
-
"messages": [AIMessage(content="All LLM services are currently unavailable. Please try again later.")],
|
| 368 |
-
"next": END
|
| 369 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 370 |
else:
|
| 371 |
-
logger.warning("No fallback LLM available")
|
| 372 |
return {
|
| 373 |
-
"messages": [AIMessage(content="
|
| 374 |
"next": END
|
| 375 |
}
|
| 376 |
-
|
| 377 |
-
|
|
|
|
| 378 |
wait_time = 60 * (retry_count + 1) # Exponential backoff
|
| 379 |
logger.warning(f"Rate limit hit, waiting {wait_time} seconds before retry...")
|
| 380 |
time.sleep(wait_time)
|
| 381 |
raise # Re-raise to trigger retry
|
| 382 |
-
|
| 383 |
-
|
| 384 |
|
| 385 |
logger.info("\n=== Model Output ===")
|
| 386 |
log_message(response, " ")
|
|
|
|
| 328 |
}]
|
| 329 |
)
|
| 330 |
|
| 331 |
+
logger.info("Attempting to use primary LLM (Gemini)")
|
| 332 |
response = self.primary_llm.invoke(
|
| 333 |
messages_with_system,
|
| 334 |
tools=[genai_tool]
|
| 335 |
)
|
| 336 |
+
logger.info("Successfully used primary LLM")
|
| 337 |
+
|
| 338 |
except Exception as e:
|
| 339 |
error_str = str(e)
|
| 340 |
+
logger.error(f"Primary LLM error: {error_str}")
|
| 341 |
+
|
| 342 |
+
# Check if we should try fallback
|
| 343 |
+
if (hasattr(self, 'fallback_llm') and self.fallback_llm is not None and
|
| 344 |
+
("429" in error_str or "object" in error_str or "string" in error_str)):
|
| 345 |
+
try:
|
| 346 |
+
logger.info("Attempting to use fallback LLM (OpenAI)")
|
| 347 |
+
# For OpenAI, we can use the system message directly
|
| 348 |
+
response = self.fallback_llm.invoke(
|
| 349 |
+
[self.sys_msg] + messages,
|
| 350 |
+
tools=[{
|
| 351 |
+
"type": "function",
|
| 352 |
+
"function": {
|
| 353 |
+
"name": "google_search",
|
| 354 |
+
"description": "Search for information on the web",
|
| 355 |
+
"parameters": {
|
| 356 |
+
"type": "object",
|
| 357 |
+
"properties": {
|
| 358 |
+
"query": {
|
| 359 |
+
"type": "string",
|
| 360 |
+
"description": "The search query"
|
|
|
|
| 361 |
}
|
| 362 |
+
},
|
| 363 |
+
"required": ["query"]
|
| 364 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 365 |
}
|
| 366 |
+
}]
|
| 367 |
+
)
|
| 368 |
+
logger.info("Successfully used fallback LLM")
|
| 369 |
+
except Exception as fallback_error:
|
| 370 |
+
logger.error(f"Fallback LLM error: {str(fallback_error)}")
|
| 371 |
+
if "429" in str(fallback_error):
|
| 372 |
+
return {
|
| 373 |
+
"messages": [AIMessage(content="All LLM services are currently rate limited. Please try again later.")],
|
| 374 |
+
"next": END
|
| 375 |
+
}
|
| 376 |
else:
|
|
|
|
| 377 |
return {
|
| 378 |
+
"messages": [AIMessage(content="All LLM services are currently unavailable. Please try again later.")],
|
| 379 |
"next": END
|
| 380 |
}
|
| 381 |
+
else:
|
| 382 |
+
# If no fallback available or error not related to rate limits
|
| 383 |
+
if "429" in error_str:
|
| 384 |
wait_time = 60 * (retry_count + 1) # Exponential backoff
|
| 385 |
logger.warning(f"Rate limit hit, waiting {wait_time} seconds before retry...")
|
| 386 |
time.sleep(wait_time)
|
| 387 |
raise # Re-raise to trigger retry
|
| 388 |
+
else:
|
| 389 |
+
raise
|
| 390 |
|
| 391 |
logger.info("\n=== Model Output ===")
|
| 392 |
log_message(response, " ")
|