Evaluation Pipeline Issue.

[SummerFall1819]

When I make my own implementation, I found a little bug in your code. Take qwen3vl as an example.

MobileWorld/src/mobile_world/agents/implementations/qwen3vl.py

Lines 274 to 297 in 41a451f

	while True:
	prediction = self.openai_chat_completions_create(
	model=self.model_name,
	messages=messages,
	retry_times=3,
	**self.runtime_conf,
	)
	if prediction is None:
	raise Exception("Error when fetching response from clients")
	try:
	parsed_response = parse_action_to_structure_output(
	prediction,
	)
	logger.info(f"Parsed response: \n{parsed_response}")
	break
	except Exception:
	if try_times > 0:
	logger.error("Error when parsing response from clients")
	logger.error(traceback.format_exc())
	prediction = None
	try_times -= 1

This code have a retry mechanism. But after three fails it won't quit the loop. Thus constantly logging openAI API use. similar issue is reported in #32 . Would you like to make a quick fix on that?