Not all the agents are LLM based. #639
Description
hi, there,
Sorry i come to this topic again in this new thread.
if conf is None:
model_name = os.getenv("LLM_MODEL_NAME")
api_key = os.getenv("LLM_API_KEY")
base_url = os.getenv("LLM_BASE_URL")
assert api_key and model_name, (
"LLM_MODEL_NAME and LLM_API_KEY (environment variables) must be set, "
"or pass AgentConfig explicitly"
)
logger.info(f"AgentConfig is empty, using env variables:\n"
f"LLM_API_KEY={'*' * min(len(api_key), 8) if api_key else 'Not set'}\n"
f"LLM_BASE_URL={base_url}\n"
f"LLM_MODEL_NAME={model_name}")
conf = AgentConfig(
llm_provider=os.getenv("LLM_PROVIDER", "openai"),
llm_model_name=model_name,
llm_api_key=api_key,
llm_base_url=base_url,
llm_temperature=float(os.getenv("LLM_TEMPERATURE", "0.7")),
)
else:
self.conf = conf
The base class should not be aware of its subclass. The BaseAgent won't care if a LLM will be used in the future.
Aworld can do much more than just to wrap a LLM. In a nutshell, Don't be tempted to LLM , IMHO.