The large language model selects a model with a thinking chain, and strictly adheres to the max_token parameter. OpenAI-API-compatible will send a request with max_token=5 during testing. In the model ...