prompt token
The number of tokens contained in the input portion sent to a model in an LLM call, including the system prompt, context, and conversation history. It is a core metric for measuring inference cost and context expansion.
Timeline (last 90 days)
In the Agent's SQL drafting loop, each round of revision resends the entire conversation history of all previous attempts verbatim, causing prompt tokens to grow with the number of attempts, i.e., context bloat.
In LLM calls, prompt tokens refer to the number of tokens contained in the input portion sent to the model, including the system prompt, schema summary, historical conversation, and current instruction; completion tokens are the output portion generated by the model.
All Facts (2)
In the Agent's SQL drafting loop, each round of revision resends the entire conversation history of all previous attempts verbatim, causing prompt tokens to grow with the number of attempts, i.e., context bloat.
50%UnverifiedIn LLM calls, prompt tokens refer to the number of tokens contained in the input portion sent to the model, including the system prompt, schema summary, historical conversation, and current instruction; completion tokens are the output portion generated by the model.
50%