Skip to main content

Nerif Token Counter

Counting token consumed by specific agent or request method like nerif()

Basic Usage​

A counter should be create seperately, and pass it into constructor of agent class or special methods.

from nerif.agent import NerifTokenCounter
from nerif.core import nerif

counter = NerifTokenCounter()

if nerif("the sky is blue", counter=counter):
    print("True")

agent = nerif.agent.SimpleChatAgent(counter=counter)

print(counter.model_token)

Classes​

NerifTokenCounter​

A class to count the token consumed by specific Agent or method.

Attributes:

  • response_parser (ResponseParserBase): Parser for response from llm backend, default: OpenAIResponseParser()

Methods:

  • set_parser(parser=ResponseParserBase): Set response parser to specific parser.
  • set_parser_based_on_model(self, model_name=str): Set response parser from name of model.
  • count_from_response(response=any): Counting tokens consumed by the model from response.

Example:

from nerif.agent import NerifTokenCounter
from nerif.core import nerif

counter = NerifTokenCounter()

if nerif("the sky is blue", counter=counter):
    print("True")

agent = nerif.agent.SimpleChatAgent(counter=counter)

print(counter.model_token)

ResponseParserBase​

Base class of response parser.

Methods:

  • __call__(response=any) -> ModelCost: Parse token usage from response.

Derived Classes:

  • OpenAIResponseParser: Parser for OpenAI compatible API.
  • OllamaResponseParser: Parser for Ollama API.

NerifTokenConsume​

warning

Do not use this class directly, plase use NerifTokenCounter

Attributes:

  • model_cost (dict{str: ModelCost}): A dict to store model name and ModelCost.

Methods:

  • __getitem__(key=str) -> ModelCost: Get ModelCost from internal dict.
  • append(consume=ModelCost): Append cost information.
  • __repr__() -> str: Prettyprint cost information.

ModelCost​

warning

Do not use this class directly, plase use NerifTokenCounter

A class to store token consumed by specific model.

Attributes:

  • model_name (str): The name of model.
  • request (int): The count of token in request.
  • response (int): The count of token in response.

Methods:

  • add_cost(request=int, response=None|int): Append token usage.