tokenizer
The tokenizer
property determines the strategy used for tokenizing text and estimating token counts within a message-processing feature.
This property allows overriding the default tokenization behavior by specifying a custom Tokenizer
implementation. By default, the NoTokenizer
instance is used, which effectively disables token counting by always returning zero.
Tokenizers play a key role in scenarios involving large language models (LLMs), where accurate token counting can be essential for understanding and managing resource usage or request limits.