Class PromptCompressor

The TypeScript implementation on original PromptCompressor, which is a class for compressing prompts using a language model.

See

Original Implementation

Index

Constructors

Methods

Constructors

constructor

new PromptCompressor(
    model: PreTrainedModel,
    tokenizer: PreTrainedTokenizer,
    getPureToken: GetPureTokenFunction,
    isBeginOfNewWord: IsBeginOfNewWordFunction,
    oaiTokenizer: { encode: (text: string) => { length: number } },
    llmlingua2Config?: LLMLingua2Config,
    logger?: Logger,
): PromptCompressor
Parameters
- model: PreTrainedModel
  The pre-trained model to use for compression.
- tokenizer: PreTrainedTokenizer
  The pre-trained tokenizer to use for compression.
- getPureToken: GetPureTokenFunction
  Function to get the pure token from a token. This is used to normalize tokens before processing.
- isBeginOfNewWord: IsBeginOfNewWordFunction
  Function to check if a token is the beginning of a new word. This is used to determine how to merge tokens into words.
- oaiTokenizer: { encode: (text: string) => { length: number } }
  The tokenizer to use calculating the compression rate. It needs to return a collection that exposes length.
- llmlingua2Config: LLMLingua2Config = DEFAULT_LLMLINGUA2_CONFIG
  Configuration for LLMLingua2.
- logger: Logger = DEFAULT_LOGGER
  Logger function to log messages.
Returns PromptCompressor
- Defined in prompt-compressor.ts:126

Methods

compress

compress(
context: string,
__namedParameters: CompressPromptOptions,
): Promise<string>
Compresses a prompt based on the given options.
Parameters
- context: string
- __namedParameters: CompressPromptOptions
Returns Promise<string>
- Defined in prompt-compressor.ts:177