Skip to content

v0.1.0

Latest
Latest

Choose a tag to compare

View all tags
@philnash philnash released this 14 Jul 07:28
v0.1.0
c4b5b4f

Added

  • Browser- and Node.js-compatible local reranking powered by Transformers.js and open Hugging Face cross-encoder models.
  • The Reranker API for ranking strings or structured documents while preserving each original document, index, and relevance score.
  • Explicit model configuration with support for Transformers.js options such as device and dtype; dtype defaults to auto.
  • Result limiting with the topK option.
  • Custom reranking strategies through strategyFactory.
  • Typed errors for invalid input, unsupported strategies, model-loading failures, and attempts to use a disposed reranker.
  • Resource cleanup through dispose() and Symbol.asyncDispose, including safe handling of active ranking calls and repeated disposal.
  • A LangChain BaseDocumentCompressor adapter exported from rerankers/langchain, with compressDocuments() and raw rerank() support.
  • A Vercel AI SDK reranking model exported from rerankers/ai-sdk, supporting text and object documents.
  • TypeScript declarations for the core library and both optional ecosystem adapters.
  • Documentation for tested English, multilingual, and Chinese-English reranking models.
  • A browser demo showing local document reranking.
Assets 2
Loading