v0.2.3

What's Changed

• feat: Add CohereEmbeddingFunction. by @wxywb in #23
• fix: Convert cohere embedding to np.float32 by @wxywb in #24
• Remove the specific version of protobuf.
• Change output of BM25EmbeddingFunction to np.float32.

Full Changelog: v0.2.2...v0.2.3

v0.2.2

What's Changed

• Use onnx as default embedding function.
• Decouple model dependencies, installing them as needed.

Full Changelog: v0.2.1...v0.2.2

v0.2.1

What's Changed

• Fix BGERerankFunction call return type. by @muazhari in #14
• feat: add JinaAI embedding and reranker models by @JoanFM in #16
• Bugfixes by @wxywb in #18
  • Fix loading bm25 config on Windows.
  • Convert sparse model output to csr_array explicitly.
  • Upgrade protobuf version to 3.20.2.
• feat: Change BGE M3 to use float32 as return by default. by @wxywb in #19

New Contributors

• @muazhari made their first contribution in #14
• @JoanFM made their first contribution in #16

Full Changelog: v0.2.0...v0.2.1

PyMilvus Model library v0.2.0 Release Notes

What's Changed

• Add several rerankers to improve search accuracy and relevance.
  BGERerankFunction
  CohereRerankFunction
  CrossEncoderRerankFunction
  VoyageRerankFunction
• BGEM3EmbeddingFunction now can load the local model from folder path.
• Optimize SpladeEmbeddingFunction implementation and reduce memory footprint when inferencing.

Full Changelog: v0.1.0...0.2.0