From 810f772959fbacb5c55c76b8871f22177a8e4d3e Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Benjamin=20Clavi=C3=A9?= Date: Fri, 16 Aug 2024 12:41:24 +0200 Subject: [PATCH] chore: prep release --- README.md | 4 +++- pyproject.toml | 2 +- rerankers/__init__.py | 2 +- 3 files changed, 5 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index 87da7c6..37f2fea 100644 --- a/README.md +++ b/README.md @@ -14,9 +14,10 @@ Welcome to `rerankers`! Our goal is to provide users with a simple API to use an ## Updates +- v0.5.0: Added support for the current state-of-the-art rerankers, BAAI's series of `BGE` layerwise LLM rerankers, based on [Gemma](https://huggingface.co/BAAI/bge-reranker-v2.5-gemma2-lightweight) and MiniCPM. These are different from RankGPT, as they're not listwise: the models are repurposed as "cross-encoders", and do output logit scores. - v0.4.0: ColBERT performance improvement! It should now be faster and result in stronger results following implementation of the JaColBERTv2.5 dynamic query length method. This version also now supports HuggingFace's Text-Embedding-Server (TEI) inference as an API reranker option, thanks to [@srisudarsan](https://github.com/srisudarsan). - v0.3.1: T5 bugfix and native default support for new Portuguese T5 rerankers. -- v0.3.0: 🆕 Many changes! Experimental support for RankLLM, directly backed by the [rank-llm library](https://github.com/castorini/rank_llm). A new `Document` object, courtesy of joint-work by [@bclavie](https://github.com/bclavie) and [Anmol6](https://github.com/Anmol6). This object is transparent, but now offers support for `metadata` stored alongside each document. Many small QoL changes (RankedResults can be itered on directly...) +- v0.3.0: Many changes! Experimental support for RankLLM, directly backed by the [rank-llm library](https://github.com/castorini/rank_llm). A new `Document` object, courtesy of joint-work by [@bclavie](https://github.com/bclavie) and [Anmol6](https://github.com/Anmol6). This object is transparent, but now offers support for `metadata` stored alongside each document. Many small QoL changes (RankedResults can be itered on directly...) - v0.2.0: [FlashRank](https://github.com/PrithivirajDamodaran/FlashRank) rerankers, Basic async support thanks to [@tarunamasa](https://github.com/tarunamasa), MixedBread.ai reranking API - v0.1.2: Voyage reranking API - v0.1.1: Langchain integration fixed! @@ -198,6 +199,7 @@ Models: - ✅ Any standard SentenceTransformer or Transformers cross-encoder - ✅ RankGPT (Available both via the original RankGPT implementation and the improved RankLLM one) - ✅ T5-based pointwise rankers (InRanker, MonoT5...) +- ✅ LLM-based pointwise rankers (BAAI/bge-reranker-v2.5-gemma2-lightweight, etc...) - ✅ Cohere, Jina, Voyage and MixedBread API rerankers - ✅ [FlashRank](https://github.com/PrithivirajDamodaran/FlashRank) rerankers (ONNX-optimised models, very fast on CPU) - ✅ ColBERT-based reranker - not a model initially designed for reranking, but does perform quite strongly in some cases. Implementation is lightweight, based only on transformers. diff --git a/pyproject.toml b/pyproject.toml index a95cced..833942b 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -14,7 +14,7 @@ packages = [ name = "rerankers" -version = "0.5.0rc1" +version = "0.5.0" description = "A unified API for various document re-ranking models." diff --git a/rerankers/__init__.py b/rerankers/__init__.py index d9a4899..b08fdb4 100644 --- a/rerankers/__init__.py +++ b/rerankers/__init__.py @@ -2,4 +2,4 @@ from rerankers.documents import Document __all__ = ["Reranker", "Document"] -__version__ = "0.5.0rc1" +__version__ = "0.5.0"