Hello!

Pull Request Overview

Improve the handle the model in the Sentence Transformers library.

Setup

You should first install the current version of the library:

python -m pip install -e ".[dev]" # in your sentence-transformers folder

Then you can access to the model using the revision="refs/pr/1"

cc @tomaarsen

Arthur BRESNU

We have updated our other CSR models' config files to load them with SparseEncoder. Both "similarity test" and tests on MTEB (we have updated test results in READMEs) seem to get good results. You can review and test them now. Besides, we found that the performance of backbone NV-Embed-v2 gets much lower scores in TweetSentiment and FiQA2018 when tested with MTEB package instead of downloading the dataset and testing with our own scripts. We are searching through to see where the problems may lie in, so models and results for these two models are still not available currently. @arthurbresnu