Summary and motivation
We have had cases that the turning fuzziness on by default in the OpenSearch integration has caused large performance issues for clients when they have a lot of documents in their index.
More details on the performance issues can be found here: https://linear.app/deepset/issue/BUI-327/turn-off-fuzziness-in-opensearchhybridretriever-in-pipeline-templates
To summarize we have had cases where search times with the OpenSearchHybridRetriever went from 30 seconds to 6 seconds when turning fuzziness off.
So the proposal is to switch the default value of fuzziness from AUTO to 0 so it's off by default in the OpenSearchHybridRetriever and the OpenSearchBm25Retriever. This would be a breaking change so we should make a new major version release.
Summary and motivation
We have had cases that the turning fuzziness on by default in the OpenSearch integration has caused large performance issues for clients when they have a lot of documents in their index.
More details on the performance issues can be found here: https://linear.app/deepset/issue/BUI-327/turn-off-fuzziness-in-opensearchhybridretriever-in-pipeline-templates
To summarize we have had cases where search times with the OpenSearchHybridRetriever went from 30 seconds to 6 seconds when turning fuzziness off.
So the proposal is to switch the default value of fuzziness from
AUTOto0so it's off by default in theOpenSearchHybridRetrieverand theOpenSearchBm25Retriever. This would be a breaking change so we should make a new major version release.