txtai has an internal mechanism to run mean and cls pooling. This adds support for sentence-transformers models without having to install the library. This component was added to limit the number of package dependencies.

One difference noted between sentence-transformers and txtai is that txtai doesn't read the max_seq_length parameter. A change should be added to optionally read this parameter. The default behavior will continue to use the tokenizer/model max length unless this parameter is specified.