How to use it?

• call fit_s3_on_corpus() to fit s3e on your corpus (or a subset)
• save results or pass them directly into the Inferencer to get your sentence embeddings

   inferencer = Inferencer(model=model, processor=processor, task_type="embeddings", gpu=use_gpu,
                       batch_size=batch_size, extraction_strategy="s3e", extraction_layer=-1,
                       s3e_stats=s3e_stats)

See example script

Known limitations:

• Speed of fit_s3_on_corpus:
  -- no multiprocessing for tokenization during
  -- the original implementation for removing PCA components is slow and was not optimized yet
• Minor differences to the original method (handling of UNK tokens)