Reproducibility bug on run_embeddings method

[DraceniY]

The run_embeddings has reproducibility issue in long text, when it runs in different machines.

I put this reproducibility_summarization.zip file with input scientific article and my embedding output for comparison, I set random.seed in the init function of the class BertEmbedding but still had differences in the embeddings between two machines.

Please could you fix this issue, it will be very helpful.

Thanks.