Why I found that the generation of embedding is very slow?

[zhuchenxi]

I found that in CPU mode, the speed of the generation of embedding is about 26 sentences per hour. Is that slow or normal?

[hoschwenk]

I haven't done speed test on CPU, but 26 sentences per hour is definitely by way too slow.
Are yo using multiple threads and bunch mode ?

[hoschwenk]

Any new insights ?

[ShunChi100]

I found that setting max_sentences to some an int number helps, e.g. 128 (depending on your GPU memory) in the SentenceEncoder(). Otherwise, SentenceEncoder() treat one sentence as one batch.

[ericmclachlan]

@hoschwenk : I'm fairly new to using LASER (which is awesome!). You mention multiple threads and bunch mode... Is there any documentation you can point me to help me take advantage of these features? The FLASK API provided by the docker deployment seems to be strictly single-threaded.
TIA