quantize only seems to use 1 core?

[dcsan]

I'm using quantize with -thread 12 argument and the process only seems to use one core (ubuntu 16.0.4)

are there any other ways to speed up the process?
I'm guessing thread is an internal C thread, but should that be able to use > 1 core?

fasttext quantize \
    -output "${MODEL_FILE}" \
    -input "${LABEL_FILE}" \
    -dim 300 \
    -loss softmax \
    -qnorm \
    -thread 12 \
    -epoch 1

[EdouardGrave]

Hi @dcsan,

As of now, the quantization of models is not multi-threaded, and there are no ways to speed up the process.

Best,
Edouard.

[dcsan]

thanks for response! closing.

[rmc135]

A further 'gotcha' is that autotuned training with the -autotune-modelsize option will invoke quantization, so the entire run will quietly switch to using a single core only. It took me a while to figure this out. Perhaps the documentation could be updated to warn of this.