OpenNMT Forum

Question about Intel MKL

A beginner question about Intel MKL. During the compiling step, Intel MKL (oneapi) would be installed to build the ctranslate2. However it wasn’t installed for the runtime docker container. CTranslate2/Dockerfile.ubuntu-gpu at 3190be529d55d1e4d4ca246bd2a9cf405107bb6d · OpenNMT/CTranslate2 · GitHub

Then what does the runtime mean below to use Intel MKL? Thanks

By default, the runtime automatically decides whether to use Intel MKL or not based on the CPU vendor.

Intel MKL is integrated in all prebuilt binaries of CTranslate2. It is installed in the first stage of the Dockerfile and linked statically to CTranslate2:

Then at runtime we only call Intel MKL when running on an Intel CPU.

1 Like