Working on the repurposed Dell r710 server I got, I mostly want to run Machine Learning algorithms on it. Specifically, I wanted to run the Universal Sentence Encoder to generate sentence embeddings. But the pip installed version of Tensorflow doesn’t work with the E5660 CPU I have because it requires AVX CPU instruction set. Without it, it will just give me Illegal instruction (core dumped) message. I can kind of get around it by using the Anaconda version of Tensorflow. One exception I found is that the multi-language pre-trained model require tensorflow_text package. This packages is not compatible with the Ananconda version of TensorFlow for some reason. So once again, I’m left with the only option for English text and cannot compare sentence embedding across different languages.