TF dependency requires specific CPU commands

[bavadim]

I try:

• build docker Image on Intel Core computer
• use it on Xeon

I get following error:

2019-03-05 19:16:44.991 INFO in 'deeppavlov.core.data.simple_vocab'['simple_vocab'] at line 103: [loading vocabulary from /root/.deeppavlov/models/ner_rus/word.dict]
2019-03-05 19:16:45.81 INFO in 'deeppavlov.core.data.simple_vocab'['simple_vocab'] at line 103: [loading vocabulary from /root/.deeppavlov/models/ner_rus/tag.dict]
2019-03-05 19:16:45.84 INFO in 'deeppavlov.core.data.simple_vocab'['simple_vocab'] at line 103: [loading vocabulary from /root/.deeppavlov/models/ner_rus/char.dict]
2019-03-05 19:16:45.94 INFO in 'deeppavlov.models.embedders.fasttext_embedder'['fasttext_embedder'] at line 52: [loading fastText embeddings from /root/.deeppavlov/downloads/embeddings/lenta_lower_100.bin]
2019-03-05 19:16:49.243475: F tensorflow/core/platform/cpu_feature_guard.cc:37] The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine.

[mu-arkhipov]

It's an information message, not an error. Do you have any errors with tracebacks?

[bavadim]

It's error, check this: 2019-03-05 19:16:49.243475: F tensorflow/core/platform/cpu_feature_guard.cc:37] The TensorFlow library was compiled to use AVX instructions, but these aren't available on your machine. After this message my App hangs without any logs, i didn't investigate this issue in details. This problem appears in this code:

from deeppavlov import configs, build_model
ner = build_model(configs.ner.ner_rus, download=False)

[mu-arkhipov]

Then it is an internal TF problem. Look at this issue openai/gpt-2#34. In short, you can try using anaconda or compile TF yourself.