What is the source repo/original model used for Section 23 Text Completion test?

[Alexey]

Hello,

I've noticed that Section 23: Text Completion (from https://ai-benchmark.com/tests.html) code link leads to https://github.com/aymericdamien/Te...xamples/3_NeuralNetworks/recurrent_network.py file which is MNIST/digit recognition network. However in the description word2vec/text completion is mentioned.

Is this a typo by any chance? Does anyone know what was the original example/repository used to generate this test?

Thank you very much for help.

 

[Andrey Ignatov]

Is this a typo by any chance? Does anyone know what was the original example/repository used to generate this test?

No, that's not a typo - this model was adapted with some small modifications to the text completion task, being applied to word embeddings instead of images.

 

[Alexey]

Thank you. Do you know by any chance if these adaptations are available externally?

 

[Andrey Ignatov]

In terms of architecture, you can just check the TFLite models from AI Benchmark APK.

 

[Alexey]

Thank you, yes, that's an option. Ideally what I want to do (please let me know if this is fine according with the license terms) is to generate TFLite for exactly the same graph as we have in AI Benchmark / from the same source but with different TFLite converter settings and benchmark one against each other. That's why I was curios if we have original code available anywhere.