In nlp.paragraghvectors, code ParagraghVectorsInferenceExample says “In this example we load previously built model, and pass raw sentences, probably never seen before, to get their vector representation.”, so i ran ParagraghVectorsClassifierExample first, it give out the classification.
then i ran ParagraghVectorsInferenceExample, there is errors
I think what you are seeing here is an instance of the bug reported here:
It is already fixed in the current SNAPSHOT version. But the linked thread also contains a workaround that you can use without switching to snapshots.
Do you means this problem is caused by Golve?
but this file don’t have relationship with Glove
No, in the linked thread I wrote:
It appears to be a problem with the mkldnn implementation. If you don’t want to go back and forth between versions, you can also disable mkldnn with
Nd4j.getEnvironment().allowHelpers(false);before loading and save the file in another format that way.
And link to: ND4J: concat op fails with 3073 or more inputs on CPU · Issue #8961 · deeplearning4j/deeplearning4j · GitHub
The bug is with vstack using a MLKDNN function that only supports up to 3073 tensors to stack.
And the screenshot you’ve posted shows that it has 97406 inputs, so I’m guessing that this is the same issue, and you should be able to workaround it with the same workaround.
