I have a model which gives me word embeddings as a sequence, say 500 tokens as a dimension of 1024, so my output is a tensor of the shape of [500, 1024].
Now I want to model the words into another tensor which has to put words in sentences, i.e. ith sentence and jth sentence Wordss[i,j] = tensors, i.e. Tensor of shape [None, None,1024] or a tensor of fixed size not at all equal to any 500 shapes say e.g. [24,37,1024]
I'm looking for something like a temporary tensor like this where I can keep a tensor only for a particular input epoch.
tensorflow/tensorflow/python/ops/math_ops.py Line 1525 in 5df4c71 var = gen_state_ops._temporary_variable(shape=shape, dtype=tensor_dtype)