question int64

[dmitrysarov]

hello!
here

ctc-ocr-tensorflow/data.py

Line 41 in 6d4b0f4

indices = np.asarray(indices, dtype=np.int64)

you choose np.int64 precision, but be default for values you choose np.int32

ctc-ocr-tensorflow/data.py

Line 27 in 6d4b0f4

def sparse_tuple_from(self, sequences, dtype=np.int32):

is there any reason to not just choose int32 everywhere?
asking because suppose that we couldn't proceed graphs with int64 tensor's type on GPU

[stardut]

I found an answer in https://devtalk.nvidia.com/default/topic/948014/forward-looking-gpu-integer-performance

32-bit integer operations are good to choose than 64-bit on GPU.