yaroslavvb / stuff Goto Github PK

just as following
_, serialized_example = reader.read_up_to(filename_queue, enqueue_many_size *batch_size)

When I run tf.svd on some matrices, I come across the same issue.

Do you have any suggestions to address it?

I've been trying to use the linearize function, however it won't run the model - it just hangs after creating the contexts.

I've taken out the staging areas to ensure it isn't them, is linearize still being updated or has tensorflow now implemented your suggestions?

System:
Ubuntu 16.04
TF 1.4

Helli @yaroslavvb,

In https://stackoverflow.com/questions/41804380/testing-gpu-with-tensorflow-matrix-multiplication, you claimed that your matmul_benchmark.py avoided common pitfalls. Can you please list them out ?

I can see that you have,

1. Avoided counting time take to transfer data to the device by making the first call outside the timing region. But, is it guaranteed that Tensorflow will not move data back to the host unless it is accessed ?
2. Averaging the time take over many iterations.

Thanks,
Sanjay

Hi, how do you modify your matmul.py to TF2?
https://github.com/yaroslavvb/stuff/blob/master/matmul_benchmark.py

I tried and replaced opt_level=tf.OptimizerOptions.L0 with the following
tf.compat.v1.OptimizerOptions(opt_level=-1)

but there is huge variation from run to run.