notebook.community
Edit and run
Keep an eye on the following:
with tf.device('device:XLA_GPU:0')
In [ ]: cat /root/src/main/python/xla/train_linear_xla.py
cat /root/src/main/python/xla/train_linear_xla.py
The cell below will report
Aborted (core dumped)
but still generates useful info.
In [ ]: %%bash train_xla
%%bash train_xla
train_linear_xla.log
hlo_graph_*
https://www.tensorflow.org/performance/xla/operation_semantics