reductionista commented on a change in pull request #462: DL: Add asymmetric
cluster support for fit and evaluate
URL: https://github.com/apache/madlib/pull/462#discussion_r352943584
##########
File path: src/ports/postgres/modules/deep_learning/madlib_keras_wrapper.py_in
##########
@@ -90,15 +110,15 @@ def clear_keras_session(sess = None):
sess.close()
-def get_gpu_memory_fraction(gpus_per_host, segments_per_host):
+def get_gpu_memory_fraction(gpu_count, segments_per_host):
"""
We cap the gpu memory usage to 90% of the total available gpu memory.
This 90% is evenly distributed among the segments per gpu.
- :param gpus_per_host:
+ :param gpu_count:
:param segments_per_host:
:return:
"""
- return 0.9 / ceil(1.0 * segments_per_host / gpus_per_host)
+ return 0.9 / ceil(1.0 * segments_per_host / gpu_count)
Review comment:
Here we are returning 90% x gpus per segment,
where gpus per segment = gpus per host / segments_per_host
What we're calling gpu_count (and earlier "gpus_per_seg") is the "gpus per
host" in above equation.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
