You are viewing a plain text version of this content. The canonical link for it is here.

Posted to issues@madlib.apache.org by "Nikhil Kak (Jira)" <ji...@apache.org> on 2021/01/22 03:13:00 UTC

[jira] [Created] (MADLIB-1463) DL: Fix gpu_memory_fraction for distribution_policy != 'all_segments'

Nikhil Kak created MADLIB-1463:
----------------------------------

             Summary: DL: Fix gpu_memory_fraction for distribution_policy != 'all_segments'
                 Key: MADLIB-1463
                 URL: https://issues.apache.org/jira/browse/MADLIB-1463
             Project: Apache MADlib
          Issue Type: Improvement
          Components: Deep Learning
            Reporter: Nikhil Kak
             Fix For: v1.18.0


The calculation of `gpu_mem_fraction` assumes that num_segments = all_segments which is not always the case. The user can pass in a distribution rules table to input preprocessor and the data can be distributed to less segments than the total number of segments on the cluster.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)