You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently evolven3fit_new evolves the PDFs sequentially both in Q2 and replicas. For many replicas this means that evolution can take up to several hours while even for the usual 100 replica fits it generally will take over 20 minutes for the evolution to complete.
Whenever resources are available (which they are on any clusters) it would be nice to have the option to parallelize parts of the the evolution to speed up the computation.
The text was updated successfully, but these errors were encountered:
As per the discussion w @felixhekhorn, the necessary change here is for eko to allow for more than one replica at once. To first approximation it should be trivial (the EKO x PDF contraction should have one extra axis for the number of replicas).
Then of course the changes will need to be propagated to evolven3fit and there might also be collateral damage somewhere.
Currently
evolven3fit_new
evolves the PDFs sequentially both in Q2 and replicas. For many replicas this means that evolution can take up to several hours while even for the usual 100 replica fits it generally will take over 20 minutes for the evolution to complete.Whenever resources are available (which they are on any clusters) it would be nice to have the option to parallelize parts of the the evolution to speed up the computation.
The text was updated successfully, but these errors were encountered: