mlcommons / logging

MLPerf™ logging library
https://mlcommons.org/en/groups/best-practices-benchmark-infra
Apache License 2.0
29 stars 46 forks source link

Why is RCP pruned based on min instead of mean values? #367

Open nv-rborkar opened 2 months ago

nv-rborkar commented 2 months ago

Analyze & fix before v4.1

pgmpablo157321 commented 1 month ago

Solution for 4.0: https://github.com/mlcommons/logging/tree/training_v4.0

pgmpablo157321 commented 1 month ago

All RCPs graph: RCPs_graph RCPs pruned by Mean: RCPs_pruned_by_mean

pgmpablo157321 commented 1 month ago

All RCPs min epochs graph: RCPs_graph_min RCPs pruned by Min epochs: RCPs_pruned_by_min

hiwotadese commented 1 month ago

@pgmpablo157321 Can you add mean and min graph per benchmark in the same graph?

pgmpablo157321 commented 1 month ago

Merged

pgmpablo157321 commented 5 days ago

RCPs pruned by Min epochs: RCPs_pruned_by_median

hiwotadese commented 4 days ago

07/18 training wg agreed to use RCP pruning with mean.