Make streaming use the correct number of unique samples with SP/TP #619
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description of changes:
The TP/SP PR assumed that streaming would track/receive an increased number of samples when using SP/TP. For example, with
replication
= 2, Composer would previously track the # of elapsed samples without taking into account replication, which meant that even if 100 unique samples had been seen in training, sincereplication
was 2, Composer would think that 200 samples had been seen, and pass that on to Streaming for resumption.This issue was previously addressed by dividing the sample offset on resumption. With Composer now correctly calculating the number of unique samples elapsed during training, this fix is not needed on the Streaming side.
This change means we also need to track # unique samples using
StreamingDataLoader
.Tested this with Shashank's help -- loss curves are overlapping on resumption.
Issue #, if available:
Merge Checklist:
Put an
x
without space in the boxes that apply. If you are unsure about any checklist, please don't hesitate to ask. We are here to help! This is simply a reminder of what we are going to look for before merging your pull request.General
Tests
pre-commit
on my change. (check out thepre-commit
section of prerequisites)