feat: use SequentialSampler if curriculum_sampling is enabled with sample_packing
#2235
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
v-dicicco
changed the title
curriculum_sampling is enabled with sample_packingcurriculum_sampling is enabled with sample_packing
v-dicicco
changed the title
curriculum_sampling is enabled with sample_packingcurriculum_sampling is enabled with sample_packing
v-dicicco
changed the title
curriculum_sampling is enabled with sample_packingcurriculum_sampling is enabled with sample_packing
|
This is a good catch! I don't think the other params of Multipack should be impacted from this, but pinging @winglian to double check. |
winglian
approved these changes
Jan 7, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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
The
curriculum_samplingflag introduced in #1567 (cc @winglian) preserves dataset ordering by disabling shuffling, which simplifies experimenting with curriculum learning strategies. However, the flag is ignored if we usesample_packing. This PR sets the sampler to aSequentialSamplerif thecurriculum_samplingflag is enabled.How has this been tested?
I've verified that the DataLoader created in this way correctly yields batches with consecutive samples packed together. However, I'm not 100% sure whether other parameters of
MultipackBatchSamplerneed adjustment if we potentially don't have a shuffled dataset (e.g.,packing_efficiency_estimate).Please note that
MultipackBatchSampleris already used with aSequentialSamplerin_get_eval_sampler().