streaming multipack for pretraining dataset #959
Merged
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
winglian
reviewed
Jan 4, 2024
|
Here's a patch file I used to test a c4 pretraiing dataset with tinnyllama. multigpu doesn't work currently w this since I think it needs a proper data collator to pad the samples to the same sequence length |
|
Would this streaming feature work with S3, GCS, Azure Blob Storage? |
winglian
force-pushed
the
pretraining_multipack
branch
from
02dc87f to
a5eb52e
Compare
|
This PR is ready for review and should resolve #1026. @mhenrichsen |
winglian
requested review from
NanoCode012,
casper-hansen,
mhenrichsen and
hamelsmu
winglian
force-pushed
the
pretraining_multipack
branch
from
eec349a to
2a49248
Compare
|
Confirmed working on single gpu. Currently fails on multi gpu. |
winglian
changed the title
djsaunde
pushed a commit
that referenced
this pull request
Dec 17, 2024
* [Feat] streaming multipack * WIP make continued pretraining work w multipack * fix up hadrcoding, lint * fix dict check * update test for updated pretraining multipack code * fix hardcoded data collator fix for multipack pretraining * fix the collator to be the max length for multipack pretraining * don't bother with latest tag for test * cleanup docker build/test --------- Co-authored-by: [email protected] <jinwonkim> Co-authored-by: Wing Lian <[email protected]>
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Outdated suggestions cannot be applied.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.
This pull request introduces support for multipacking in streaming pretraining datasets.
Due to the immense size of these datasets, traditional methods of loading them entirely into memory are not feasible.
The proposed solution aims to enhance efficiency and scalability.
need guide for making config.
this multipack does not use BatchSamplerDataCollatorForSeq2Seq, only use DataCollatorForSeq2Seq due to huggingface dataset map function.