Repository navigation
Conversation
karhu
force-pushed
the
chore/shared-frozen-splits
branch
5 times, most recently
from
November 17, 2026 23:58
1e4fe29 to
197c245
Compare
karhu
marked this pull request as ready for review
November 18, 2026 00:17
karhu
force-pushed
the
chore/shared-frozen-splits
branch
from
November 18, 2026 01:04
00bbd4e to
061aa8d
Compare
✅ CI checks passed
Updated for |
karhu
force-pushed
the
chore/shared-frozen-splits
branch
2 times, most recently
from
November 18, 2026 01:23
e2d1e84 to
74cdde6
Compare
RealLast
requested changes
Nov 18, 2026
RealLast
left a comment
Contributor
There was a problem hiding this comment.
@TimoStoff good to get your eyes on this one please
Imo this can be consolidated with existing split functionality, a predefined split is a splitter derived from common base class
RealLast
requested changes
Nov 18, 2026
karhu
force-pushed
the
chore/shared-frozen-splits
branch
4 times, most recently
from
November 20, 2026 17:30
e584050 to
16279e8
Compare
sleep-cot and m4-caption each had their own copy of a frozen split: the validation and test samples that a script wrote to a file, with the digest of the source file that the split was made from. FrozenSplits in timenet_connectors.splitting now owns this. It names samples by row position or by source id, reads plain JSON or JSON compressed with xz, and takes the pinned source digest as an argument. Both connectors use it. The real split files of both connectors give the same splits as before.
karhu
force-pushed
the
chore/shared-frozen-splits
branch
from
November 20, 2026 17:31
16279e8 to
5f537da
Compare
K becomes SampleKeyT, like RowT of the readers. The docstring no longer says that the split comes from a library the connector should not pin. It now says that the split depends on a library version, so that making it again at build time could change it.
PrecomputedSplits becomes PrecomputedSplitter, a Splitter like StratifiedSplitter. PrecomputedSplitter.from_file reads the split file and checks it against the pinned source digest. split() returns DatasetSplits. The file names a sample by its position, or by the key that the key function returns. The file format does not change. sleep_cot and m4_caption now loop over the DatasetSplits, like tsqa. So their records come split by split, each split in source order. M4 keeps one series class, M4CaptionSeries, and the reader gets split_series. The connector docs describe the new splitter.
Contributor
Author
|
Agreed, done. |
This branch has not been deployed
This file contains hidden or 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
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.
sleep_cotandm4_captioneach had their own copy of a precomputed split: the validation and test samples that a script wrote to a file, with the digest of the source file that the split was made from.PrecomputedSplitterintimenet_connectors.splittingnow owns this. It is aSplitter, likeStratifiedSplitter, and returnsDatasetSplits.PrecomputedSplitter.from_filereads plain JSON or JSON compressed with xz, and checks the split against the pinned source digest. The file names a sample by its position, or by the key that akeyfunction returns. The file format does not change.Both connectors use it and loop over the splits like
tsqa. So their records come split by split, each split in source order. The error messages ofsleep_cotfor a bad split file change wording.docs/connectors.mddescribes the new splitter.The tests for a bad split file are now in
test_precomputed_splits.py. Each connector keeps one test, which proves that the connector checks the split file against its own pinned digest.The real split files of both connectors give the same splits as before.