[bugfix] slice rows per rank and add data_config.min_batch_size - #677

Merged

tiankongdeguiji merged 9 commits into

alibaba:masterfrom

tiankongdeguiji:bugfix/rank-level-slicing

Sep 20, 2026

Commits on Sep 19, 2026

[bugfix] slice rows per rank and add data_config.min_batch_size

tiankongdeguiji
and
claude
committed
332d1a8

Commits on Sep 20, 2026

[bugfix] pass world_size to to_batches instead of inferring it

tiankongdeguiji
and
claude
committed
6a17bd9
[bugfix] apply drop_remainder and min_batch_size to training only

tiankongdeguiji
and
claude
committed
05d69c3
[bugfix] plan a rank's sources as one stream and cache odps session row counts

tiankongdeguiji
and
claude
committed
91e147d
[chore] bump version to 1.4.12

tiankongdeguiji
and
claude
committed
b1bad3c
Merge remote-tracking branch 'origin/master' into bugfix/rank-level-slicing
tiankongdeguiji
committed
0207b22
[bugfix] fetch odps session counts per rank and tighten planner docs and tests

tiankongdeguiji
and
claude
committed
7a7e9ce
[refactor] read one contiguous chunk per worker and source

tiankongdeguiji
and
claude
committed
0373c9d
[bugfix] keep unstarted leading rows of a source when resuming

tiankongdeguiji
and
claude
committed
1008de9