don't sort multipack sampler (#2657)

* don't sort multipack sampler

* increased packing efficiency increases loss

---------

Co-authored-by: Wing Lian <wing@axolotl.ai>
This commit is contained in:
Dan Saunders
2025-05-09 20:28:58 -04:00
committed by Wing Lian
parent 8cda9e93c1
commit 27fec49083
3 changed files with 5 additions and 8 deletions

View File

@@ -106,3 +106,4 @@ class TestBatchedSamplerPacking:
original_idxs = set(range(len(train_dataset)))
assert original_idxs == set(batch_idxs)
assert len(batch_idxs) == len(set(batch_idxs))