Expose DistributedSampler RNG seed argument

Relies on mosaicml/composer#3724.
janEbert · Nov 28, 2024 · 68c0d6c · 68c0d6c
1 parent d1f12f0
commit 68c0d6c
Showing 1 changed file with 1 addition and 0 deletions.
diff --git a/llmfoundry/data/finetuning/dataloader.py b/llmfoundry/data/finetuning/dataloader.py
@@ -336,6 +336,7 @@ def build_finetuning_dataloader(
             replication_factor if replication_factor > 1 else None,
             rank=dist.get_global_rank() //
             replication_factor if replication_factor > 1 else None,
+            seed=dataset_cfg.get('shuffle_seed', 0),
         )
 
     assert streaming_dataset is not None  # for pyright