From ff3d9018027ad8f8e47f4f0c246364f8887b520d Mon Sep 17 00:00:00 2001
From: janEbert <janpublicebert@posteo.net>
Date: Thu, 5 Dec 2024 16:03:16 +0000
Subject: [PATCH] Expose `DistributedSampler` RNG seed argument (#1677)

Co-authored-by: Mihir Patel <mihir.v.patel7@gmail.com>
Co-authored-by: Saaketh Narayan <narayan.saaketh@gmail.com>
---
 llmfoundry/data/finetuning/dataloader.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/llmfoundry/data/finetuning/dataloader.py b/llmfoundry/data/finetuning/dataloader.py
index 661729ff8a..fce694f160 100644
--- 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