diff --git a/egs/librispeech/ASR/tdnn_lstm_ctc/asr_datamodule.py b/egs/librispeech/ASR/tdnn_lstm_ctc/asr_datamodule.py
index dc957fd04..8dd1459ca 100644
--- a/egs/librispeech/ASR/tdnn_lstm_ctc/asr_datamodule.py
+++ b/egs/librispeech/ASR/tdnn_lstm_ctc/asr_datamodule.py
@@ -42,11 +42,11 @@ from icefall.utils import str2bool
 
 
 class _SeedWorkers:
+    def __init__(self, seed: int):
+        self.seed = seed
+
     def __call__(self, worker_id: int):
-        # 'seed' is derived from the current random state, which will have
-        # previously been set in the main process.
-        seed = torch.randint(0, 100000, ()).item()
-        fix_random_seed(seed + worker_id)
+        fix_random_seed(self.seed + worker_id)
 
 
 class LibriSpeechAsrDataModule:
@@ -311,13 +311,18 @@ class LibriSpeechAsrDataModule:
             logging.info("Loading sampler state dict")
             train_sampler.load_state_dict(sampler_state_dict)
 
+        # 'seed' is derived from the current random state, which will have
+        # previously been set in the main process.
+        seed = torch.randint(0, 100000, ()).item()
+        worker_init_fn = _SeedWorkers(seed)
+
         train_dl = DataLoader(
             train,
             sampler=train_sampler,
             batch_size=None,
             num_workers=self.args.num_workers,
             persistent_workers=False,
-            worker_init_fn=_SeedWorkers(),
+            worker_init_fn=worker_init_fn,
         )
 
         return train_dl
diff --git a/egs/librispeech/ASR/transducer_stateless_multi_datasets/asr_datamodule.py b/egs/librispeech/ASR/transducer_stateless_multi_datasets/asr_datamodule.py
index c3aaea782..c6cf739fb 100644
--- a/egs/librispeech/ASR/transducer_stateless_multi_datasets/asr_datamodule.py
+++ b/egs/librispeech/ASR/transducer_stateless_multi_datasets/asr_datamodule.py
@@ -42,11 +42,11 @@ from icefall.utils import str2bool
 
 
 class _SeedWorkers:
+    def __init__(self, seed: int):
+        self.seed = seed
+
     def __call__(self, worker_id: int):
-        # 'seed' is derived from the current random state, which will have
-        # previously been set in the main process.
-        seed = torch.randint(0, 100000, ()).item()
-        fix_random_seed(seed + worker_id)
+        fix_random_seed(self.seed + worker_id)
 
 
 class AsrDataModule:
@@ -264,13 +264,18 @@ class AsrDataModule:
 
         logging.info("About to create train dataloader")
 
+        # 'seed' is derived from the current random state, which will have
+        # previously been set in the main process.
+        seed = torch.randint(0, 100000, ()).item()
+        worker_init_fn = _SeedWorkers(seed)
+
         train_dl = DataLoader(
             train,
             sampler=train_sampler,
             batch_size=None,
             num_workers=self.args.num_workers,
             persistent_workers=False,
-            worker_init_fn=_SeedWorkers(),
+            worker_init_fn=worker_init_fn,
         )
         return train_dl