remove jax.device_put from imagenet test pipeline because it results in an OOM for subsequent training steps

Author: priyakasimbeg

algoperf/workloads/imagenet_resnet/imagenet_v2.py

@@ -6,10 +6,9 @@
 import functools
 from typing import Dict, Iterator, Tuple
 
-import jax
 import tensorflow_datasets as tfds
 
-from algoperf import data_utils, jax_sharding_utils, spec
+from algoperf import data_utils, spec
 from algoperf.workloads.imagenet_resnet.imagenet_jax import input_pipeline
 
 
@@ -47,10 +46,4 @@ def _decode_example(example: Dict[str, float]) -> Dict[str, float]:
   if framework == 'pytorch':
     it = map(data_utils.shard, it)
 
-  elif framework == 'jax':
-    f = functools.partial(
-      jax.device_put, device=jax_sharding_utils.get_batch_dim_sharding()
-    )
-    it = map(f, it)
-
   return it