diff --git a/python/kubeflow/trainer/api/trainer_client_test.py b/python/kubeflow/trainer/api/trainer_client_test.py
index e3f56c69..9c3dbb39 100644
--- a/python/kubeflow/trainer/api/trainer_client_test.py
+++ b/python/kubeflow/trainer/api/trainer_client_test.py
@@ -450,9 +450,6 @@ def create_cluster_training_runtime(
         metadata=models.IoK8sApimachineryPkgApisMetaV1ObjectMeta(
             name=name,
             namespace=namespace,
-            labels={
-                "trainer.kubeflow.org/accelerator": "gpu-tesla-v100-16gb",
-            },
         ),
         spec=models.TrainerV1alpha1TrainingRuntimeSpec(
             mlPolicy=models.TrainerV1alpha1MLPolicy(
@@ -514,7 +511,6 @@ def create_runtime_type(
             trainer_type=types.TrainerType.CUSTOM_TRAINER,
             framework=types.Framework.TORCH,
             entrypoint=[constants.TORCH_ENTRYPOINT],
-            accelerator="gpu-tesla-v100-16gb",
             accelerator_count=4,
         ),
     )
@@ -541,7 +537,6 @@ def get_train_job_data_type(
                 trainer_type=types.TrainerType.CUSTOM_TRAINER,
                 framework=types.Framework.TORCH,
                 entrypoint=["torchrun"],
-                accelerator="gpu-tesla-v100-16gb",
                 accelerator_count=4,
             ),
         ),
diff --git a/python/kubeflow/trainer/constants/constants.py b/python/kubeflow/trainer/constants/constants.py
index fbcd24c1..d21da279 100644
--- a/python/kubeflow/trainer/constants/constants.py
+++ b/python/kubeflow/trainer/constants/constants.py
@@ -72,10 +72,6 @@
 # single VM where distributed training code is executed.
 NODE = "node"
 
-# The label key to identify the accelerator type for model training (e.g. GPU-Tesla-V100-16GB).
-# TODO: Potentially, we should take this from the Node selectors.
-ACCELERATOR_LABEL = "trainer.kubeflow.org/accelerator"
-
 # Unknown indicates that the value can't be identified.
 UNKNOWN = "Unknown"
 
diff --git a/python/kubeflow/trainer/types/types.py b/python/kubeflow/trainer/types/types.py
index d3390957..0d9c5ed9 100644
--- a/python/kubeflow/trainer/types/types.py
+++ b/python/kubeflow/trainer/types/types.py
@@ -167,7 +167,6 @@ class Trainer:
     trainer_type: TrainerType
     framework: Framework
     entrypoint: Optional[List[str]] = None
-    accelerator: str = constants.UNKNOWN
     accelerator_count: Union[str, float, int] = constants.UNKNOWN
 
 
diff --git a/python/kubeflow/trainer/utils/utils.py b/python/kubeflow/trainer/utils/utils.py
index 4225bb8a..85783732 100644
--- a/python/kubeflow/trainer/utils/utils.py
+++ b/python/kubeflow/trainer/utils/utils.py
@@ -140,14 +140,6 @@ def get_runtime_trainer(
     if isinstance(trainer.accelerator_count, (int, float)) and ml_policy.num_nodes:
         trainer.accelerator_count *= ml_policy.num_nodes
 
-    # TODO (andreyvelich): Currently, we get the accelerator type from
-    # the runtime labels.
-    if (
-        runtime_metadata.labels
-        and constants.ACCELERATOR_LABEL in runtime_metadata.labels
-    ):
-        trainer.accelerator = runtime_metadata.labels[constants.ACCELERATOR_LABEL]
-
     return trainer