diff --git a/src/transformers/modeling_utils.py b/src/transformers/modeling_utils.py
index 962a8c016f0b6c..369788e7369888 100755
--- a/src/transformers/modeling_utils.py
+++ b/src/transformers/modeling_utils.py
@@ -1477,9 +1477,6 @@ def _from_config(cls, config, **kwargs):
         else:
             model = cls(config, **kwargs)
 
-        # Flag for if we init with `zero3`, add an attr to the model so we can check downstream for issues
-        model._transformers_zero3_init_used = is_deepspeed_zero3_enabled()
-
         # restore default dtype if it was modified
         if dtype_orig is not None:
             torch.set_default_dtype(dtype_orig)
@@ -3809,9 +3806,6 @@ def from_pretrained(
             # Let's make sure we don't run the init function of buffer modules
             model = cls(config, *model_args, **model_kwargs)
 
-        # If we init with `zero3`, add an attr to the model so we can check downstream for issues
-        model._transformers_zero3_init_used = is_deepspeed_zero3_enabled() and not is_quantized
-
         # make sure we use the model's config since the __init__ call might have copied it
         config = model.config
 
diff --git a/src/transformers/trainer.py b/src/transformers/trainer.py
index 59f0ed438bf7e4..68ba7babfc5dde 100755
--- a/src/transformers/trainer.py
+++ b/src/transformers/trainer.py
@@ -100,7 +100,6 @@
     get_model_param_count,
     get_module_class_from_name,
     get_parameter_names,
-    is_deepspeed_zero3_enabled,
     nested_concat,
     nested_detach,
     nested_numpify,
@@ -436,15 +435,6 @@ def __init__(
                 )
             self.model_init = model_init
 
-        # Will reach this branch if the user has
-        # 1. Used `.from_pretrained` or `.from_config` to initialize their model
-        # 2. Did not configure Zero-3 via `TrainingArguments` or `accelerate launch` beforehand
-        # New models init such as `MyModel()` will not hit this step
-        if is_deepspeed_zero3_enabled() and not getattr(model, "_transformers_zero3_init_used", True):
-            raise ValueError(
-                "Model was not initialized with `Zero-3` despite being configured for DeepSpeed Zero-3. Please re-initialize your model via `Model.from_pretrained(...)` or `Model.from_config(...)` after creating your `TrainingArguments`!"
-            )
-
         if model.__class__.__name__ in MODEL_MAPPING_NAMES:
             raise ValueError(
                 f"The model you have picked ({model.__class__.__name__}) cannot be used as is for training: it only "
diff --git a/tests/deepspeed/test_deepspeed.py b/tests/deepspeed/test_deepspeed.py
index 7b81ba40e47b01..b635833706d136 100644
--- a/tests/deepspeed/test_deepspeed.py
+++ b/tests/deepspeed/test_deepspeed.py
@@ -709,30 +709,34 @@ def test_gradient_accumulation(self, stage, dtype):
         # Relative difference. See the note above how to get identical loss on a small bs
         self.assertTrue((no_grad_accum_loss - yes_grad_accum_loss) / (no_grad_accum_loss + 1e-15) <= 1e-3)
 
-    def test_missed_zero3_init(self):
-        from transformers import Trainer  # noqa
-
-        with mockenv_context(**self.dist_env_1_gpu):
-            model = AutoModel.from_pretrained(T5_TINY)
-            training_args = TrainingArguments(
-                output_dir="./test_missed_zero3_init",
-                deepspeed=self.get_config_dict(ZERO3),
-            )
-            with self.assertRaises(
-                ValueError, msg="Model was not initialized with `Zero-3` despite being configured."
-            ):
-                _ = Trainer(
-                    model=model,
-                    args=training_args,
-                )
-            # Now do it properly, triggered from our `TrainingArguments` earlier
-            model = AutoModel.from_pretrained(T5_TINY)
-            trainer = Trainer(
-                model=model,
-                args=training_args,
-            )
-            assert trainer.is_deepspeed_enabled
-            assert model._transformers_zero3_init_used
+    # NOTE: Currently a disabled test. In the future we should re-enable it.
+    # Issue resolves around Zero-3 w/ DPO/TRL + DeepSpeed
+    # As well as Zero-3 inference
+    # Related PR: https://github.com/huggingface/transformers/pull/32299
+    # def test_missed_zero3_init(self):
+    #     from transformers import Trainer  # noqa
+
+    #     with mockenv_context(**self.dist_env_1_gpu):
+    #         model = AutoModel.from_pretrained(T5_TINY)
+    #         training_args = TrainingArguments(
+    #             output_dir="./test_missed_zero3_init",
+    #             deepspeed=self.get_config_dict(ZERO3),
+    #         )
+    #         with self.assertRaises(
+    #             ValueError, msg="Model was not initialized with `Zero-3` despite being configured."
+    #         ):
+    #             _ = Trainer(
+    #                 model=model,
+    #                 args=training_args,
+    #             )
+    #         # Now do it properly, triggered from our `TrainingArguments` earlier
+    #         model = AutoModel.from_pretrained(T5_TINY)
+    #         trainer = Trainer(
+    #             model=model,
+    #             args=training_args,
+    #         )
+    #         assert trainer.is_deepspeed_enabled
+    #         assert model._transformers_zero3_init_used
 
     def check_saved_checkpoints_deepspeed(self, output_dir, freq, total, stage, dtype):
         # adapted from TrainerIntegrationCommon.check_saved_checkpoints