diff --git a/src/accelerate/optimizer.py b/src/accelerate/optimizer.py
index c1f8faa1543..a318f2f1d1d 100644
--- a/src/accelerate/optimizer.py
+++ b/src/accelerate/optimizer.py
@@ -141,6 +141,13 @@ def eval(self):
         """
         if hasattr(self.optimizer, "eval") and callable(self.optimizer.eval):
             self.optimizer.eval()
+        elif (
+            hasattr(self.optimizer, "optimizer")
+            and hasattr(self.optimizer.optimizer, "eval")
+            and callable(self.optimizer.optimizer.eval)
+        ):
+            # the deepspeed optimizer further wraps the optimizer
+            self.optimizer.optimizer.eval()
 
     def step(self, closure=None):
         if is_lomo_available():
diff --git a/tests/test_optimizer.py b/tests/test_optimizer.py
index 8bb324f0ec5..01665563d0b 100644
--- a/tests/test_optimizer.py
+++ b/tests/test_optimizer.py
@@ -17,6 +17,7 @@
 import torch
 
 from accelerate import Accelerator
+from accelerate.optimizer import AcceleratedOptimizer
 from accelerate.test_utils import require_cpu, require_fp16, require_non_cpu
 from accelerate.test_utils.testing import AccelerateTestCase
 
@@ -34,6 +35,36 @@ def test_accelerated_optimizer_pickling(self):
             self.fail(f"Accelerated optimizer pickling failed with {e}")
 
 
+class OptimizerModeTester(AccelerateTestCase):
+    def test_accelerated_optimizer_eval_reaches_deepspeed_wrapper(self):
+        class ScheduleFreeLikeOptimizer(torch.optim.SGD):
+            def __init__(self, *args, **kwargs):
+                super().__init__(*args, **kwargs)
+                self.mode = None
+
+            def eval(self):
+                self.mode = "eval"
+
+        class DeepSpeedLikeOptimizerWrapper:
+            def __init__(self, optimizer):
+                self.optimizer = optimizer
+
+            def state_dict(self):
+                return self.optimizer.state_dict()
+
+            def load_state_dict(self, state_dict):
+                self.optimizer.load_state_dict(state_dict)
+
+        Accelerator()
+        model = torch.nn.Linear(10, 10)
+        inner = ScheduleFreeLikeOptimizer(model.parameters(), lr=0.1)
+        optimizer = AcceleratedOptimizer(DeepSpeedLikeOptimizerWrapper(inner), device_placement=False)
+
+        optimizer.eval()
+
+        assert inner.mode == "eval"
+
+
 @require_fp16
 @require_non_cpu
 class OptimizerTester(AccelerateTestCase):