diff --git a/tests/algorithms/testers/torch_dynamic.py b/tests/algorithms/testers/torch_dynamic.py
index a0c9c4da..3e376b1f 100644
--- a/tests/algorithms/testers/torch_dynamic.py
+++ b/tests/algorithms/testers/torch_dynamic.py
@@ -1,4 +1,7 @@
+from typing import Any
+
 from pruna.algorithms.torch_dynamic import TorchDynamic
+from pruna.engine.utils import get_device_type
 
 from .base_tester import AlgorithmTesterBase
 
@@ -11,3 +14,8 @@ class TestTorchDynamic(AlgorithmTesterBase):
     allow_pickle_files = False
     algorithm_class = TorchDynamic
     metrics = ["latency"]
+
+    def pre_smash_hook(self, model: Any) -> None:
+        """Ensure CPU dynamic quantized linear layers receive float32 activations."""
+        if get_device_type(model) == "cpu":
+            model.float()