diff --git a/contrastive-pretraining/tests/test_vision_encoder.py b/contrastive-pretraining/tests/test_vision_encoder.py
index 316f7b9..16f866b 100644
--- a/contrastive-pretraining/tests/test_vision_encoder.py
+++ b/contrastive-pretraining/tests/test_vision_encoder.py
@@ -3,6 +3,8 @@
 import torch
 import torch.nn as nn
 from vision_encoder.vjepa_encoder import ResidualTemporalDownsample
+from vision_encoder.optimizer import get_optimizer
+
 
 
 class TestResidualTemporalDownsample:
@@ -143,8 +145,20 @@ class TestOptimizerImport:
     """Test that the optimizer utility from vision_encoder works."""
 
     def test_get_optimizer_creates_adam(self):
-        from vision_encoder.optimizer import get_optimizer
         params = nn.Linear(10, 10).parameters()
         opt = get_optimizer(set(params), lr=1e-3, wd=0.01)
         assert opt is not None
         assert len(opt.param_groups) > 0
+
+    def test_get_optimizer_zero_weight_decay(self):
+        from torch.optim import Adam
+        params = nn.Linear(10, 10).parameters()
+        opt = get_optimizer(set(params), lr=1e-3, wd=0)
+        assert isinstance(opt, Adam)
+
+    def test_get_optimizer_filter_by_requires_grad(self):
+        p1 = nn.Parameter(torch.randn(10, 10), requires_grad=True)
+        p2 = nn.Parameter(torch.randn(10, 10), requires_grad=False)
+        opt = get_optimizer(params=[p1, p2], lr=1e-3, filter_by_requires_grad=True)
+        all_params = [p for group in opt.param_groups for p in group['params']]
+        assert all_params == [p1]