diff --git a/dsv4/layers/linear.py b/dsv4/layers/linear.py
index c67cadad..fb007010 100644
--- a/dsv4/layers/linear.py
+++ b/dsv4/layers/linear.py
@@ -176,6 +176,7 @@ class Nvfp4Linear:
             x_fp4, x_sf, gsa_gpu = quantize_nvfp4_gpu_fused(hidden_states)
             self._gsa_buf.copy_(gsa_gpu[:1].reshape(1))  # GPU → GPU, no sync
         else:
+            from dsv4.ops.quantize import quantize_nvfp4_gpu
             self._gsa_buf.fill_(self._activation_global_scale)
             x_fp4, x_sf = quantize_nvfp4_gpu(hidden_states, self._activation_global_scale)