diff --git a/cutedsl/nvfp4_linear.py b/cutedsl/nvfp4_linear.py
index 7d36bf8f..2119176f 100644
--- a/cutedsl/nvfp4_linear.py
+++ b/cutedsl/nvfp4_linear.py
@@ -110,6 +110,7 @@ class CuTeDSLNvfp4Linear:
             self._activation_global_scale = gs
 
 
+    @torch.compiler.disable
     def run(self, hidden_states: torch.Tensor) -> torch.Tensor:
         """Forward: BF16 input → NVFP4 GEMM → BF16 output."""
         self._ensure_initialized()
diff --git a/cutedsl/shared_expert_pipeline.py b/cutedsl/shared_expert_pipeline.py
index 99913f1f..224ca1c7 100644
--- a/cutedsl/shared_expert_pipeline.py
+++ b/cutedsl/shared_expert_pipeline.py
@@ -274,6 +274,7 @@ class CuTeDSLSharedExpertRunner:
 
         return out[:num_tokens]
 
+    @torch.compiler.disable
     def run(self, hidden_states: torch.Tensor) -> torch.Tensor:
         """Full shared expert forward: L1 → SiLU → L2 → output."""
         self._ensure_initialized()
diff --git a/vllm/nvfp4_cutedsl.py b/vllm/nvfp4_cutedsl.py
index 979efa17..f1ce8b67 100644
--- a/vllm/nvfp4_cutedsl.py
+++ b/vllm/nvfp4_cutedsl.py
@@ -374,6 +374,7 @@ class CuTeDSLMoERunner:
         
 
 
+    @torch.compiler.disable
     def run(self, hidden_states, topk_weights, topk_ids, expert_indices=None):
         """Run the NVFP4 MoE forward pass.