diff --git a/vllm/nvfp4_cutedsl.py b/vllm/nvfp4_cutedsl.py
index e39cf8d4..a440a42f 100644
--- a/vllm/nvfp4_cutedsl.py
+++ b/vllm/nvfp4_cutedsl.py
@@ -86,6 +86,10 @@ class CuTeDSLMoERunner:
         self._max_chunks_per_expert = cutedsl_ceil_div(
             self.max_num_tokens * self.top_k, self.num_experts * 128
         )
+        import os
+        print(f"[CLAWMINE] Runner init: max_num_tokens={self.max_num_tokens} top_k={self.top_k} "
+              f"num_experts={self.num_experts} max_chunks={self._max_chunks_per_expert} "
+              f"pid={os.getpid()}", flush=True)
         self._buffers_allocated = False
 
     def _fill_token_indices(self):