diff --git a/tensorrt_llm/llmapi/rlhf_utils.py b/tensorrt_llm/llmapi/rlhf_utils.py
index 47135c765678..958529827025 100644
--- a/tensorrt_llm/llmapi/rlhf_utils.py
+++ b/tensorrt_llm/llmapi/rlhf_utils.py
@@ -124,6 +124,11 @@ def update_weights(self, ipc_handles: Optional[dict] = None):
                     self.engine.model_engine.model, weights, allow_partial_loading=True
                 )
                 del weights
+                # ipc_collect only reclaims mappings whose Python refs are
+                # already gone; force GC so cudaIpcOpenMemHandle storages are
+                # released before the sweep, or the IPC handle table saturates
+                # across repeated RPCs.
+                gc.collect()
                 torch.cuda.ipc_collect()
             else:
                 logger.info("Finalize update weights")