diff --git a/unirl/rollout/engine/fastvideo/engine.py b/unirl/rollout/engine/fastvideo/engine.py
index 5fb1fedc1..24cdc24ab 100644
--- a/unirl/rollout/engine/fastvideo/engine.py
+++ b/unirl/rollout/engine/fastvideo/engine.py
@@ -181,6 +181,7 @@ def __init__(
         )
 
         self._version = 0
+        self._weight_update_calls = 0
         self._generate_lock = threading.Lock()
         self._shutdown_lock = threading.Lock()
         self._shutdown_requested = False
@@ -620,7 +621,7 @@ def update_weights_from_path(self, checkpoint_path: str, *, track_prefix: str =
             require(self._generator is not None, "fastvideo engine is offloaded/not initialized")
             self._generator.update_transformer_weights_from_path(checkpoint_path)
             self._last_weights_path = checkpoint_path
-            self._version += 1
+            self._weight_update_calls += 1
             logger.info("fastvideo transformer weights updated from %s", checkpoint_path)
 
 
diff --git a/unirl/rollout/engine/sglang/engine.py b/unirl/rollout/engine/sglang/engine.py
index 06557a160..a885aedef 100644
--- a/unirl/rollout/engine/sglang/engine.py
+++ b/unirl/rollout/engine/sglang/engine.py
@@ -152,6 +152,7 @@ def __init__(
         )
 
         self._version = 0
+        self._weight_update_calls = 0
 
     def _prepare_generation(self, sample: Sample) -> Any:
         require(
@@ -282,7 +283,7 @@ def update_weights_from_tensor(
             load_format=load_format,
             flush_cache=flush_cache,
         )
-        self._version += 1
+        self._weight_update_calls += 1
 
     def init_weights_update_group(
         self,
@@ -329,7 +330,7 @@ def update_weights_from_distributed(
             group_name=group_name,
             flush_cache=flush_cache,
         )
-        self._version += 1
+        self._weight_update_calls += 1
 
     def destroy_weights_update_group(
         self,
diff --git a/unirl/rollout/engine/sglang_diffusion/engine.py b/unirl/rollout/engine/sglang_diffusion/engine.py
index a6a8c859e..2ab166119 100644
--- a/unirl/rollout/engine/sglang_diffusion/engine.py
+++ b/unirl/rollout/engine/sglang_diffusion/engine.py
@@ -98,6 +98,7 @@ def __init__(
         self.schedule_policy = self.adapter.schedule_policy()
 
         self._version = 0
+        self._weight_update_calls = 0
         self._generate_lock = threading.Lock()
         self._shutdown_lock = threading.Lock()
         self._shutdown_requested = False
@@ -236,7 +237,7 @@ def update_weights_from_tensor(
             load_format=load_format,
             flush_cache=flush_cache,
         )
-        self._version += 1
+        self._weight_update_calls += 1
 
     def init_weights_update_group(
         self,
@@ -279,7 +280,7 @@ def update_weights_from_distributed(
             target_modules=target_modules,
             flush_cache=flush_cache,
         )
-        self._version += 1
+        self._weight_update_calls += 1
 
     def destroy_weights_update_group(
         self,
diff --git a/unirl/rollout/engine/vllm_omni/engine.py b/unirl/rollout/engine/vllm_omni/engine.py
index 194b38139..bf19ceeb9 100644
--- a/unirl/rollout/engine/vllm_omni/engine.py
+++ b/unirl/rollout/engine/vllm_omni/engine.py
@@ -38,6 +38,7 @@ def __init__(
     ) -> None:
         self.cfg = config
         self._version = 0
+        self._weight_update_calls = 0
         self._generate_lock = threading.Lock()
         self._shutdown_lock = threading.Lock()
         self._shutdown_requested = False
@@ -256,7 +257,7 @@ def update_weights_from_ipc(
             use_shm=use_shm,
             replica_rank=replica_rank,
         )
-        self._version += 1
+        self._weight_update_calls += 1
 
     def init_weights_update_group(
         self,
@@ -299,7 +300,7 @@ def update_weights_from_distributed(
             target_modules=target_modules,
             flush_cache=flush_cache,
         )
-        self._version += 1
+        self._weight_update_calls += 1
 
     def destroy_weights_update_group(
         self,
@@ -326,7 +327,7 @@ def update_weights_from_tensor(
             load_format=load_format,
             flush_cache=flush_cache,
         )
-        self._version += 1
+        self._weight_update_calls += 1
 
     def set_lora_from_tensors(
         self,