--- /tmp/evalscope-sync/src-v191/evalscope/utils/function_utils.py	2026-07-21 03:18:18.000000000 +0000
+++ /data1/syy/evalscope/evalstone/evalscope/evalscope/utils/function_utils.py	2026-09-01 03:48:30.955904239 +0000
@@ -68,12 +68,22 @@
     return wrapper
 
 
-def retry_call(func, *args, retries=3, sleep_interval=0, **kwargs):
-    """Function that retries a function call up to `retries` times if an exception occurs."""
+def retry_call(func, *args, retries=3, sleep_interval=0, on_attempt=None, **kwargs):
+    """Function that retries a function call up to `retries` times if an exception occurs.
+
+    ``on_attempt(ok, exc=None)`` is invoked once per try (including retries) so
+    callers can count vendor HTTP success/failure independently of the final
+    ``generate()`` outcome.
+    """
     for attempt in range(retries):
         try:
-            return func(*args, **kwargs)
+            result = func(*args, **kwargs)
+            if on_attempt is not None:
+                on_attempt(True, None)
+            return result
         except Exception as e:
+            if on_attempt is not None:
+                on_attempt(False, e)
             if attempt < retries - 1:
                 if sleep_interval > 0:
                     logger.warning(f'Attempt {attempt + 1} / {retries} failed: {e}. Retrying...')
@@ -83,13 +93,23 @@
 
 
 async def async_retry_call(
-    func: Callable[..., Awaitable[T]], *args, retries: int = 3, sleep_interval: float = 0, **kwargs
+    func: Callable[..., Awaitable[T]],
+    *args,
+    retries: int = 3,
+    sleep_interval: float = 0,
+    on_attempt=None,
+    **kwargs,
 ) -> T:
     """Async version of retry_call. Retries an async function call up to `retries` times if an exception occurs."""
     for attempt in range(retries):
         try:
-            return await func(*args, **kwargs)
+            result = await func(*args, **kwargs)
+            if on_attempt is not None:
+                on_attempt(True, None)
+            return result
         except Exception as e:
+            if on_attempt is not None:
+                on_attempt(False, e)
             if attempt < retries - 1:
                 if sleep_interval > 0:
                     logger.warning(f'Attempt {attempt + 1} / {retries} failed: {e}. Retrying...')