From 5f24c5bf4b670591d67585aabb4eee781ead63fe Mon Sep 17 00:00:00 2001
From: Atharvanair09 <atharvanair09.ns@gmail.com>
Date: Wed, 22 Jul 2026 21:26:46 +0530
Subject: [PATCH] fix(proxy): fall back to local tokenizer on provider token
 count errors

---
 litellm/proxy/proxy_server.py                 |  8 +-
 .../test_proxy_token_counter.py               | 76 +++++++++++++++++++
 2 files changed, 82 insertions(+), 2 deletions(-)

diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index 0d425fac1a0..dfa69248cbe 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -10781,9 +10781,13 @@ async def _try_provider_token_count(
             tools=tools,
             system=system,
         )
-    except httpx.HTTPStatusError as e:
+    except Exception as e:
         error_message = getattr(e, "message", None) or str(e)
-        status_code = getattr(e, "status_code", None) or e.response.status_code
+        status_code = getattr(e, "status_code", None)
+        if status_code is None and hasattr(e, "response") and hasattr(e.response, "status_code"):
+            status_code = e.response.status_code
+        status_code = status_code or 500
+
         if litellm.disable_token_counter is True:
             raise ProxyException(
                 message=error_message,
diff --git a/tests/proxy_unit_tests/test_proxy_token_counter.py b/tests/proxy_unit_tests/test_proxy_token_counter.py
index e7a25af2659..232aae0cb10 100644
--- a/tests/proxy_unit_tests/test_proxy_token_counter.py
+++ b/tests/proxy_unit_tests/test_proxy_token_counter.py
@@ -1446,3 +1446,79 @@ async def test_try_provider_token_count_http_error_disabled():
         assert exc_info.value.code == "500"
     finally:
         litellm.disable_token_counter = original_disable
+
+
+@pytest.mark.asyncio
+async def test_try_provider_token_count_generic_error_fallback():
+    """
+    Test that _try_provider_token_count catches a generic Exception 
+    and returns None (falling back to local tokenizer) when litellm.disable_token_counter != True.
+    """
+    from litellm.proxy.proxy_server import _try_provider_token_count
+    from unittest.mock import AsyncMock, MagicMock
+    import litellm
+
+    original_disable = getattr(litellm, "disable_token_counter", False)
+    litellm.disable_token_counter = False
+
+    try:
+        mock_provider_counter = MagicMock()
+        mock_provider_counter.should_use_token_counting_api.return_value = True
+        
+        # Make count_tokens raise a generic Exception
+        error = ValueError("Some weird provider error")
+        mock_provider_counter.count_tokens = AsyncMock(side_effect=error)
+
+        result = await _try_provider_token_count(
+            provider_counter=mock_provider_counter,
+            custom_llm_provider="openai",
+            model_to_use="gpt-4",
+            messages=[{"role": "user", "content": "hello"}],
+            contents=None,
+            deployment=None,
+            request_model="gpt-4"
+        )
+        
+        # It should return None on generic Exception instead of raising Exception
+        assert result is None
+    finally:
+        litellm.disable_token_counter = original_disable
+
+
+@pytest.mark.asyncio
+async def test_try_provider_token_count_generic_error_disabled():
+    """
+    Test that _try_provider_token_count raises ProxyException on generic Exception
+    when litellm.disable_token_counter == True.
+    """
+    from litellm.proxy.proxy_server import _try_provider_token_count
+    from litellm.proxy._types import ProxyException
+    from unittest.mock import AsyncMock, MagicMock
+    import litellm
+
+    original_disable = getattr(litellm, "disable_token_counter", False)
+    litellm.disable_token_counter = True
+
+    try:
+        mock_provider_counter = MagicMock()
+        mock_provider_counter.should_use_token_counting_api.return_value = True
+        
+        # Make count_tokens raise a generic Exception
+        error = ValueError("Some weird provider error")
+        mock_provider_counter.count_tokens = AsyncMock(side_effect=error)
+
+        with pytest.raises(ProxyException) as exc_info:
+            await _try_provider_token_count(
+                provider_counter=mock_provider_counter,
+                custom_llm_provider="openai",
+                model_to_use="gpt-4",
+                messages=[{"role": "user", "content": "hello"}],
+                contents=None,
+                deployment=None,
+                request_model="gpt-4"
+            )
+        
+        # Default status code for generic exceptions is 500
+        assert exc_info.value.code == "500"
+    finally:
+        litellm.disable_token_counter = original_disable