fix(count_tokens): verify azure provider without counter

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/proxy/common_utils/token_count_settings.py

@@ -123,7 +123,7 @@ def unreliable_local_estimate(
     tools: Sequence[object] | None,
 ) -> str | None:
     if _has_document_content_block(messages):
-        return f"{model} has no server-side count_tokens and LiteLLM cannot count document content locally"
+        return f"{model} returned no server-side token count and LiteLLM cannot count document content locally"
     trigger: Final = "tools" if tools else "no_tools"
     matched_rule: Final = next(
         (

litellm/proxy/proxy_server.py

@@ -150,9 +150,7 @@ from litellm.proxy.common_utils.token_count_settings import (
     TokenCountSettings,
     cost_map_entry,
     resolve_token_count_settings,
-)
-from litellm.proxy.common_utils.token_count_settings import (
-    unreliable_local_estimate as _unreliable_local_estimate,
+    unreliable_local_estimate,
 )
 from litellm.proxy.management_helpers.auto_router_availability import AutoRouterCatalogEntry, build_auto_router_catalog
 from litellm.router_utils.access_windows import access_windows_config_error
@@ -14114,7 +14112,7 @@ async def count_request_tokens(
         return ladder
 
     if call_endpoint is True:
-        unreliable_reason: Final = _unreliable_local_estimate(
+        unreliable_reason: Final = unreliable_local_estimate(
             custom_llm_provider=custom_llm_provider,
             model=model_to_use,
             messages=messages,

tests/unit/proxy/test_proxy_token_counter.py

@@ -2754,6 +2754,38 @@ async def test_count_request_tokens_rejects_only_known_unreliable_estimates(
             assert response.count_source == "local_estimate"
 
 
+@pytest.mark.asyncio
+async def test_azure_ai_estimate_guard_runs_without_a_provider_counter(monkeypatch: pytest.MonkeyPatch) -> None:
+    router: Final = Router(
+        model_list=[
+            {
+                "model_name": "azure-ai-count-target",
+                "litellm_params": {
+                    "model": "azure_ai/deepseek-v4-pro",
+                    "api_key": "test-key",
+                    "api_base": "https://count.test",
+                },
+            }
+        ]
+    )
+    monkeypatch.setattr(proxy_server, "llm_router", router)
+
+    with pytest.raises(ProxyException) as raised:
+        await count_request_tokens(
+            request=TokenCountRequest(
+                model="azure-ai-count-target",
+                messages=[{"role": "user", "content": "Summarize."}],
+                tools=[{"name": "search"}],
+            ),
+            call_endpoint=True,
+        )
+
+    assert raised.value.code == "400"
+    assert raised.value.type == "invalid_request_error"
+    assert raised.value.param == "model"
+    assert "deepseek-v4-pro" in raised.value.message
+
+
 @pytest.mark.asyncio
 async def test_document_block_keeps_successful_provider_count(monkeypatch: pytest.MonkeyPatch) -> None:
     router: Final = Router(