fix(caching): stamp provider on embedding cache-hit spend logs

litellm/caching/caching_handler.py

@@ -517,6 +517,7 @@ def _process_async_embedding_cached_response(
                 cached_result=final_embedding_cached_response,
                 is_async=True,
                 is_embedding=True,
+                custom_llm_provider=custom_llm_provider,
             )
             self._async_log_cache_hit_on_callbacks(
                 logging_obj=logging_obj,

tests/test_litellm/caching/test_caching_handler.py

@@ -558,6 +558,44 @@ async def test_embedding_cache_falls_back_to_token_counter_for_legacy_entries():
     assert response.usage.prompt_tokens > 0
 
 
+@pytest.mark.asyncio
+async def test_embedding_cache_hit_sets_custom_llm_provider_on_logging_obj():
+    """A full embedding cache hit must stamp the resolved provider onto the logging
+    obj so spend logs record the provider instead of None/unknown."""
+    from litellm.types.utils import CallTypes
+
+    llm_caching_handler = LLMCachingHandler(
+        original_function=MagicMock(),
+        request_kwargs={},
+        start_time=datetime.now(),
+    )
+
+    cached_result = [
+        {
+            "embedding": [-0.025, -0.019],
+            "index": 0,
+            "object": "embedding",
+            "model": "text-embedding-3-small",
+            "prompt_tokens": 5,
+        }
+    ]
+
+    logging_obj = _build_logging_obj(CallTypes.aembedding.value, stream=False)
+    logging_obj.async_success_handler = AsyncMock()
+
+    response, cache_hit = llm_caching_handler._process_async_embedding_cached_response(
+        final_embedding_cached_response=None,
+        cached_result=cached_result,
+        kwargs={"model": "text-embedding-3-small", "input": "hello world"},
+        logging_obj=logging_obj,
+        start_time=datetime.now(),
+        model="text-embedding-3-small",
+    )
+
+    assert cache_hit
+    assert logging_obj.model_call_details["custom_llm_provider"] == "openai"
+
+
 def test_request_kwargs_does_not_retain_logging_obj():
     """
     The caching handler lives on logging_obj._llm_caching_handler, so keeping