fix(vertex): only bill image rate without modality details when every input is an image

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

litellm/llms/vertex_ai/gemini_embeddings/batch_embed_content_transformation.py

@@ -337,11 +337,12 @@ def _is_image_element(
     return False
 
 
-def _count_input_images(
+def _is_image_only_input(
     input: GeminiEmbeddingInput,
     resolved_files: Mapping[str, Mapping[str, str]],
-) -> int:
-    return sum(1 for element in _flatten_input(input) if _is_image_element(element, resolved_files))
+) -> bool:
+    elements: Final = _flatten_input(input)
+    return bool(elements) and all(_is_image_element(element, resolved_files) for element in elements)
 
 
 def _tokens_for_modality(details: Sequence[PromptTokensDetails], modality: str) -> int:
@@ -376,7 +377,7 @@ def _usage_from_embed_content_response(
             total_tokens=total_tokens,
             prompt_tokens_details=PromptTokensDetailsWrapper(
                 text_tokens=0,
-                image_tokens=prompt_tokens if _count_input_images(input, resolved_files) else 0,
+                image_tokens=prompt_tokens if _is_image_only_input(input, resolved_files) else 0,
             ),
         )
 

tests/test_litellm/llms/vertex_ai/gemini_embeddings/test_batch_embed_content_transformation.py

@@ -524,6 +524,29 @@ class TestProcessEmbedContentResponseUsage:
         )
         assert prompt_cost == pytest.approx(258 * 4.5e-7)
 
+    def test_mixed_text_and_image_without_modality_details_not_billed_as_image(self):
+        response_json = {
+            "embedding": {"values": [0.1]},
+            "usageMetadata": {
+                "promptTokenCount": 270,
+                "totalTokenCount": 270,
+            },
+        }
+        result = process_embed_content_response(
+            input=["a short caption", IMAGE_DATA_URI],
+            model_response=EmbeddingResponse(),
+            model=self.MODEL,
+            response_json=response_json,
+        )
+        assert result.usage.prompt_tokens_details.image_tokens == 0
+
+        prompt_cost, _ = generic_cost_per_token(
+            model=self.MODEL,
+            usage=result.usage,
+            custom_llm_provider="vertex_ai",
+        )
+        assert prompt_cost == pytest.approx(270 * 2e-7)
+
     def test_text_without_modality_details_uses_text_rate(self):
         response_json = {
             "embedding": {"values": [0.1]},