fix: clear error for nested lists on embedContent path, add validation tests

litellm/llms/vertex_ai/gemini_embeddings/batch_embed_content_transformation.py

@@ -269,6 +269,11 @@ def transform_openai_input_gemini_embed_content(
     parts: List[PartType] = []
 
     for element in input_list:
+        if isinstance(element, list):
+            raise ValueError(
+                "Nested (combined) embeddings are not supported on the embedContent path. "
+                "Use the batchEmbedContents path or pass a flat list instead."
+            )
         if not isinstance(element, str):
             raise ValueError(f"Unsupported input type: {type(element)}")
         parts.append(_build_part_for_input(element, resolved_files=resolved_files))

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

@@ -257,3 +257,33 @@ class TestProcessResponse:
         assert result.data[1]["index"] == 1
         # Should count tokens only for the text element, not the image
         assert result.usage.prompt_tokens > 0
+
+    def test_nested_input_token_counting(self):
+        """Nested list: only plain-text sub-elements should be counted."""
+        predictions: VertexAIBatchEmbeddingsResponseObject = {
+            "embeddings": [{"values": [0.1, 0.2]}]
+        }
+        result = process_response(
+            input=[["a red shoe", IMAGE_DATA_URI]],
+            model_response=EmbeddingResponse(),
+            model="gemini-embedding-2-preview",
+            _predictions=predictions,
+        )
+        assert len(result.data) == 1
+        assert result.usage.prompt_tokens > 0
+
+    def test_nested_empty_list_raises(self):
+        with pytest.raises(ValueError, match="must not be empty"):
+            transform_openai_input_gemini_content(
+                input=[[]],
+                model="gemini-embedding-2-preview",
+                optional_params={},
+            )
+
+    def test_nested_non_string_element_raises(self):
+        with pytest.raises(ValueError, match="must be strings"):
+            transform_openai_input_gemini_content(
+                input=[[["doubly", "nested"]]],
+                model="gemini-embedding-2-preview",
+                optional_params={},
+            )