fix(s3-vectors): use selected embedding model during ingest

litellm/rag/ingestion/s3_vectors_ingestion.py

@@ -67,6 +67,9 @@ class S3VectorsRAGIngestion(BaseRAGIngestion, BaseAWSLLM):
 
         # Extract config
         self.vector_bucket_name = self.vector_store_config["vector_bucket_name"]
+        embedding_model = self.vector_store_config.get("embedding_model")
+        if not self.embedding_config and embedding_model:
+            self.embedding_config = {"model": embedding_model}
         self.index_name = self.vector_store_config.get("index_name")
         self.distance_metric = self.vector_store_config.get(
             "distance_metric", S3_VECTORS_DEFAULT_DISTANCE_METRIC

tests/test_litellm/proxy/rag_endpoints/test_s3_vectors_ingestion.py

@@ -0,0 +1,54 @@
+from types import SimpleNamespace
+from unittest.mock import AsyncMock, patch
+
+import pytest
+
+from litellm.rag.ingestion.s3_vectors_ingestion import S3VectorsRAGIngestion
+
+
+def _s3_vectors_ingest_options(**overrides):
+    vector_store = {
+        "custom_llm_provider": "s3_vectors",
+        "vector_bucket_name": "test-vector-bucket",
+        "aws_region_name": "us-east-1",
+    }
+    vector_store.update(overrides.pop("vector_store", {}))
+    return {
+        "vector_store": vector_store,
+        **overrides,
+    }
+
+
+@pytest.mark.asyncio
+async def test_s3_vectors_ingestion_uses_vector_store_embedding_model():
+    ingestion = S3VectorsRAGIngestion(
+        ingest_options=_s3_vectors_ingest_options(
+            vector_store={"embedding_model": "text-embedding-3-large"}
+        )
+    )
+
+    with patch(
+        "litellm.rag.ingestion.s3_vectors_ingestion.litellm.aembedding",
+        new_callable=AsyncMock,
+    ) as mock_aembedding:
+        mock_aembedding.return_value = SimpleNamespace(
+            data=[{"embedding": [0.1, 0.2, 0.3]}]
+        )
+
+        embeddings = await ingestion.embed(["hello"])
+
+    mock_aembedding.assert_awaited_once_with(
+        model="text-embedding-3-large", input=["hello"]
+    )
+    assert embeddings == [[0.1, 0.2, 0.3]]
+
+
+def test_s3_vectors_ingestion_prefers_top_level_embedding_config():
+    ingestion = S3VectorsRAGIngestion(
+        ingest_options=_s3_vectors_ingest_options(
+            embedding={"model": "text-embedding-3-small"},
+            vector_store={"embedding_model": "text-embedding-3-large"},
+        )
+    )
+
+    assert ingestion.embedding_config == {"model": "text-embedding-3-small"}