Merge pull request #70 from Azure-Samples/embedcolumn

Add embedding with ollama

Author: pamelafox

.env.sample

@@ -18,13 +18,17 @@ AZURE_OPENAI_CHAT_MODEL=gpt-35-turbo
 AZURE_OPENAI_EMBED_DEPLOYMENT=text-embedding-ada-002
 AZURE_OPENAI_EMBED_MODEL=text-embedding-ada-002
 AZURE_OPENAI_EMBED_MODEL_DIMENSIONS=1536
+AZURE_OPENAI_EMBEDDING_COLUMN=embedding_ada002
 # Only needed when using key-based Azure authentication:
 AZURE_OPENAI_KEY=
 # Needed for OpenAI.com:
 OPENAICOM_KEY=YOUR-OPENAI-API-KEY
 OPENAICOM_CHAT_MODEL=gpt-3.5-turbo
 OPENAICOM_EMBED_MODEL=text-embedding-ada-002
 OPENAICOM_EMBED_MODEL_DIMENSIONS=1536
+OPENAICOM_EMBEDDING_COLUMN=embedding_ada002
 # Needed for Ollama:
 OLLAMA_ENDPOINT=http://host.docker.internal:11434/v1
-OLLAMA_CHAT_MODEL=phi3:3.8b
+OLLAMA_CHAT_MODEL=llama3.1
+OLLAMA_EMBED_MODEL=nomic-embed-text
+OLLAMA_EMBEDDING_COLUMN=embedding_nomic

.vscode/settings.json

@@ -22,5 +22,10 @@
                 "ssl": true
             }
         }
-    ]
+    ],
+    "python.testing.pytestArgs": [
+        "tests"
+    ],
+    "python.testing.unittestEnabled": false,
+    "python.testing.pytestEnabled": true
 }

src/backend/fastapi_app/dependencies.py

@@ -28,9 +28,10 @@ class FastAPIAppContext(BaseModel):
 
     openai_chat_model: str
     openai_embed_model: str
-    openai_embed_dimensions: int
+    openai_embed_dimensions: int | None
     openai_chat_deployment: str | None
     openai_embed_deployment: str | None
+    embedding_column: str
 
 
 async def common_parameters():
@@ -43,16 +44,24 @@ async def common_parameters():
         openai_embed_deployment = os.getenv("AZURE_OPENAI_EMBED_DEPLOYMENT", "text-embedding-ada-002")
         openai_embed_model = os.getenv("AZURE_OPENAI_EMBED_MODEL", "text-embedding-ada-002")
         openai_embed_dimensions = int(os.getenv("AZURE_OPENAI_EMBED_DIMENSIONS", 1536))
+        embedding_column = os.getenv("AZURE_OPENAI_EMBEDDING_COLUMN", "embedding_ada002")
+    elif OPENAI_EMBED_HOST == "ollama":
+        openai_embed_deployment = None
+        openai_embed_model = os.getenv("OLLAMA_EMBED_MODEL", "nomic-embed-text")
+        openai_embed_dimensions = None
+        embedding_column = os.getenv("OLLAMA_EMBEDDING_COLUMN", "embedding_nomic")
     else:
-        openai_embed_deployment = "text-embedding-ada-002"
+        openai_embed_deployment = None
         openai_embed_model = os.getenv("OPENAICOM_EMBED_MODEL", "text-embedding-ada-002")
         openai_embed_dimensions = int(os.getenv("OPENAICOM_EMBED_DIMENSIONS", 1536))
+        embedding_column = os.getenv("OPENAICOM_EMBEDDING_COLUMN", "embedding_ada002")
     if OPENAI_CHAT_HOST == "azure":
         openai_chat_deployment = os.getenv("AZURE_OPENAI_CHAT_DEPLOYMENT", "gpt-35-turbo")
         openai_chat_model = os.getenv("AZURE_OPENAI_CHAT_MODEL", "gpt-35-turbo")
     elif OPENAI_CHAT_HOST == "ollama":
         openai_chat_deployment = None
         openai_chat_model = os.getenv("OLLAMA_CHAT_MODEL", "phi3:3.8b")
+        openai_embed_model = os.getenv("OLLAMA_EMBED_MODEL", "nomic-embed-text")
     else:
         openai_chat_deployment = None
         openai_chat_model = os.getenv("OPENAICOM_CHAT_MODEL", "gpt-3.5-turbo")
@@ -62,6 +71,7 @@ async def common_parameters():
         openai_embed_dimensions=openai_embed_dimensions,
         openai_chat_deployment=openai_chat_deployment,
         openai_embed_deployment=openai_embed_deployment,
+        embedding_column=embedding_column,
     )
 
 

src/backend/fastapi_app/embeddings.py

@@ -10,7 +10,7 @@ async def compute_text_embedding(
     openai_client: AsyncOpenAI | AsyncAzureOpenAI,
     embed_model: str,
     embed_deployment: str | None = None,
-    embedding_dimensions: int = 1536,
+    embedding_dimensions: int | None = None,
 ) -> list[float]:
     SUPPORTED_DIMENSIONS_MODEL = {
         "text-embedding-ada-002": False,
@@ -21,7 +21,12 @@ async def compute_text_embedding(
     class ExtraArgs(TypedDict, total=False):
         dimensions: int
 
-    dimensions_args: ExtraArgs = {"dimensions": embedding_dimensions} if SUPPORTED_DIMENSIONS_MODEL[embed_model] else {}
+    dimensions_args: ExtraArgs = {}
+    if SUPPORTED_DIMENSIONS_MODEL.get(embed_model):
+        if embedding_dimensions is None:
+            raise ValueError(f"Model {embed_model} requires embedding dimensions")
+        else:
+            dimensions_args = {"dimensions": embedding_dimensions}
 
     embedding = await openai_client.embeddings.create(
         # Azure OpenAI takes the deployment name as the model name

src/backend/fastapi_app/openai_clients.py

@@ -76,7 +76,13 @@ async def create_openai_embed_client(
                 azure_deployment=azure_deployment,
                 azure_ad_token_provider=token_provider,
             )
-
+    elif OPENAI_EMBED_HOST == "ollama":
+        logger.info("Authenticating to OpenAI using Ollama...")
+        openai_embed_client = openai.AsyncOpenAI(
+            base_url=os.getenv("OLLAMA_ENDPOINT"),
+            api_key="nokeyneeded",
+        )
     else:
+        logger.info("Authenticating to OpenAI using OpenAI.com API key...")
         openai_embed_client = openai.AsyncOpenAI(api_key=os.getenv("OPENAICOM_KEY"))
     return openai_embed_client

src/backend/fastapi_app/postgres_engine.py

@@ -30,7 +30,7 @@ def get_password_from_azure_credential():
 
     engine = create_async_engine(
         DATABASE_URI,
-        echo=False,
+        echo=True,
     )
 
     @event.listens_for(engine.sync_engine, "do_connect")

src/backend/fastapi_app/postgres_models.py

@@ -20,14 +20,17 @@ class Item(Base):
     name: Mapped[str] = mapped_column()
     description: Mapped[str] = mapped_column()
     price: Mapped[float] = mapped_column()
-    embedding: Mapped[Vector] = mapped_column(Vector(1536))  # ada-002
+    embedding_ada002: Mapped[Vector] = mapped_column(Vector(1536))  # ada-002
+    embedding_nomic: Mapped[Vector] = mapped_column(Vector(768))  # nomic-embed-text
 
     def to_dict(self, include_embedding: bool = False):
         model_dict = asdict(self)
         if include_embedding:
-            model_dict["embedding"] = model_dict["embedding"].tolist()
+            model_dict["embedding_ada002"] = model_dict.get("embedding_ada002", [])
+            model_dict["embedding_nomic"] = model_dict.get("embedding_nomic", [])
         else:
-            del model_dict["embedding"]
+            del model_dict["embedding_ada002"]
+            del model_dict["embedding_nomic"]
         return model_dict
 
     def to_str_for_rag(self):
@@ -38,10 +41,18 @@ def to_str_for_embedding(self):
 
 
 # Define HNSW index to support vector similarity search through the vector_cosine_ops access method (cosine distance).
-index = Index(
-    "hnsw_index_for_innerproduct_item_embedding",
-    Item.embedding,
+index_ada002 = Index(
+    "hnsw_index_for_innerproduct_item_embedding_ada002",
+    Item.embedding_ada002,
     postgresql_using="hnsw",
     postgresql_with={"m": 16, "ef_construction": 64},
-    postgresql_ops={"embedding": "vector_ip_ops"},
+    postgresql_ops={"embedding_ada002": "vector_ip_ops"},
+)
+
+index_nomic = Index(
+    "hnsw_index_for_innerproduct_item_embedding_nomic",
+    Item.embedding_nomic,
+    postgresql_using="hnsw",
+    postgresql_with={"m": 16, "ef_construction": 64},
+    postgresql_ops={"embedding_nomic": "vector_ip_ops"},
 )

src/backend/fastapi_app/postgres_searcher.py

@@ -14,13 +14,15 @@ def __init__(
         openai_embed_client: AsyncOpenAI | AsyncAzureOpenAI,
         embed_deployment: str | None,  # Not needed for non-Azure OpenAI or for retrieval_mode="text"
         embed_model: str,
-        embed_dimensions: int,
+        embed_dimensions: int | None,
+        embedding_column: str,
     ):
         self.db_session = db_session
         self.openai_embed_client = openai_embed_client
         self.embed_model = embed_model
         self.embed_deployment = embed_deployment
         self.embed_dimensions = embed_dimensions
+        self.embedding_column = embedding_column
 
     def build_filter_clause(self, filters) -> tuple[str, str]:
         if filters is None:
@@ -36,19 +38,15 @@ def build_filter_clause(self, filters) -> tuple[str, str]:
         return "", ""
 
     async def search(
-        self,
-        query_text: str | None,
-        query_vector: list[float] | list,
-        top: int = 5,
-        filters: list[dict] | None = None,
+        self, query_text: str | None, query_vector: list[float] | list, top: int = 5, filters: list[dict] | None = None
     ):
         filter_clause_where, filter_clause_and = self.build_filter_clause(filters)
 
         vector_query = f"""
-            SELECT id, RANK () OVER (ORDER BY embedding <=> :embedding) AS rank
+            SELECT id, RANK () OVER (ORDER BY {self.embedding_column} <=> :embedding) AS rank
                 FROM items
                 {filter_clause_where}
-                ORDER BY embedding <=> :embedding
+                ORDER BY {self.embedding_column} <=> :embedding
                 LIMIT 20
             """
 

src/backend/fastapi_app/routes/api_routes.py

@@ -45,15 +45,18 @@ async def item_handler(database_session: DBSession, id: int) -> ItemPublic:
 
 
 @router.get("/similar", response_model=list[ItemWithDistance])
-async def similar_handler(database_session: DBSession, id: int, n: int = 5) -> list[ItemWithDistance]:
+async def similar_handler(
+    context: CommonDeps, database_session: DBSession, id: int, n: int = 5
+) -> list[ItemWithDistance]:
     """A similarity API to find items similar to items with given ID."""
     item = (await database_session.scalars(select(Item).where(Item.id == id))).first()
     if not item:
         raise HTTPException(detail=f"Item with ID {id} not found.", status_code=404)
+
     closest = await database_session.execute(
-        select(Item, Item.embedding.l2_distance(item.embedding))
+        select(Item, Item.embedding_ada002.l2_distance(item.embedding_ada002))
         .filter(Item.id != id)
-        .order_by(Item.embedding.l2_distance(item.embedding))
+        .order_by(Item.embedding_ada002.l2_distance(item.embedding_ada002))
         .limit(n)
     )
     return [
@@ -78,6 +81,7 @@ async def search_handler(
         embed_deployment=context.openai_embed_deployment,
         embed_model=context.openai_embed_model,
         embed_dimensions=context.openai_embed_dimensions,
+        embedding_column=context.embedding_column,
     )
     results = await searcher.search_and_embed(
         query, top=top, enable_vector_search=enable_vector_search, enable_text_search=enable_text_search
@@ -99,6 +103,7 @@ async def chat_handler(
         embed_deployment=context.openai_embed_deployment,
         embed_model=context.openai_embed_model,
         embed_dimensions=context.openai_embed_dimensions,
+        embedding_column=context.embedding_column,
     )
     rag_flow: SimpleRAGChat | AdvancedRAGChat
     if chat_request.context.overrides.use_advanced_flow:
@@ -139,6 +144,7 @@ async def chat_stream_handler(
         embed_deployment=context.openai_embed_deployment,
         embed_model=context.openai_embed_model,
         embed_dimensions=context.openai_embed_dimensions,
+        embedding_column=context.embedding_column,
     )
 
     rag_flow: SimpleRAGChat | AdvancedRAGChat

src/backend/fastapi_app/seed_data.json

@@ -36,17 +36,18 @@ async def seed_data(engine):
         with open(os.path.join(current_dir, "seed_data.json")) as f:
             catalog_items = json.load(f)
             for catalog_item in catalog_items:
-                db_item = await session.execute(select(Item).filter(Item.id == catalog_item["Id"]))
+                db_item = await session.execute(select(Item).filter(Item.id == catalog_item["id"]))
                 if db_item.scalars().first():
                     continue
                 item = Item(
-                    id=catalog_item["Id"],
-                    type=catalog_item["Type"],
-                    brand=catalog_item["Brand"],
-                    name=catalog_item["Name"],
-                    description=catalog_item["Description"],
-                    price=catalog_item["Price"],
-                    embedding=catalog_item["Embedding"],
+                    id=catalog_item["id"],
+                    type=catalog_item["type"],
+                    brand=catalog_item["brand"],
+                    name=catalog_item["name"],
+                    description=catalog_item["description"],
+                    price=catalog_item["price"],
+                    embedding_ada002=catalog_item["embedding_ada002"],
+                    embedding_nomic=catalog_item.get("embedding_nomic"),
                 )
                 session.add(item)
             try:

src/backend/fastapi_app/setup_postgres_seeddata.py

@@ -1,4 +1,7 @@
 import asyncio
+import json
+import logging
+import os
 
 from dotenv import load_dotenv
 from sqlalchemy import select
@@ -10,28 +13,75 @@
 from fastapi_app.postgres_engine import create_postgres_engine_from_env
 from fastapi_app.postgres_models import Item
 
+logger = logging.getLogger("ragapp")
 
-async def update_embeddings():
+
+async def update_embeddings(in_seed_data=False):
     azure_credential = await get_azure_credentials()
     engine = await create_postgres_engine_from_env(azure_credential)
     openai_embed_client = await create_openai_embed_client(azure_credential)
     common_params = await common_parameters()
 
-    async with async_sessionmaker(engine, expire_on_commit=False)() as session:
-        async with session.begin():
-            items = (await session.scalars(select(Item))).all()
-
-            for item in items:
-                item.embedding = await compute_text_embedding(
+    embedding_column = ""
+    OPENAI_EMBED_HOST = os.getenv("OPENAI_EMBED_HOST")
+    if OPENAI_EMBED_HOST == "azure":
+        embedding_column = os.getenv("AZURE_OPENAI_EMBEDDING_COLUMN", "embedding_ada002")
+    elif OPENAI_EMBED_HOST == "ollama":
+        embedding_column = os.getenv("OLLAMA_EMBEDDING_COLUMN", "embedding_nomic")
+    else:
+        embedding_column = os.getenv("OPENAICOM_EMBEDDING_COLUMN", "embedding_ada002")
+    logger.info(f"Updating embeddings in column: {embedding_column}")
+    if in_seed_data:
+        current_dir = os.path.dirname(os.path.realpath(__file__))
+        items = []
+        with open(os.path.join(current_dir, "seed_data.json")) as f:
+            catalog_items = json.load(f)
+            for catalog_item in catalog_items:
+                item = Item(
+                    id=catalog_item["id"],
+                    type=catalog_item["type"],
+                    brand=catalog_item["brand"],
+                    name=catalog_item["name"],
+                    description=catalog_item["description"],
+                    price=catalog_item["price"],
+                    embedding_ada002=catalog_item["embedding_ada002"],
+                    embedding_nomic=catalog_item.get("embedding_nomic"),
+                )
+                embedding = await compute_text_embedding(
                     item.to_str_for_embedding(),
                     openai_client=openai_embed_client,
                     embed_model=common_params.openai_embed_model,
+                    embed_deployment=common_params.openai_embed_deployment,
                     embedding_dimensions=common_params.openai_embed_dimensions,
                 )
+                setattr(item, embedding_column, embedding)
+                items.append(item)
+            # write to the file
+            with open(os.path.join(current_dir, "seed_data.json"), "w") as f:
+                json.dump([item.to_dict(include_embedding=True) for item in items], f, indent=4)
+        return
 
+    async with async_sessionmaker(engine, expire_on_commit=False)() as session:
+        async with session.begin():
+            items_to_update = (await session.scalars(select(Item))).all()
+
+            for item in items_to_update:
+                setattr(
+                    item,
+                    embedding_column,
+                    await compute_text_embedding(
+                        item.to_str_for_embedding(),
+                        openai_client=openai_embed_client,
+                        embed_model=common_params.openai_embed_model,
+                        embed_deployment=common_params.openai_embed_deployment,
+                        embedding_dimensions=common_params.openai_embed_dimensions,
+                    ),
+                )
             await session.commit()
 
 
 if __name__ == "__main__":
+    logging.basicConfig(level=logging.WARNING)
+    logger.setLevel(logging.INFO)
     load_dotenv(override=True)
     asyncio.run(update_embeddings())