Port to embedding-3-large

.env.sample

@@ -15,10 +15,10 @@ AZURE_OPENAI_ENDPOINT=https://YOUR-AZURE-OPENAI-SERVICE-NAME.openai.azure.com
 AZURE_OPENAI_VERSION=2024-03-01-preview
 AZURE_OPENAI_CHAT_DEPLOYMENT=gpt-4o-mini
 AZURE_OPENAI_CHAT_MODEL=gpt-4o-mini
-AZURE_OPENAI_EMBED_DEPLOYMENT=text-embedding-ada-002
-AZURE_OPENAI_EMBED_MODEL=text-embedding-ada-002
-AZURE_OPENAI_EMBED_DIMENSIONS=1536
-AZURE_OPENAI_EMBEDDING_COLUMN=embedding_ada002
+AZURE_OPENAI_EMBED_DEPLOYMENT=text-embedding-3-large
+AZURE_OPENAI_EMBED_MODEL=text-embedding-3-large
+AZURE_OPENAI_EMBED_DIMENSIONS=1024
+AZURE_OPENAI_EMBEDDING_COLUMN=embedding_3l
 AZURE_OPENAI_EVAL_DEPLOYMENT=gpt-4
 AZURE_OPENAI_EVAL_MODEL=gpt-4
 AZURE_TENANT_ID=
@@ -27,9 +27,9 @@ 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
+OPENAICOM_EMBED_MODEL=text-embedding-3-large
+OPENAICOM_EMBED_MODEL_DIMENSIONS=1024
+OPENAICOM_EMBEDDING_COLUMN=embedding_3l
 # Needed for Ollama:
 OLLAMA_ENDPOINT=http://host.docker.internal:11434/v1
 OLLAMA_CHAT_MODEL=llama3.1

README.md

@@ -123,7 +123,7 @@ Once you've opened the project in [Codespaces](#github-codespaces), [Dev Contain
     azd up
     ```
 
-    You will be asked to select two locations, first a region for most of the resources (Container Apps, PostgreSQL), then a region specifically for the Azure OpenAI models. This project uses the gpt-4o-mini and text-embedding-ada-002 models which may not be available in all Azure regions. Check for [up-to-date region availability](https://learn.microsoft.com/azure/ai-services/openai/concepts/models#standard-deployment-model-availability) and select a region accordingly.
+    You will be asked to select two locations, first a region for most of the resources (Container Apps, PostgreSQL), then a region specifically for the Azure OpenAI models. This project uses the gpt-4o-mini and text-embedding-3-large models which may not be available in all Azure regions. Check for [up-to-date region availability](https://learn.microsoft.com/azure/ai-services/openai/concepts/models#standard-deployment-model-availability) and select a region accordingly.
 
 ## Local Development
 

infra/main.bicep

@@ -303,11 +303,11 @@ var webAppEnv = union(azureOpenAIKeyEnv, openAIComKeyEnv, [
   }
   {
     name: 'OPENAICOM_EMBED_MODEL_DIMENSIONS'
-    value: openAIEmbedHost == 'openaicom' ? '1536' : ''
+    value: openAIEmbedHost == 'openaicom' ? '1024' : ''
   }
   {
     name: 'OPENAICOM_EMBED_MODEL'
-    value: openAIEmbedHost == 'openaicom' ? 'text-embedding-ada-002' : ''
+    value: openAIEmbedHost == 'openaicom' ? 'text-embedding-3-large' : ''
   }
   {
     name: 'AZURE_OPENAI_EMBED_MODEL'

infra/main.parameters.json

@@ -69,22 +69,22 @@
         "value": "${OPENAI_EMBED_HOST=azure}"
       },
       "embedModelName":{
-        "value": "${AZURE_OPENAI_EMBED_MODEL=text-embedding-ada-002}"
+        "value": "${AZURE_OPENAI_EMBED_MODEL=text-embedding-3-large}"
       },
       "embedDeploymentName": {
-        "value": "${AZURE_OPENAI_EMBED_DEPLOYMENT=text-embedding-ada-002}"
+        "value": "${AZURE_OPENAI_EMBED_DEPLOYMENT=text-embedding-3-large}"
       },
       "embedDeploymentVersion":{
-        "value": "${AZURE_OPENAI_EMBED_DEPLOYMENT_VERSION=2}"
+        "value": "${AZURE_OPENAI_EMBED_DEPLOYMENT_VERSION=1}"
       },
       "embedDeploymentSku": {
-        "value": "${AZURE_OPENAI_EMBED_DEPLOYMENT_SKU=Standard}"
+        "value": "${AZURE_OPENAI_EMBED_DEPLOYMENT_SKU=GlobalStandard}"
       },
       "embedDeploymentCapacity":{
         "value": "${AZURE_OPENAI_EMBED_DEPLOYMENT_CAPACITY=30}"
       },
       "embedDimensions": {
-        "value": "${AZURE_OPENAI_EMBED_DIMENSIONS=1536}"
+        "value": "${AZURE_OPENAI_EMBED_DIMENSIONS=1024}"
       },
       "openAIComKey": {
         "value": "${OPENAICOM_KEY}"

src/backend/fastapi_app/dependencies.py

@@ -42,20 +42,20 @@ async def common_parameters():
     OPENAI_EMBED_HOST = os.getenv("OPENAI_EMBED_HOST")
     OPENAI_CHAT_HOST = os.getenv("OPENAI_CHAT_HOST")
     if OPENAI_EMBED_HOST == "azure":
-        openai_embed_deployment = os.getenv("AZURE_OPENAI_EMBED_DEPLOYMENT") or "text-embedding-ada-002"
-        openai_embed_model = os.getenv("AZURE_OPENAI_EMBED_MODEL") or "text-embedding-ada-002"
-        openai_embed_dimensions = int(os.getenv("AZURE_OPENAI_EMBED_DIMENSIONS") or 1536)
-        embedding_column = os.getenv("AZURE_OPENAI_EMBEDDING_COLUMN") or "embedding_ada002"
+        openai_embed_deployment = os.getenv("AZURE_OPENAI_EMBED_DEPLOYMENT") or "text-embedding-3-large"
+        openai_embed_model = os.getenv("AZURE_OPENAI_EMBED_MODEL") or "text-embedding-3-large"
+        openai_embed_dimensions = int(os.getenv("AZURE_OPENAI_EMBED_DIMENSIONS") or 1024)
+        embedding_column = os.getenv("AZURE_OPENAI_EMBEDDING_COLUMN") or "embedding_3l"
     elif OPENAI_EMBED_HOST == "ollama":
         openai_embed_deployment = None
         openai_embed_model = os.getenv("OLLAMA_EMBED_MODEL") or "nomic-embed-text"
         openai_embed_dimensions = None
         embedding_column = os.getenv("OLLAMA_EMBEDDING_COLUMN") or "embedding_nomic"
     else:
         openai_embed_deployment = None
-        openai_embed_model = os.getenv("OPENAICOM_EMBED_MODEL") or "text-embedding-ada-002"
-        openai_embed_dimensions = int(os.getenv("OPENAICOM_EMBED_DIMENSIONS", 1536))
-        embedding_column = os.getenv("OPENAICOM_EMBEDDING_COLUMN") or "embedding_ada002"
+        openai_embed_model = os.getenv("OPENAICOM_EMBED_MODEL") or "text-embedding-3-large"
+        openai_embed_dimensions = int(os.getenv("OPENAICOM_EMBED_DIMENSIONS", 1024))
+        embedding_column = os.getenv("OPENAICOM_EMBEDDING_COLUMN") or "embedding_3l"
     if OPENAI_CHAT_HOST == "azure":
         openai_chat_deployment = os.getenv("AZURE_OPENAI_CHAT_DEPLOYMENT") or "gpt-4o-mini"
         openai_chat_model = os.getenv("AZURE_OPENAI_CHAT_MODEL") or "gpt-4o-mini"

src/backend/fastapi_app/postgres_models.py

@@ -19,16 +19,16 @@ class Item(Base):
     description: Mapped[str] = mapped_column()
     price: Mapped[float] = mapped_column()
     # Embeddings for different models:
-    embedding_ada002: Mapped[Vector] = mapped_column(Vector(1536), nullable=True)  # ada-002
+    embedding_3l: Mapped[Vector] = mapped_column(Vector(1024), nullable=True)  # text-embedding-3-large
     embedding_nomic: Mapped[Vector] = mapped_column(Vector(768), nullable=True)  # nomic-embed-text
 
     def to_dict(self, include_embedding: bool = False):
         model_dict = {column.name: getattr(self, column.name) for column in self.__table__.columns}
         if include_embedding:
-            model_dict["embedding_ada002"] = model_dict.get("embedding_ada002", [])
+            model_dict["embedding_3l"] = model_dict.get("embedding_3l", [])
             model_dict["embedding_nomic"] = model_dict.get("embedding_nomic", [])
         else:
-            del model_dict["embedding_ada002"]
+            del model_dict["embedding_3l"]
             del model_dict["embedding_nomic"]
         return model_dict
 
@@ -51,12 +51,12 @@ def to_str_for_embedding(self):
 
 table_name = Item.__tablename__
 
-index_ada002 = Index(
-    f"hnsw_index_for_cosine_{table_name}_embedding_ada002",
-    Item.embedding_ada002,
+index_3l = Index(
+    f"hnsw_index_for_cosine_{table_name}_embedding_3l",
+    Item.embedding_3l,
     postgresql_using="hnsw",
     postgresql_with={"m": 16, "ef_construction": 64},
-    postgresql_ops={"embedding_ada002": "vector_cosine_ops"},
+    postgresql_ops={"embedding_3l": "vector_cosine_ops"},
 )
 
 index_nomic = Index(

src/backend/fastapi_app/routes/api_routes.py

@@ -68,7 +68,7 @@ async def similar_handler(
                 f"SELECT *, {context.embedding_column} <=> :embedding as DISTANCE FROM {Item.__tablename__} "
                 "WHERE id <> :item_id ORDER BY distance LIMIT :n"
             ),
-            {"embedding": item.embedding_ada002, "n": n, "item_id": id},
+            {"embedding": item.embedding_3l, "n": n, "item_id": id},
         )
     ).fetchall()