Merge pull request #51 from whyhow-ai/40-feature-seperate-embeddings-and-llm-inference

WIP: Updating references and separating concerns.

Author: tomsmoker

backend/CHANGELOG.md

@@ -12,6 +12,10 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added support for queries without source data in vector database
 - Graceful failure of triple export when no chunks are found
 
+### Changed
+
+- Separated embedding service from LLM service
+
 ## [v0.1.5] - 2024-10-29
 
 ### Changed

backend/docker-compose.yml

@@ -14,6 +14,8 @@ services:
     environment:
       - ENVIRONMENT=dev
       - TESTING=0
+    env_file:
+      - .env
     # Add healthcheck for better orchestration
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:8000/health"]

backend/pyproject.toml

@@ -65,8 +65,6 @@ dependencies = [
     "pydantic-settings>=2.5.2",
     "pydantic_core>=2.23.3",
     "pymilvus>=2.4.6",
-    "PyMuPDF>=1.24.10",
-    "PyMuPDFb>=1.24.10",
     "pypdf>=5.0.0",
     "PyPDF2>=3.0.1",
     "python-dateutil>=2.9.0",

backend/src/app/api/v1/endpoints/graph.py

@@ -18,7 +18,7 @@
     ExportTriplesResponseSchema,
 )
 from app.services.graph_service import generate_triples
-from app.services.llm.base import LLMService
+from app.services.llm.base import CompletionService
 from app.services.llm_service import generate_schema
 
 router = APIRouter(tags=["Graph"])
@@ -28,7 +28,7 @@
 @router.post("/export-triples", response_model=ExportTriplesResponseSchema)
 async def export_triples(
     request: ExportTriplesRequestSchema,
-    llm_service: LLMService = Depends(get_llm_service),
+    llm_service: CompletionService = Depends(get_llm_service),
 ) -> ExportTriplesResponseSchema:
     """
     Generate and export triples from a table.
@@ -40,7 +40,7 @@ async def export_triples(
     ----------
     request : ExportTriplesRequestSchema
         The request body containing the table data (columns, rows, and cells).
-    llm_service : LLMService
+    llm_service : CompletionService
         The language model service used for generating the schema, injected by FastAPI.
 
     Returns

backend/src/app/api/v1/endpoints/query.py

@@ -12,7 +12,7 @@
     QueryRequestSchema,
     QueryResult,
 )
-from app.services.llm.base import LLMService
+from app.services.llm.base import CompletionService
 from app.services.query_service import (
     decomposition_query,
     hybrid_query,
@@ -31,7 +31,7 @@
 @router.post("", response_model=QueryAnswerResponse)
 async def run_query(
     request: QueryRequestSchema,
-    llm_service: LLMService = Depends(get_llm_service),
+    llm_service: CompletionService = Depends(get_llm_service),
     vector_db_service: VectorDBService = Depends(get_vector_db_service),
 ) -> QueryAnswerResponse:
     """
@@ -45,7 +45,7 @@ async def run_query(
     ----------
     request : QueryRequestSchema
         The incoming query request.
-    llm_service : LLMService
+    llm_service : CompletionService
         The language model service.
     vector_db_service : VectorDBService
         The vector database service.

backend/src/app/core/dependencies.py

@@ -4,29 +4,46 @@
 
 from app.core.config import Settings, get_settings
 from app.services.document_service import DocumentService
-from app.services.llm.base import LLMService
-from app.services.llm.factory import LLMFactory
+from app.services.embedding.base import EmbeddingService
+from app.services.embedding.factory import EmbeddingServiceFactory
+from app.services.llm.base import CompletionService
+from app.services.llm.factory import CompletionServiceFactory
 from app.services.vector_db.base import VectorDBService
 from app.services.vector_db.factory import VectorDBFactory
 
 
-def get_llm_service(settings: Settings = Depends(get_settings)) -> LLMService:
+def get_llm_service(
+    settings: Settings = Depends(get_settings),
+) -> CompletionService:
     """Get the LLM service for the application."""
-    llm_service = LLMFactory.create_llm_service(settings)
+    llm_service = CompletionServiceFactory.create_service(settings)
     if llm_service is None:
         raise ValueError(
             f"Failed to create LLM service for provider: {settings.llm_provider}"
         )
     return llm_service
 
 
+def get_embedding_service(
+    settings: Settings = Depends(get_settings),
+) -> EmbeddingService:
+    """Get the embedding service for the application."""
+    embedding_service = EmbeddingServiceFactory.create_service(settings)
+    if embedding_service is None:
+        raise ValueError(
+            f"Failed to create embedding service for provider: {settings.embedding_provider}"
+        )
+    return embedding_service
+
+
 def get_vector_db_service(
     settings: Settings = Depends(get_settings),
-    llm_service: LLMService = Depends(get_llm_service),
+    embedding_service: EmbeddingService = Depends(get_embedding_service),
+    llm_service: CompletionService = Depends(get_llm_service),
 ) -> VectorDBService:
     """Get the vector database service for the application."""
     vector_db_service = VectorDBFactory.create_vector_db_service(
-        llm_service, settings
+        embedding_service, llm_service, settings
     )
     if vector_db_service is None:
         raise ValueError(
@@ -38,7 +55,7 @@ def get_vector_db_service(
 def get_document_service(
     settings: Settings = Depends(get_settings),
     vector_db_service: VectorDBService = Depends(get_vector_db_service),
-    llm_service: LLMService = Depends(get_llm_service),
+    llm_service: CompletionService = Depends(get_llm_service),
 ) -> DocumentService:
     """Get the document service for the application."""
     return DocumentService(vector_db_service, llm_service, settings)

backend/src/app/services/document_service.py

@@ -10,7 +10,7 @@
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 
 from app.core.config import Settings
-from app.services.llm.base import LLMService
+from app.services.llm.base import CompletionService
 from app.services.loaders.factory import LoaderFactory
 from app.services.vector_db.base import VectorDBService
 
@@ -23,7 +23,7 @@ class DocumentService:
     def __init__(
         self,
         vector_db_service: VectorDBService,
-        llm_service: LLMService,
+        llm_service: CompletionService,
         settings: Settings,
     ):
         """Document service."""

backend/src/app/services/embedding/__init__.py

@@ -0,0 +1,6 @@
+"""Embedding service module."""
+
+from app.services.embedding.base import EmbeddingService
+from app.services.embedding.factory import EmbeddingServiceFactory
+
+__all__ = ["EmbeddingService", "EmbeddingServiceFactory"]

backend/src/app/services/embedding/base.py

@@ -0,0 +1,13 @@
+"""Abstract base class for embedding services."""
+
+from abc import ABC, abstractmethod
+from typing import List
+
+
+class EmbeddingService(ABC):
+    """Abstract base class for embedding services."""
+
+    @abstractmethod
+    async def get_embeddings(self, texts: List[str]) -> List[List[float]]:
+        """Get the embeddings for the given text."""
+        pass

backend/src/app/services/embedding/factory.py

@@ -0,0 +1,27 @@
+"""Factory for creating embedding services."""
+
+import logging
+from typing import Optional
+
+from app.core.config import Settings
+from app.services.embedding.base import EmbeddingService
+from app.services.embedding.openai_embedding_service import (
+    OpenAIEmbeddingService,
+)
+
+logger = logging.getLogger(__name__)
+
+
+class EmbeddingServiceFactory:
+    """Factory for creating embedding services."""
+
+    @staticmethod
+    def create_service(settings: Settings) -> Optional[EmbeddingService]:
+        """Create an embedding service."""
+        logger.info(
+            f"Creating embedding service for provider: {settings.embedding_provider}"
+        )
+        if settings.embedding_provider == "openai":
+            return OpenAIEmbeddingService(settings)
+        # Add more providers here when needed
+        return None