From a48ac6a20973d0d433e49ff5176e7d0704d84d14 Mon Sep 17 00:00:00 2001
From: Jun Siang Cheah <git@jscheah.me>
Date: Mon, 1 Jul 2024 08:13:56 +0800
Subject: [PATCH] refac: lazily load sentence_transformers to reduce start up
 memory usage

---
 backend/apps/rag/main.py  | 6 ++++--
 backend/apps/rag/utils.py | 4 ++--
 2 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/backend/apps/rag/main.py b/backend/apps/rag/main.py
index 7c6974535..5e4ec03c3 100644
--- a/backend/apps/rag/main.py
+++ b/backend/apps/rag/main.py
@@ -48,8 +48,6 @@ import mimetypes
 import uuid
 import json
 
-import sentence_transformers
-
 from apps.webui.models.documents import (
     Documents,
     DocumentForm,
@@ -190,6 +188,8 @@ def update_embedding_model(
     update_model: bool = False,
 ):
     if embedding_model and app.state.config.RAG_EMBEDDING_ENGINE == "":
+        import sentence_transformers
+
         app.state.sentence_transformer_ef = sentence_transformers.SentenceTransformer(
             get_model_path(embedding_model, update_model),
             device=DEVICE_TYPE,
@@ -204,6 +204,8 @@ def update_reranking_model(
     update_model: bool = False,
 ):
     if reranking_model:
+        import sentence_transformers
+
         app.state.sentence_transformer_rf = sentence_transformers.CrossEncoder(
             get_model_path(reranking_model, update_model),
             device=DEVICE_TYPE,
diff --git a/backend/apps/rag/utils.py b/backend/apps/rag/utils.py
index 7b4324d9a..3a3dad4a2 100644
--- a/backend/apps/rag/utils.py
+++ b/backend/apps/rag/utils.py
@@ -442,8 +442,6 @@ from langchain_core.documents import BaseDocumentCompressor, Document
 from langchain_core.callbacks import Callbacks
 from langchain_core.pydantic_v1 import Extra
 
-from sentence_transformers import util
-
 
 class RerankCompressor(BaseDocumentCompressor):
     embedding_function: Any
@@ -468,6 +466,8 @@ class RerankCompressor(BaseDocumentCompressor):
                 [(query, doc.page_content) for doc in documents]
             )
         else:
+            from sentence_transformers import util
+
             query_embedding = self.embedding_function(query)
             document_embedding = self.embedding_function(
                 [doc.page_content for doc in documents]