add lighten control (#2567)

### What problem does this PR solve?

#2295

### Type of change

- [ ] Bug Fix (non-breaking change which fixes an issue)
- [x] New Feature (non-breaking change which adds functionality)
- [ ] Documentation Update
- [ ] Refactoring
- [ ] Performance Improvement
- [ ] Other (please describe):

rag/llm/embedding_model.py

@@ -25,10 +25,10 @@ from abc import ABC
 from ollama import Client
 import dashscope
 from openai import OpenAI
-from FlagEmbedding import FlagModel
-import torch
 import numpy as np
 import asyncio
+
+from api.settings import LIGHTEN
 from api.utils.file_utils import get_home_cache_dir
 from rag.utils import num_tokens_from_string, truncate
 import google.generativeai as genai 
@@ -60,8 +60,10 @@ class DefaultEmbedding(Base):
         ^_-
 
         """
-        if not DefaultEmbedding._model:
+        if not LIGHTEN and not DefaultEmbedding._model:
             with DefaultEmbedding._model_lock:
+                from FlagEmbedding import FlagModel
+                import torch
                 if not DefaultEmbedding._model:
                     try:
                         DefaultEmbedding._model = FlagModel(os.path.join(get_home_cache_dir(), re.sub(r"^[a-zA-Z]+/", "", model_name)),

rag/llm/rerank_model.py

@@ -14,14 +14,14 @@
 #  limitations under the License.
 #
 import re
-import  threading
+import threading
 import requests
-import torch
-from FlagEmbedding import FlagReranker
 from huggingface_hub import snapshot_download
 import os
 from abc import ABC
 import numpy as np
+
+from api.settings import LIGHTEN
 from api.utils.file_utils import get_home_cache_dir
 from rag.utils import num_tokens_from_string, truncate
 import json
@@ -53,7 +53,9 @@ class DefaultRerank(Base):
         ^_-
 
         """
-        if not DefaultRerank._model:
+        if not LIGHTEN and not DefaultRerank._model:
+            import torch
+            from FlagEmbedding import FlagReranker
             with DefaultRerank._model_lock:
                 if not DefaultRerank._model:
                     try: