fix: replace sentence-transformers with fastembed to eliminate torch/CUDA dependency

src/config.py

@@ -11,4 +11,4 @@ ADMIN_TELEGRAM_ID = int(os.environ["ADMIN_TELEGRAM_ID"])
 DB_PATH = os.getenv("DB_PATH", "/app/data/second_brain.db")
 CHROMA_PATH = os.getenv("CHROMA_PATH", "/app/data/chromadb")
 UPLOAD_PATH = os.getenv("UPLOAD_PATH", "/app/data/uploads")
-EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL", "all-MiniLM-L6-v2")
+EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL", "sentence-transformers/all-MiniLM-L6-v2")

src/rag.py

@@ -1,10 +1,10 @@
 import asyncio
 import chromadb
-from sentence_transformers import SentenceTransformer
+from fastembed import TextEmbedding
 from config import CHROMA_PATH, EMBEDDING_MODEL
 
 _chroma = chromadb.PersistentClient(path=CHROMA_PATH)
-_embedding_model = SentenceTransformer(EMBEDDING_MODEL)
+_embedding_model = TextEmbedding(model_name=EMBEDDING_MODEL)
 
 
 def _get_collection(name: str):
@@ -12,8 +12,8 @@ def _get_collection(name: str):
 
 
 async def _embed(text: str) -> list:
-    embedding = await asyncio.to_thread(_embedding_model.encode, text)
-    return embedding.tolist()
+    result = await asyncio.to_thread(lambda: list(_embedding_model.embed([text]))[0])
+    return result.tolist()
 
 
 async def add_memory(user_id: int, memory_id: int, content: str, scope: str = "private"):