force local st_model folder (no internet, no cache)

rag.py

@@ -104,15 +104,18 @@ def get_texts() -> List[str]:
         return []
 
 ## ------------------------------------------------------------------
-# ------# ------------------------------------------------------------------
-# ------------------------------------------------------------------
+# ------# --------# ------------------------------------------------------------------
 @lru_cache(maxsize=1)
 def get_vectorstore() -> FAISS:
     texts = get_texts()
 
-    # --- FINAL: load 384 KB model once into RAM (no cache, no disk) --------
+    # --- FINAL: load model from repo folder (no internet, no cache) --------
+    import os
+    local_model_path = os.path.join(os.path.dirname(__file__), "st_model")
+
     from sentence_transformers import SentenceTransformer
-    model = SentenceTransformer("all-MiniLM-L6-v2", device="cpu", cache_folder=None)
+    model = SentenceTransformer(local_model_path, device="cpu", cache_folder=None)
+
     from langchain.embeddings import SentenceTransformerEmbeddings
     embeddings = SentenceTransformerEmbeddings(model=model)
     # ------------------------------------------------------------------------
@@ -123,7 +126,6 @@ def get_vectorstore() -> FAISS:
     splitter = RecursiveCharacterTextSplitter(chunk_size=600, chunk_overlap=50)
     docs = splitter.create_documents(texts, metadatas=[{"source": DATASET}] * len(texts))
     return FAISS.from_documents(docs, embeddings)
-
 # ------------------------------------------------------------------# LLM
 # ------------------------------------------------------------------
 @lru_cache(maxsize=1)