load local ONNX MiniLM with absolute path (no cache)

rag.py

@@ -109,9 +109,17 @@ def get_texts() -> List[str]:
 def get_vectorstore() -> FAISS:
     texts = get_texts()
 
-    # --- FINAL: use optimum ONNX MiniLM (already on disk in image) --------
-    from langchain.embeddings import OptimumEmbeddings
-    embeddings = OptimumEmbeddings.from_pretrained("optimum/all-MiniLM-L6-v2")
+    # --- FINAL: load local ONNX MiniLM (no internet, no disk write) -------
+    import os
+    local_model_path = os.path.abspath(
+        os.path.join(os.path.dirname(__file__), "st_model")
+    )
+
+    from sentence_transformers import SentenceTransformer
+    model = SentenceTransformer(local_model_path, device="cpu", cache_folder=None)
+
+    from langchain.embeddings import SentenceTransformerEmbeddings
+    embeddings = SentenceTransformerEmbeddings(model=model)
     # ------------------------------------------------------------------------
 
     if not texts:                       # no data → empty FAISS