use optimum ONNX MiniLM (no disk, no internet)

rag.py

@@ -108,22 +108,21 @@ def get_texts() -> List[str]:
 def get_vectorstore() -> FAISS:
     texts = get_texts()
 
-    # --- FINAL: load local MiniLM (no internet, no cache) -----------------
-    import os
-    local_model_path = os.path.abspath(
-        os.path.join(os.path.dirname(__file__), "st_model")
-    )
-
-    # force offline + local only
-    os.environ["TRANSFORMERS_OFFLINE"] = "1"
-    os.environ["HF_DATASETS_OFFLINE"] = "1"
-    os.environ["SENTENCE_TRANSFORMERS_HOME"] = local_model_path
-
-    from sentence_transformers import SentenceTransformer
-    model = SentenceTransformer(local_model_path, device="cpu", cache_folder=None)
-
-    from langchain.embeddings import SentenceTransformerEmbeddings
-    embeddings = SentenceTransformerEmbeddings(model=model)
+    # --- FINAL: optimum ONNX MiniLM (already on disk) ---------------------
+    from optimum.pipelines import pipeline
+    from langchain.embeddings.base import Embeddings
+
+    class OptimumMiniLM(Embeddings):
+        def __init__(self):
+            self.pipe = pipeline("feature-extraction",
+                                 model="optimum/all-MiniLM-L6-v2",
+                                 device="cpu")
+        def embed_documents(self, texts):
+            return [self.pipe(t)[0][0] for t in texts]
+        def embed_query(self, text):
+            return self.embed_documents([text])[0]
+
+    embeddings = OptimumMiniLM()
     # ------------------------------------------------------------------------
 
     if not texts:                       # no data → empty FAISS