force cache into writable /code/.cache

rag.py

@@ -117,17 +117,28 @@ def get_texts() -> List[str]:
 @lru_cache(maxsize=1)
 def get_vectorstore() -> FAISS:
     texts = get_texts()
-    if not texts:                       # ← no data → return empty FAISS
+    if not texts:                       # no data → empty FAISS
+        import os
+        os.environ["HF_HOME"] = "/code/.cache"
+        os.environ["TRANSFORMERS_CACHE"] = "/code/.cache"
         embeddings = HuggingFaceEmbeddings(model_name=EMBED_MODEL)
-        return FAISS.from_texts([""], embeddings)  # dummy, retriever will be empty
+        return FAISS.from_texts([""], embeddings)  # dummy
 
     splitter = RecursiveCharacterTextSplitter(chunk_size=600, chunk_overlap=50)
     docs = splitter.create_documents(texts, metadatas=[{"source": DATASET}] * len(texts))
+
+    import os
+    os.environ["HF_HOME"] = "/code/.cache"
+    os.environ["TRANSFORMERS_CACHE"] = "/code/.cache"
     embeddings = HuggingFaceEmbeddings(model_name=EMBED_MODEL)
     return FAISS.from_documents(docs, embeddings)
 
-# ------------------------------------------------------------------
-# LLM
+
+
+
+
+
+# ------------------------------------------------------------------# LLM
 # ------------------------------------------------------------------
 @lru_cache(maxsize=1)
 def get_llm():