torch give me my memory

app.py

@@ -3,7 +3,9 @@ app.py  –  Tiny-RAG (Gradio playground)  +  REST API (/ingest, /query)
 """
 
 # ---------- 1. imports & global helpers -------------
-import os, math, torch, uvicorn, gradio as gr
+import os
+os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
+import math, torch, uvicorn, gradio as gr
 from fastapi import FastAPI, HTTPException
 from pydantic import BaseModel
 from transformers import (