Update model_utils.py

model_utils.py

@@ -2,6 +2,7 @@
 from typing import List, Optional
 import re
 
+import os
 import numpy as np
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
@@ -23,6 +24,9 @@ from loader import (
 MODEL_NAME = "SeaLLMs/SeaLLMs-v3-1.5B-Chat"
 EMBED_MODEL_NAME = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
 
+BASE_DIR = os.path.dirname(os.path.abspath(__file__))
+CACHE_FILE = os.path.join(BASE_DIR, "data", "cached_embeddings.pt")
+
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
@@ -44,15 +48,30 @@ MAX_CONTEXT_ENTRIES = 4
 # -----------------------------
 def _build_entry_embeddings() -> None:
     """
-    Build embeddings for each textbook entry using chapter + section + text
-    and store them in qa_store.TEXT_EMBEDDINGS.
-
-    Call this after loading / reloading curriculum.
+    Load pre-computed embeddings if available, otherwise build them.
     """
     if not getattr(qa_store, "ENTRIES", None):
         qa_store.TEXT_EMBEDDINGS = None
         return
 
+    # 1. Try Loading from Cache
+    if os.path.exists(CACHE_FILE):
+        try:
+            print(f"[INFO] Loading cached embeddings from {CACHE_FILE}...")
+            cache = torch.load(CACHE_FILE, map_location=device)
+            if "textbook" in cache and cache["textbook"] is not None:
+                # Validate size matches
+                if len(cache["textbook"]) == len(qa_store.ENTRIES):
+                    qa_store.TEXT_EMBEDDINGS = cache["textbook"].to(device)
+                    print("[INFO] Textbook embeddings loaded successfully.")
+                    return
+                else:
+                    print("[WARN] Cache size mismatch (Data changed?). Re-computing...")
+        except Exception as e:
+            print(f"[WARN] Failed to load cache: {e}")
+
+    # 2. Fallback: Compute from scratch (The old slow way)
+    print("[INFO] Computing textbook embeddings from scratch...")
     texts: List[str] = []
     for e in qa_store.ENTRIES:
         chapter = e.get("chapter_title", "") or e.get("chapter", "") or ""