Update src/qa.py

src/qa.py

@@ -5,14 +5,14 @@ from vectorstore import search_faiss
 
 print("✅ qa.py loaded from:", __file__)
 
-# Always redirect Hugging Face caches to /tmp
+# Force Hugging Face to use /tmp for cache
 CACHE_DIR = "/tmp/huggingface"
 os.environ["HF_HOME"] = CACHE_DIR
 os.environ["TRANSFORMERS_CACHE"] = CACHE_DIR
 os.environ["HF_DATASETS_CACHE"] = CACHE_DIR
 
 # ----------------------------
-# Embedding model for queries
+# Query embedding model
 # ----------------------------
 _query_model = SentenceTransformer(
     "sentence-transformers/all-MiniLM-L6-v2",
@@ -23,25 +23,13 @@ _query_model = SentenceTransformer(
 # LLM for answers
 # ----------------------------
 MODEL_NAME = "google/flan-t5-small"
-MODEL_PATH = os.path.join(CACHE_DIR, MODEL_NAME)
-
-if not os.path.exists(MODEL_PATH):
-    print(f"⬇️ Downloading {MODEL_NAME} to {MODEL_PATH}")
-    _answer_model = pipeline(
-        "text2text-generation",
-        model=MODEL_NAME,
-        cache_dir=CACHE_DIR
-    )
-    # Save pipeline model locally
-    _answer_model.model.save_pretrained(MODEL_PATH)
-    _answer_model.tokenizer.save_pretrained(MODEL_PATH)
-else:
-    print(f"✅ Loading {MODEL_NAME} from {MODEL_PATH}")
-    _answer_model = pipeline(
-        "text2text-generation",
-        model=MODEL_PATH,
-        cache_dir=CACHE_DIR
-    )
+
+# Make sure model downloads into /tmp
+_answer_model = pipeline(
+    "text2text-generation",
+    model=MODEL_NAME,
+    cache_dir=CACHE_DIR
+)
 
 # ----------------------------
 # Functions