chore: sync app/ and src/ from GitHub

app/app.py

@@ -55,15 +55,17 @@ def load_vector_store_cached():
     login(token=HF_TOKEN, add_to_git_credential=False)
     VECTOR_STORE_DIR.mkdir(parents=True, exist_ok=True)
 
-    snapshot_path = snapshot_download(
-        repo_id="rishadaz/amazon_retriever-storage",
-        repo_type="dataset",
-        local_dir=str(VECTOR_STORE_DIR),
-        split='full',
-        token=HF_TOKEN,
-    )
+    if not any(VECTOR_STORE_DIR.iterdir()):
+        snapshot_path = Path(snapshot_download(
+            repo_id="rishadaz/amazon_retriever-storage",
+            repo_type="dataset",
+            local_dir=str(VECTOR_STORE_DIR),
+            token=HF_TOKEN,
+        ))
+    else:
+        snapshot_path = VECTOR_STORE_DIR
 
-    mini_index_path = Path(snapshot_path) / "tokenisation" / "bm25_index_mini.pkl"
+    mini_index_path = Path(snapshot_path) / "tokenisation" / "bm25_index.pkl"
     embeddings_dir  = Path(snapshot_path) / "embeddings"
 
     vector_store    = load_vector_store(embeddings_dir)

src/semantic.py

@@ -17,6 +17,7 @@ Typical usage
 
 import logging
 from typing import Any
+import streamlit as st
 
 import torch
 import json, os, sys
@@ -45,17 +46,19 @@ DEFAULT_TOP_REVIEWS = 5
 DEFAULT_TOP_K = 5
 
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
-EMBEDDINGS = HuggingFaceEmbeddings(
-    model_name=DEFAULT_EMBEDDING_MODEL,
-    model_kwargs={
-        "device": DEVICE,
-        "model_kwargs": {"torch_dtype": torch.float16},
-    },
-    encode_kwargs={
-        "batch_size": 128 if DEVICE == 'cpu' else 512,
-        "normalize_embeddings": True,
-    },
-)
+@st.cache_resource(show_spinner=False)
+def get_embeddings():
+    return HuggingFaceEmbeddings(
+        model_name=DEFAULT_EMBEDDING_MODEL,
+        model_kwargs={
+            "device": DEVICE,
+            "model_kwargs": {"torch_dtype": torch.float16},
+        },
+        encode_kwargs={
+            "batch_size": 128 if DEVICE == "cpu" else 512,
+            "normalize_embeddings": True,
+        },
+    )
 
 # ---------------------------------------------------------------------------
 # Document construction
@@ -178,11 +181,11 @@ def build_vector_store(
 
     # --- Create new store if needed ---
     if existing_store is None:
-        dim = len(EMBEDDINGS.embed_query("probe"))
+        dim = len(get_embeddings().embed_query("probe"))
         index = faiss.IndexFlatL2(dim)
 
         vector_store = FAISS(
-            embedding_function=EMBEDDINGS,
+            embedding_function=get_embeddings(),
             index=index,
             docstore=InMemoryDocstore(),
             index_to_docstore_id={},
@@ -208,7 +211,7 @@ def build_and_save_vector_store(
     # --- Resume / initialize ---
     if os.path.exists(os.path.join(save_path, "index.faiss")):
         vector_store = FAISS.load_local(
-            save_path, EMBEDDINGS, allow_dangerous_deserialization=True
+            save_path, get_embeddings(), allow_dangerous_deserialization=True
         )
         already_indexed = set(vector_store.index_to_docstore_id.values())
         print(f"Resuming — {len(already_indexed)} docs already indexed.")
@@ -297,6 +300,6 @@ def load_vector_store(
 
     return FAISS.load_local(
         load_path,
-        embeddings=EMBEDDINGS,
+        embeddings=get_embeddings(),
         allow_dangerous_deserialization=True,
     )