Updated app with pre-built assets

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+assets/faiss.index filter=lfs diff=lfs merge=lfs -text

app.py

@@ -76,7 +76,7 @@ def _ensure_index():
 
 def recommend(query_text: str, top_k: int = 3) -> pd.DataFrame:
     _ensure_index()
-    _ensure_models() # Make sure the embedder is ready
+    _ensure_models()
     q_vec = _EMBED_MODEL.encode([query_text], normalize_embeddings=True).astype("float32")
     scores, idxs = _INDEX.search(q_vec, top_k)
     out = _DATA_DF.iloc[idxs[0]].copy()
@@ -168,7 +168,7 @@ def _copies_neighbor(s: str, neighbors_df: pd.DataFrame) -> bool:
         if _jaccard(s_toks, _tokens(t_low)) >= 0.7:
             return True
     try:
-        _ensure_models() # Make sure the embedder is ready
+        _ensure_models()
         s_vec = _EMBED_MODEL.encode([s])[0]; s_vec = s_vec / np.linalg.norm(s_vec)
         for _, row in neighbors_df.head(3).iterrows():
             t = str(row.get("tagline", "")).strip()
@@ -193,14 +193,14 @@ def _clean_slogan(text: str, max_words: int = 8) -> str:
 def _score_candidates(query: str, cands: List[str], neighbors_df: pd.DataFrame) -> List[tuple]:
     if not cands:
         return []
-    _ensure_models() # Make sure the cross-encoder is ready
+    _ensure_models()
     ce_scores = np.asarray(_RERANKER.predict([(query, s) for s in cands]), dtype=np.float32) / 5.0
     q_toks = _tokens(query)
     results = []
 
     neighbor_vecs = []
     if neighbors_df is not None and not neighbors_df.empty:
-        _ensure_models() # Make sure the embedder is ready
+        _ensure_models()
         for _, row in neighbors_df.head(3).iterrows():
             t = str(row.get("tagline","")).strip()
             if t:
@@ -220,7 +220,7 @@ def _score_candidates(query: str, cands: List[str], neighbors_df: pd.DataFrame)
         n_pen = 0.0
         if neighbor_vecs:
             try:
-                _ensure_models() # Make sure the embedder is ready
+                _ensure_models()
                 s_vec = _EMBED_MODEL.encode([s])[0]; s_vec = s_vec / np.linalg.norm(s_vec)
                 sim_max = max(float(np.dot(s_vec, nv)) for nv in neighbor_vecs) if neighbor_vecs else 0.0
                 n_pen = sim_max

assets/data.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77fbc4451a8048cbe3c037b337fa90e6d52fb5dc0d815844803de185f2843ce5
+size 5942482

assets/embeddings.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3f4cfcdb193612010ec8507ef90766eb95be0845e45a5640a46602e4b745da7d
+size 106346624

assets/faiss.index

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4cd2c7bf43c63dd857c96af91a599b19f1e964aebe123eee743dabbaceb3b78
+size 106346541

assets/meta.json

@@ -0,0 +1 @@
+{"row_count": 34618, "dim": 768, "model": "sentence-transformers/all-mpnet-base-v2"}