Spaces:
Sleeping
Sleeping
Update vector_store.py
Browse files- vector_store.py +8 -13
vector_store.py
CHANGED
|
@@ -1,4 +1,3 @@
|
|
| 1 |
-
# vector_store.py
|
| 2 |
import os
|
| 3 |
import logging
|
| 4 |
import numpy as np
|
|
@@ -9,20 +8,20 @@ from sentence_transformers import SentenceTransformer
|
|
| 9 |
|
| 10 |
logger = logging.getLogger(__name__)
|
| 11 |
|
| 12 |
-
#
|
| 13 |
STORAGE_DIR = "storage"
|
| 14 |
EMB_FILE = os.path.join(STORAGE_DIR, "embeddings_float16.npz")
|
| 15 |
FAISS_FILE = os.path.join(STORAGE_DIR, "faiss_index.idx")
|
| 16 |
|
| 17 |
|
| 18 |
-
#
|
| 19 |
def init_model(model_name: str = "sentence-transformers/all-MiniLM-L6-v2"):
|
| 20 |
"""Initialize SentenceTransformer model (cached by HuggingFace)."""
|
| 21 |
logger.info(f"Loading embedding model: {model_name}")
|
| 22 |
return SentenceTransformer(model_name)
|
| 23 |
|
| 24 |
|
| 25 |
-
#
|
| 26 |
def build_embeddings(documents: List[Dict], model) -> np.ndarray:
|
| 27 |
"""
|
| 28 |
Creates or loads embeddings.
|
|
@@ -52,7 +51,7 @@ def build_embeddings(documents: List[Dict], model) -> np.ndarray:
|
|
| 52 |
return embeddings.astype(np.float32)
|
| 53 |
|
| 54 |
|
| 55 |
-
#
|
| 56 |
def build_faiss_index(embeddings: np.ndarray):
|
| 57 |
"""
|
| 58 |
Create or load a FAISS index.
|
|
@@ -77,7 +76,7 @@ def build_faiss_index(embeddings: np.ndarray):
|
|
| 77 |
return index
|
| 78 |
|
| 79 |
|
| 80 |
-
#
|
| 81 |
def build_bm25(documents: List[Dict]):
|
| 82 |
"""
|
| 83 |
Build BM25 sparse index.
|
|
@@ -90,11 +89,7 @@ def build_bm25(documents: List[Dict]):
|
|
| 90 |
return bm25
|
| 91 |
|
| 92 |
|
| 93 |
-
#
|
| 94 |
-
# 🔍 SEARCH METHODS
|
| 95 |
-
# ============================================================
|
| 96 |
-
|
| 97 |
-
# -------- Semantic Search (via FAISS) --------
|
| 98 |
def semantic_search(query: str, model, faiss_index, documents, k=5):
|
| 99 |
"""
|
| 100 |
Returns top-k documents ranked by dense semantic similarity (FAISS).
|
|
@@ -114,7 +109,7 @@ def semantic_search(query: str, model, faiss_index, documents, k=5):
|
|
| 114 |
return results
|
| 115 |
|
| 116 |
|
| 117 |
-
#
|
| 118 |
def bm25_search(query: str, bm25, documents, k=5):
|
| 119 |
"""
|
| 120 |
Returns top-k documents ranked by sparse lexical BM25 similarity.
|
|
@@ -132,7 +127,7 @@ def bm25_search(query: str, bm25, documents, k=5):
|
|
| 132 |
return results
|
| 133 |
|
| 134 |
|
| 135 |
-
#
|
| 136 |
def hybrid_search(query: str, model, faiss_index, bm25, documents, k=5, alpha=0.5):
|
| 137 |
"""
|
| 138 |
Combines semantic FAISS + lexical BM25 search.
|
|
|
|
|
|
|
| 1 |
import os
|
| 2 |
import logging
|
| 3 |
import numpy as np
|
|
|
|
| 8 |
|
| 9 |
logger = logging.getLogger(__name__)
|
| 10 |
|
| 11 |
+
# Constants
|
| 12 |
STORAGE_DIR = "storage"
|
| 13 |
EMB_FILE = os.path.join(STORAGE_DIR, "embeddings_float16.npz")
|
| 14 |
FAISS_FILE = os.path.join(STORAGE_DIR, "faiss_index.idx")
|
| 15 |
|
| 16 |
|
| 17 |
+
# Model Init
|
| 18 |
def init_model(model_name: str = "sentence-transformers/all-MiniLM-L6-v2"):
|
| 19 |
"""Initialize SentenceTransformer model (cached by HuggingFace)."""
|
| 20 |
logger.info(f"Loading embedding model: {model_name}")
|
| 21 |
return SentenceTransformer(model_name)
|
| 22 |
|
| 23 |
|
| 24 |
+
# Embeddings
|
| 25 |
def build_embeddings(documents: List[Dict], model) -> np.ndarray:
|
| 26 |
"""
|
| 27 |
Creates or loads embeddings.
|
|
|
|
| 51 |
return embeddings.astype(np.float32)
|
| 52 |
|
| 53 |
|
| 54 |
+
# FAISS Index
|
| 55 |
def build_faiss_index(embeddings: np.ndarray):
|
| 56 |
"""
|
| 57 |
Create or load a FAISS index.
|
|
|
|
| 76 |
return index
|
| 77 |
|
| 78 |
|
| 79 |
+
# BM25
|
| 80 |
def build_bm25(documents: List[Dict]):
|
| 81 |
"""
|
| 82 |
Build BM25 sparse index.
|
|
|
|
| 89 |
return bm25
|
| 90 |
|
| 91 |
|
| 92 |
+
# Semantic Search (via FAISS)
|
|
|
|
|
|
|
|
|
|
|
|
|
| 93 |
def semantic_search(query: str, model, faiss_index, documents, k=5):
|
| 94 |
"""
|
| 95 |
Returns top-k documents ranked by dense semantic similarity (FAISS).
|
|
|
|
| 109 |
return results
|
| 110 |
|
| 111 |
|
| 112 |
+
# BM25 Search
|
| 113 |
def bm25_search(query: str, bm25, documents, k=5):
|
| 114 |
"""
|
| 115 |
Returns top-k documents ranked by sparse lexical BM25 similarity.
|
|
|
|
| 127 |
return results
|
| 128 |
|
| 129 |
|
| 130 |
+
# Hybrid Search (FAISS + BM25)
|
| 131 |
def hybrid_search(query: str, model, faiss_index, bm25, documents, k=5, alpha=0.5):
|
| 132 |
"""
|
| 133 |
Combines semantic FAISS + lexical BM25 search.
|