devjas1
commited on
Commit
·
eae02ee
1
Parent(s):
cede51b
(FEAT)[Implement FAISS Indexing]: enhance document embedding process by integrating FAISS indexing and saving metadata.
Browse files- src/embedder.py +52 -13
src/embedder.py
CHANGED
|
@@ -1,32 +1,71 @@
|
|
| 1 |
"""
|
| 2 |
This script handles document embedding using EmbeddingGemma.
|
| 3 |
This is the entry point for indexing documents.
|
| 4 |
-
TODO: Wire this to FAISS
|
| 5 |
"""
|
| 6 |
|
| 7 |
import os
|
|
|
|
|
|
|
|
|
|
| 8 |
from sentence_transformers import SentenceTransformer
|
| 9 |
|
| 10 |
|
| 11 |
def embed_documents(path: str, config: dict):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
try:
|
| 13 |
model = SentenceTransformer(config["embedding"]["model_path"])
|
| 14 |
-
|
| 15 |
-
|
|
|
|
|
|
|
| 16 |
|
| 17 |
-
model = SentenceTransformer(config["embedding"]["model_path"])
|
| 18 |
embeddings = []
|
|
|
|
|
|
|
| 19 |
|
|
|
|
| 20 |
for fname in os.listdir(path):
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 28 |
|
|
|
|
| 29 |
print(f"Total embeddings created: {len(embeddings)}")
|
| 30 |
-
return embeddings
|
| 31 |
|
| 32 |
-
|
|
|
|
| 1 |
"""
|
| 2 |
This script handles document embedding using EmbeddingGemma.
|
| 3 |
This is the entry point for indexing documents.
|
|
|
|
| 4 |
"""
|
| 5 |
|
| 6 |
import os
|
| 7 |
+
import pickle
|
| 8 |
+
import faiss
|
| 9 |
+
import numpy as np
|
| 10 |
from sentence_transformers import SentenceTransformer
|
| 11 |
|
| 12 |
|
| 13 |
def embed_documents(path: str, config: dict):
|
| 14 |
+
"""
|
| 15 |
+
Embed documents from a directory and save to FAISS index.
|
| 16 |
+
|
| 17 |
+
Args:
|
| 18 |
+
path (str): Path to the directory containing the documents to embed.
|
| 19 |
+
config (dict): Configuration dictionary.
|
| 20 |
+
"""
|
| 21 |
try:
|
| 22 |
model = SentenceTransformer(config["embedding"]["model_path"])
|
| 23 |
+
print(f"Initalized embedding model: {config['embedding']['model_path']}")
|
| 24 |
+
except ValueError as e:
|
| 25 |
+
print(f"Error initializing embedding model: {e}")
|
| 26 |
+
return []
|
| 27 |
|
|
|
|
| 28 |
embeddings = []
|
| 29 |
+
texts = []
|
| 30 |
+
filenames = []
|
| 31 |
|
| 32 |
+
# Read all documents
|
| 33 |
for fname in os.listdir(path):
|
| 34 |
+
fpath = os.path.join(path, fname)
|
| 35 |
+
if os.path.isfile(fpath):
|
| 36 |
+
try:
|
| 37 |
+
with open(fpath, "r", encoding="utf-8") as f:
|
| 38 |
+
text = f.read()
|
| 39 |
+
if text.strip(): # Only process non-empty files
|
| 40 |
+
emb = model.encode(text)
|
| 41 |
+
embeddings.append(emb)
|
| 42 |
+
texts.append(text)
|
| 43 |
+
filenames.append(fname)
|
| 44 |
+
except Exception as e:
|
| 45 |
+
print(f"Error reading file {fpath}: {e}")
|
| 46 |
+
|
| 47 |
+
if not embeddings:
|
| 48 |
+
print("No documents were successfully embedded.")
|
| 49 |
+
return []
|
| 50 |
+
|
| 51 |
+
# Create FAISS index
|
| 52 |
+
dimension = embeddings[0].shape[0]
|
| 53 |
+
index = faiss.IndexFlatIP(dimension)
|
| 54 |
+
|
| 55 |
+
# Normalize embeddings for cosine similarity
|
| 56 |
+
embeddings_matrix = np.array(embeddings).astype("float32")
|
| 57 |
+
faiss.normalize_L2(embeddings_matrix)
|
| 58 |
+
# Add embeddings to index
|
| 59 |
+
index.add(embeddings_matrix)
|
| 60 |
+
|
| 61 |
+
# Save FAISS index and metadata
|
| 62 |
+
os.makedirs("vector_cache", exist_ok=True)
|
| 63 |
+
faiss.write_index(index, "vector_cache/faiss_index.bin")
|
| 64 |
+
|
| 65 |
+
with open("vector_cache/metadata.pkl", "wb") as f:
|
| 66 |
+
pickle.dump({"texts": texts, "filenames": filenames}, f)
|
| 67 |
|
| 68 |
+
print(f"Saved FAISS index to vector_cache/ with {len(embeddings)} documents.")
|
| 69 |
print(f"Total embeddings created: {len(embeddings)}")
|
|
|
|
| 70 |
|
| 71 |
+
return list(zip(filenames, embeddings))
|