Hugging Face's logo

Spaces:
JTS-AI
/
hf_rabbit_life_poc
Sleeping

App Files Community
hf_rabbit_life_poc
File size: 4,730 Bytes 
30adccc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
 
from langchain_huggingface import HuggingFaceEmbeddings
from langchain_mongodb.vectorstores import MongoDBAtlasVectorSearch
from langchain_mongodb.retrievers.hybrid_search import MongoDBAtlasHybridSearchRetriever
import os
from dotenv import load_dotenv

load_dotenv()

# ---- MongoDB credentials ----
# mongo_username = os.getenv('MONGO_USERNAME')
# mongo_password = os.getenv('MONGO_PASSWORD')
mongo_database = os.getenv('MONGO_DATABASE')
mongo_connection_str = os.getenv('MONGO_CONNECTION_STRING')
mongo_collection_name = os.getenv('MONGO_COLLECTION')

# ---- Common Configurations & Hybrid Retrieval Configuration ----
MODEL_KWARGS = {"device": "cpu"}
ENCODE_KWARGS = {"normalize_embeddings": True,
                 "batch_size": 32}
EMBEDDING_DIMENSIONS = 1024
MODEL_NAME = "BAAI/bge-m3"
FINAL_TOP_K = 50 # 30
HYBRID_FULLTEXT_PENALTY = 0 # 60
HYBRID_VECTOR_PENALTY = 0.8 # 60

# ---- Embedding model ----
embed_model = HuggingFaceEmbeddings(
    model_name=MODEL_NAME, 
    model_kwargs=MODEL_KWARGS, 
    encode_kwargs=ENCODE_KWARGS
)

# ---- Vectore Search ----
num_vector_candidates = max(20, 2 * FINAL_TOP_K)
num_text_candidates = max(20, 2 * FINAL_TOP_K)
vector_k = num_vector_candidates
vector_num_candidates_for_operator = vector_k * 10

# ---- Vectore Store ----
vector_store = MongoDBAtlasVectorSearch.from_connection_string(
    connection_string=mongo_connection_str,
    namespace=f"{mongo_database}.{mongo_collection_name}",
    embedding=embed_model,
    index_name="search_index_v1",
)

# ---- Retriever (Hybrid) ----

def get_retriever(**kwargs):
    """
    สร้าง Retriever โดยสามารถรับ filter พิเศษสำหรับ Vector Search ได้
    """
    # ดึง vector_search_filter ออกมาจาก kwargs
    vector_search_filter = kwargs.pop('vector_search_filter', None)
    
    # kwargs ที่เหลือ (ถ้ามี) จะถูกใช้เป็น pre_filter
    pre_filter = kwargs if kwargs else None

    retriever = MongoDBAtlasHybridSearchRetriever(
        vectorstore=vector_store,
        search_index_name='search_index_v1',
        embedding=embed_model,
        text_key= 'text',
        embedding_key='embedding',
        top_k=FINAL_TOP_K,
        vector_penalty=HYBRID_VECTOR_PENALTY,
        fulltext_penalty=HYBRID_FULLTEXT_PENALTY,
        vector_search_params={
            "k": vector_k,
            "numCandidates": vector_num_candidates_for_operator,
            # --- ส่ง filter ที่ถูกต้องเข้าไปในตำแหน่งที่ถูกต้อง ---
            "filter": vector_search_filter
        },
        text_search_params={
            "limit": max(20, 2 * FINAL_TOP_K)
        },
        pre_filter=pre_filter
    )
    return retriever

# def get_retriever(**kwargs):
#     retriever = MongoDBAtlasHybridSearchRetriever(
#         vectorstore=vector_store,
#         search_index_name='search_index_v1',
#         embedding=embed_model,
#         text_key= 'text', #'token',
#         embedding_key='embedding',
#         top_k=FINAL_TOP_K,
#         vector_penalty=HYBRID_VECTOR_PENALTY,
#         fulltext_penalty=HYBRID_FULLTEXT_PENALTY,
#         vector_search_params={
#             "k": vector_k,
#             "numCandidates": vector_num_candidates_for_operator
#         },
#         text_search_params={
#             "limit": num_text_candidates
#         },
#         pre_filter=kwargs
#     )
#     return retriever



# ---------- FILTER METAAAAA ----------
# ---------- FILTER METAAAAA ----------
# ---------- FILTER METAAAAA ----------
# ---------- FILTER METAAAAA ----------

# def get_retriever(**kwargs):
#     # ดึง filter ที่เราจะส่งมาจาก tool ออกมาจาก kwargs
#     # เราใช้ .pop() เพื่อเอามันออกมา จะได้ไม่ถูกส่งไปที่ pre_filter ซ้ำซ้อน
#     search_filter = kwargs.pop('filter', None)

#     retriever = MongoDBAtlasHybridSearchRetriever(
#         vectorstore=vector_store,
#         search_index_name='search_index_v1',
#         embedding=embed_model,
#         text_key= 'text',
#         embedding_key='embedding',
#         top_k=FINAL_TOP_K,
#         vector_penalty=HYBRID_VECTOR_PENALTY,
#         fulltext_penalty=HYBRID_FULLTEXT_PENALTY,
#         vector_search_params={
#             "k": vector_k,
#             "numCandidates": vector_num_candidates_for_operator,
#             "filter": search_filter
#         },
#         text_search_params={
#             "limit": num_text_candidates
#         },
#         pre_filter=kwargs 
#     )
#     return retriever