File size: 11,302 Bytes
10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 79b8126 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 10d6a86 ae92cb7 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 |
import os, logging
from app.engine.logger import logger
from typing import List, Any
import pandas as pd
from weaviate.classes.config import Property, DataType
from .weaviate_interface_v4 import WeaviateWCS, WeaviateIndexer
from ..settings import parquet_file
from weaviate.classes.query import Filter
from torch import cuda
if os.path.exists('.we_are_local'):
COLLECTION = 'MultiRAG_local_mr'
else:
COLLECTION = 'MultiRAG'
class dummyWeaviate:
""" Created to pass on HF since I had again the client creation issue
Temporary solution
"""
def __init__(self,
endpoint: str=None,
api_key: str=None,
model_name_or_path: str='sentence-transformers/all-MiniLM-L6-v2',
embedded: bool=False,
openai_api_key: str=None,
skip_init_checks: bool=False,
**kwargs
):
return
def _connect(self) -> None:
return
def _client(self):
return
def create_collection(self,
collection_name: str,
properties: list[Property],
description: str=None,
**kwargs
) -> None:
return
def show_all_collections(self,
detailed: bool=False,
max_details: bool=False
) -> list[str] | dict:
return ['abc', 'def']
def show_collection_config(self, collection_name: str):
return
def show_collection_properties(self, collection_name: str):
return
def delete_collection(self, collection_name: str):
return
def get_doc_count(self, collection_name: str):
return
def keyword_search(self,
request: str,
collection_name: str,
query_properties: list[str]=['content'],
limit: int=10,
filter: Filter=None,
return_properties: list[str]=None,
return_raw: bool=False
):
return
def vector_search(self,
request: str,
collection_name: str,
limit: int=10,
return_properties: list[str]=None,
filter: Filter=None,
return_raw: bool=False,
device: str='cuda:0' if cuda.is_available() else 'cpu'
):
return
def hybrid_search(self,
request: str,
collection_name: str,
query_properties: list[str]=['content'],
alpha: float=0.5,
limit: int=10,
filter: Filter=None,
return_properties: list[str]=None,
return_raw: bool=False,
device: str='cuda:0' if cuda.is_available() else 'cpu'
):
return
class VectorStore:
def __init__(self, model_path: str = 'sentence-transformers/all-mpnet-base-v2'):
# we can create several instances to test various models, especially if we finetune one
self.MultiRAG_properties = [
Property(name='file',
data_type=DataType.TEXT,
description='Name of the file',
index_filterable=True,
index_searchable=True),
# Property(name='keywords',
# data_type=DataType.TEXT_ARRAY,
# description='Keywords associated with the file',
# index_filterable=True,
# index_searchable=True),
Property(name='content',
data_type=DataType.TEXT,
description='Splits of the article',
index_filterable=True,
index_searchable=True),
]
self.class_name = "MultiRAG_all-mpnet-base-v2"
self.class_config = {'classes': [
{"class": self.class_name,
"description": "multiple types of docs",
"vectorIndexType": "hnsw",
# Vector index specific app.settings for HSNW
"vectorIndexConfig": {
"ef": 64, # higher is better quality vs slower search
"efConstruction": 128, # higher = better index but slower build
"maxConnections": 32, # max conn per layer - higher = more memory
},
"vectorizer": "none",
"properties": self.MultiRAG_properties}
]
}
self.model_path = model_path
try:
self.api_key = os.environ.get('FINRAG_WEAVIATE_API_KEY')
logger(f"API key: {self.api_key[:5]}")
self.url = os.environ.get('FINRAG_WEAVIATE_ENDPOINT')
logger(f"URL: {self.url[8:15]}")
self.client = WeaviateWCS(
endpoint=self.url,
api_key=self.api_key,
model_name_or_path=self.model_path,
)
assert self.client._client.is_live(), "Weaviate is not live"
assert self.client._client.is_ready(), "Weaviate is not ready"
logger(f"Weaviate client created")
except Exception as e:
# raise Exception(f"Could not create Weaviate client: {e}")
self.client = dummyWeaviate() # used when issue with HF client creation, to continue on HF
logger(f"Could not create Weaviate client: {e}")
# if we fail these tests 'VectorStore' object has no attribute 'client'
# it's prob not the env var but the model missing
# assert self.client._client.is_live(), "Weaviate is not live"
# assert self.client._client.is_ready(), "Weaviate is not ready"
# careful with accessing '_client' since the weaviate helper usually closes the connection every time
self.indexer = None
self.create_collection()
@property
def collections(self):
return self.client.show_all_collections()
def create_collection(self,
collection_name: str=COLLECTION,
description: str='Documents'):
self.collection_name = collection_name
if collection_name not in self.collections:
self.client.create_collection(collection_name=collection_name,
properties=self.MultiRAG_properties,
description=description)
# self.collection_name = collection_name
else:
logger(f"Collection {collection_name} already exists")
def empty_collection(self, collection_name: str=COLLECTION) -> bool:
# not in the library yet, so I simply delete and recreate it
if collection_name in self.collections:
self.client.delete_collection(collection_name=collection_name)
self.create_collection()
return True
else:
logger(f"Collection {collection_name} doesn't exist")
return False
def index_data(self, data: List[dict]= None, collection_name: str=COLLECTION):
if self.indexer is None:
self.indexer = WeaviateIndexer(self.client)
if data is None:
# use the parquet file, otherwise use the data passed
data = pd.read_parquet(parquet_file).to_dict('records')
# the parquet file was created/incremented when a new article was uploaded
# it is a dataframe with columns: file, content, content_embedding
# and reflects exactly the data that we want to index at all times
self.status = self.indexer.batch_index_data(data, collection_name, 256)
self.num_errors, self.error_messages, self.doc_ids = self.status
# in this case with few articles, we don't tolerate errors
# batch_index_data already tests errors against a threshold
# assert self.num_errors == 0, f"Errors: {self.num_errors}"
def keyword_search(self,
query: str,
limit: int=5,
return_properties: List[str]=['file', 'content'],
alpha=None # dummy parameter to match the hybrid_search signature
) -> List[str]:
response = self.client.keyword_search(
request=query,
collection_name=self.collection_name,
query_properties=['file', 'content'],
limit=limit,
filter=None,
return_properties=return_properties,
return_raw=False)
return [(res['file'], res['content'], res['score']) for res in response]
def vector_search(self,
query: str,
limit: int=5,
return_properties: List[str]=['file', 'content'],
alpha=None # dummy parameter to match the hybrid_search signature
) -> List[str]:
response = self.client.vector_search(
request=query,
collection_name=self.collection_name,
limit=limit,
filter=None,
return_properties=return_properties,
return_raw=False)
return [(res['file'], res['content'], res['score']) for res in response]
def hybrid_search(self,
query: str,
limit: int=10,
alpha=0.5, # higher = more vector search
return_properties: List[str]=['file', 'content']
) -> List[str]:
response = self.client.hybrid_search(
request=query,
collection_name=self.collection_name,
query_properties=['file', 'content'],
alpha=alpha,
limit=limit,
filter=None,
return_properties=return_properties,
return_raw=False)
return [(res['file'], res['content'], res['score']) for res in response] |