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
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308 | class DocumentSummaryIndex(BaseIndex[IndexDocumentSummary]):
"""文档摘要索引。
Args:
response_synthesizer (BaseSynthesizer): 用于生成摘要的响应合成器。
summary_query (str): 用于为每个文档生成摘要的查询。
show_progress (bool): 是否显示tqdm进度条。
默认为False。
embed_summaries (bool): 是否嵌入摘要。
这是运行默认基于嵌入的检索器所必需的。
默认为True。
"""
index_struct_cls = IndexDocumentSummary
def __init__(
self,
nodes: Optional[Sequence[BaseNode]] = None,
objects: Optional[Sequence[IndexNode]] = None,
index_struct: Optional[IndexDocumentSummary] = None,
llm: Optional[LLM] = None,
embed_model: Optional[BaseEmbedding] = None,
storage_context: Optional[StorageContext] = None,
response_synthesizer: Optional[BaseSynthesizer] = None,
summary_query: str = DEFAULT_SUMMARY_QUERY,
show_progress: bool = False,
embed_summaries: bool = True,
# deprecated
service_context: Optional[ServiceContext] = None,
**kwargs: Any,
) -> None:
"""初始化参数。"""
self._llm = llm or llm_from_settings_or_context(Settings, service_context)
self._embed_model = embed_model or embed_model_from_settings_or_context(
Settings, service_context
)
self._response_synthesizer = response_synthesizer or get_response_synthesizer(
llm=self._llm, response_mode=ResponseMode.TREE_SUMMARIZE
)
self._summary_query = summary_query
self._embed_summaries = embed_summaries
super().__init__(
nodes=nodes,
index_struct=index_struct,
service_context=service_context,
storage_context=storage_context,
show_progress=show_progress,
objects=objects,
**kwargs,
)
@property
def vector_store(self) -> BasePydanticVectorStore:
return self._vector_store
def as_retriever(
self,
retriever_mode: Union[str, _RetrieverMode] = _RetrieverMode.EMBEDDING,
**kwargs: Any,
) -> BaseRetriever:
"""获取检索器。
Args:
retriever_mode（Union[str，DocumentSummaryRetrieverMode]）：检索器模式。
默认为DocumentSummaryRetrieverMode.EMBEDDING。
"""
from llama_index.core.indices.document_summary.retrievers import (
DocumentSummaryIndexEmbeddingRetriever,
DocumentSummaryIndexLLMRetriever,
)
LLMRetriever = DocumentSummaryIndexLLMRetriever
EmbeddingRetriever = DocumentSummaryIndexEmbeddingRetriever
if retriever_mode == _RetrieverMode.EMBEDDING:
if not self._embed_summaries:
raise ValueError(
"Cannot use embedding retriever if embed_summaries is False"
)
return EmbeddingRetriever(
self,
object_map=self._object_map,
embed_model=self._embed_model,
**kwargs,
)
if retriever_mode == _RetrieverMode.LLM:
return LLMRetriever(
self, object_map=self._object_map, llm=self._llm, **kwargs
)
else:
raise ValueError(f"Unknown retriever mode: {retriever_mode}")
def get_document_summary(self, doc_id: str) -> str:
"""根据文档ID获取文档摘要。
Args:
doc_id（str）：文档ID。
"""
if doc_id not in self._index_struct.doc_id_to_summary_id:
raise ValueError(f"doc_id {doc_id} not in index")
summary_id = self._index_struct.doc_id_to_summary_id[doc_id]
return self.docstore.get_node(summary_id).get_content()
def _add_nodes_to_index(
self,
index_struct: IndexDocumentSummary,
nodes: Sequence[BaseNode],
show_progress: bool = False,
) -> None:
"""将节点添加到索引。"""
doc_id_to_nodes = defaultdict(list)
for node in nodes:
if node.ref_doc_id is None:
raise ValueError(
"ref_doc_id of node cannot be None when building a document "
"summary index"
)
doc_id_to_nodes[node.ref_doc_id].append(node)
summary_node_dict = {}
items = doc_id_to_nodes.items()
iterable_with_progress = get_tqdm_iterable(
items, show_progress, "Summarizing documents"
)
for doc_id, nodes in iterable_with_progress:
print(f"current doc id: {doc_id}")
nodes_with_scores = [NodeWithScore(node=n) for n in nodes]
# get the summary for each doc_id
summary_response = self._response_synthesizer.synthesize(
query=self._summary_query,
nodes=nodes_with_scores,
)
summary_response = cast(Response, summary_response)
metadata = doc_id_to_nodes.get(doc_id, [TextNode()])[0].metadata
summary_node_dict[doc_id] = TextNode(
text=summary_response.response,
relationships={
NodeRelationship.SOURCE: RelatedNodeInfo(node_id=doc_id)
},
metadata=metadata,
)
self.docstore.add_documents([summary_node_dict[doc_id]])
logger.info(
f"> Generated summary for doc {doc_id}: " f"{summary_response.response}"
)
for doc_id, nodes in doc_id_to_nodes.items():
index_struct.add_summary_and_nodes(summary_node_dict[doc_id], nodes)
if self._embed_summaries:
summary_nodes = list(summary_node_dict.values())
id_to_embed_map = embed_nodes(
summary_nodes, self._embed_model, show_progress=show_progress
)
summary_nodes_with_embedding = []
for node in summary_nodes:
node_with_embedding = node.copy()
node_with_embedding.embedding = id_to_embed_map[node.node_id]
summary_nodes_with_embedding.append(node_with_embedding)
self._vector_store.add(summary_nodes_with_embedding)
def _build_index_from_nodes(
self, nodes: Sequence[BaseNode]
) -> IndexDocumentSummary:
"""从节点构建索引。"""
# first get doc_id to nodes_dict, generate a summary for each doc_id,
# then build the index struct
index_struct = IndexDocumentSummary()
self._add_nodes_to_index(index_struct, nodes, self._show_progress)
return index_struct
def _insert(self, nodes: Sequence[BaseNode], **insert_kwargs: Any) -> None:
"""插入一个文档。"""
self._add_nodes_to_index(self._index_struct, nodes)
def _delete_node(self, node_id: str, **delete_kwargs: Any) -> None:
pass
def delete_nodes(
self,
node_ids: List[str],
delete_from_docstore: bool = False,
**delete_kwargs: Any,
) -> None:
"""从索引中删除节点列表。
Args:
node_ids（List[str]）：要删除的节点的node_ids列表
"""
index_nodes = self._index_struct.node_id_to_summary_id.keys()
for node in node_ids:
if node not in index_nodes:
logger.warning(f"node_id {node} not found, will not be deleted.")
node_ids.remove(node)
self._index_struct.delete_nodes(node_ids)
remove_summary_ids = [
summary_id
for summary_id in self._index_struct.summary_id_to_node_ids
if len(self._index_struct.summary_id_to_node_ids[summary_id]) == 0
]
remove_docs = [
doc_id
for doc_id in self._index_struct.doc_id_to_summary_id
if self._index_struct.doc_id_to_summary_id[doc_id] in remove_summary_ids
]
for doc_id in remove_docs:
self.delete_ref_doc(doc_id)
def delete_ref_doc(
self, ref_doc_id: str, delete_from_docstore: bool = False, **delete_kwargs: Any
) -> None:
"""从索引中删除一个文档。
与该文档相关的所有节点将被删除。
"""
ref_doc_info = self.docstore.get_ref_doc_info(ref_doc_id)
if ref_doc_info is None:
logger.warning(f"ref_doc_id {ref_doc_id} not found, nothing deleted.")
return
self._index_struct.delete(ref_doc_id)
self._vector_store.delete(ref_doc_id)
if delete_from_docstore:
self.docstore.delete_ref_doc(ref_doc_id, raise_error=False)
self._storage_context.index_store.add_index_struct(self._index_struct)
@property
def ref_doc_info(self) -> Dict[str, RefDocInfo]:
"""获取已摄取文档及其节点和元数据的字典映射。"""
ref_doc_ids = list(self._index_struct.doc_id_to_summary_id.keys())
all_ref_doc_info = {}
for ref_doc_id in ref_doc_ids:
ref_doc_info = self.docstore.get_ref_doc_info(ref_doc_id)
if not ref_doc_info:
continue
all_ref_doc_info[ref_doc_id] = ref_doc_info
return all_ref_doc_info
|