Bases: BaseNodePostprocessor
前/后节点后处理器。
允许用户从文档存储中获取额外的节点,基于节点之间的前/后关系。
注意:与PrevNextPostprocessor的区别在于,这个推断了前进/后退的方向。
注意:这是一个测试版功能。
Parameters:
Name |
Type |
Description |
Default |
docstore |
BaseDocumentStore
|
|
required
|
num_nodes |
int
|
|
required
|
infer_prev_next_tmpl |
str
|
用于推断的模板。
必需字段为{context_str}和{query_str}。
|
required
|
Source code in llama_index/core/postprocessor/node.py
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
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355 | class AutoPrevNextNodePostprocessor(BaseNodePostprocessor):
"""前/后节点后处理器。
允许用户从文档存储中获取额外的节点,基于节点之间的前/后关系。
注意:与PrevNextPostprocessor的区别在于,这个推断了前进/后退的方向。
注意:这是一个测试版功能。
Args:
docstore (BaseDocumentStore): 文档存储。
num_nodes (int): 要返回的节点数(默认值:1)
infer_prev_next_tmpl (str): 用于推断的模板。
必需字段为{context_str}和{query_str}。"""
docstore: BaseDocumentStore
service_context: Optional[ServiceContext] = None
llm: Optional[LLM] = None
num_nodes: int = Field(default=1)
infer_prev_next_tmpl: str = Field(default=DEFAULT_INFER_PREV_NEXT_TMPL)
refine_prev_next_tmpl: str = Field(default=DEFAULT_REFINE_INFER_PREV_NEXT_TMPL)
verbose: bool = Field(default=False)
response_mode: ResponseMode = Field(default=ResponseMode.TREE_SUMMARIZE)
class Config:
"""此为pydantic对象的配置。"""
arbitrary_types_allowed = True
@classmethod
def class_name(cls) -> str:
return "AutoPrevNextNodePostprocessor"
def _parse_prediction(self, raw_pred: str) -> str:
"""解析预测。"""
pred = raw_pred.strip().lower()
if "previous" in pred:
return "previous"
elif "next" in pred:
return "next"
elif "none" in pred:
return "none"
raise ValueError(f"Invalid prediction: {raw_pred}")
def _postprocess_nodes(
self,
nodes: List[NodeWithScore],
query_bundle: Optional[QueryBundle] = None,
) -> List[NodeWithScore]:
"""后处理节点。"""
llm = self.llm or llm_from_settings_or_context(Settings, self.service_context)
if query_bundle is None:
raise ValueError("Missing query bundle.")
infer_prev_next_prompt = PromptTemplate(
self.infer_prev_next_tmpl,
)
refine_infer_prev_next_prompt = PromptTemplate(self.refine_prev_next_tmpl)
all_nodes: Dict[str, NodeWithScore] = {}
for node in nodes:
all_nodes[node.node.node_id] = node
# use response builder instead of llm directly
# to be more robust to handling long context
response_builder = get_response_synthesizer(
llm=llm,
text_qa_template=infer_prev_next_prompt,
refine_template=refine_infer_prev_next_prompt,
response_mode=self.response_mode,
)
raw_pred = response_builder.get_response(
text_chunks=[node.node.get_content()],
query_str=query_bundle.query_str,
)
raw_pred = cast(str, raw_pred)
mode = self._parse_prediction(raw_pred)
logger.debug(f"> Postprocessor Predicted mode: {mode}")
if self.verbose:
print(f"> Postprocessor Predicted mode: {mode}")
if mode == "next":
all_nodes.update(get_forward_nodes(node, self.num_nodes, self.docstore))
elif mode == "previous":
all_nodes.update(
get_backward_nodes(node, self.num_nodes, self.docstore)
)
elif mode == "none":
pass
else:
raise ValueError(f"Invalid mode: {mode}")
sorted_nodes = sorted(all_nodes.values(), key=lambda x: x.node.node_id)
return list(sorted_nodes)
|
Config
此为pydantic对象的配置。
Source code in llama_index/core/postprocessor/node.py
| class Config:
"""此为pydantic对象的配置。"""
arbitrary_types_allowed = True
|