KAG/knext/common/base/component.py
zhuzhongshu123 e1d818dfaa refactor(all): kag v0.6 (#174)
* add path find

* fix find path

* spg guided relation extraction

* fix dict parse with same key

* rename graphalgoclient to graphclient

* rename graphalgoclient to graphclient

* file reader supports http url

* add checkpointer class

* parser supports checkpoint

* add build

* remove incorrect logs

* remove logs

* update examples

* update chain checkpointer

* vectorizer batch size set to 32

* add a zodb backended checkpointer

* add a zodb backended checkpointer

* fix zodb based checkpointer

* add thread for zodb IO

* fix(common): resolve mutlithread conflict in zodb IO

* fix(common): load existing zodb checkpoints

* update examples

* update examples

* fix zodb writer

* add docstring

* fix jieba version mismatch

* commit kag_config-tc.yaml

1、rename type to register_name
2、put a uniqe & specific name to register_name
3、rename reader to scanner
4、rename parser to reader
5、rename num_parallel to num_parallel_file, rename chain_level_num_paralle to num_parallel_chain_of_file
6、rename kag_extractor to schema_free_extractor, schema_base_extractor to schema_constraint_extractor
7、pre-define llm & vectorize_model and refer them in the yaml file

Issues to be resolved:
1、examples of event extract & spg extract
2、statistic of indexer, such as nums of nodes & edges extracted, ratio of llm invoke.
3、Exceptions such as Debt, account does not exist should be thrown in llm invoke.
4、conf of solver need to be re-examined.

* commit kag_config-tc.yaml

1、rename type to register_name
2、put a uniqe & specific name to register_name
3、rename reader to scanner
4、rename parser to reader
5、rename num_parallel to num_parallel_file, rename chain_level_num_paralle to num_parallel_chain_of_file
6、rename kag_extractor to schema_free_extractor, schema_base_extractor to schema_constraint_extractor
7、pre-define llm & vectorize_model and refer them in the yaml file

Issues to be resolved:
1、examples of event extract & spg extract
2、statistic of indexer, such as nums of nodes & edges extracted, ratio of llm invoke.
3、Exceptions such as Debt, account does not exist should be thrown in llm invoke.
4、conf of solver need to be re-examined.

* 1、fix bug in base_table_splitter

* 1、fix bug in base_table_splitter

* 1、fix bug in default_chain

* 增加solver

* add kag

* update outline splitter

* add main test

* add op

* code refactor

* add tools

* fix outline splitter

* fix outline prompt

* graph api pass

* commit with page rank

* add search api and graph api

* add markdown report

* fix vectorizer num batch compute

* add retry for vectorize model call

* update markdown reader

* update markdown reader

* update pdf reader

* raise extractor failure

* add default expr

* add log

* merge jc reader features

* rm import

* add build

* fix zodb based checkpointer

* add thread for zodb IO

* fix(common): resolve mutlithread conflict in zodb IO

* fix(common): load existing zodb checkpoints

* update examples

* update examples

* fix zodb writer

* add docstring

* fix jieba version mismatch

* commit kag_config-tc.yaml

1、rename type to register_name
2、put a uniqe & specific name to register_name
3、rename reader to scanner
4、rename parser to reader
5、rename num_parallel to num_parallel_file, rename chain_level_num_paralle to num_parallel_chain_of_file
6、rename kag_extractor to schema_free_extractor, schema_base_extractor to schema_constraint_extractor
7、pre-define llm & vectorize_model and refer them in the yaml file

Issues to be resolved:
1、examples of event extract & spg extract
2、statistic of indexer, such as nums of nodes & edges extracted, ratio of llm invoke.
3、Exceptions such as Debt, account does not exist should be thrown in llm invoke.
4、conf of solver need to be re-examined.

* commit kag_config-tc.yaml

1、rename type to register_name
2、put a uniqe & specific name to register_name
3、rename reader to scanner
4、rename parser to reader
5、rename num_parallel to num_parallel_file, rename chain_level_num_paralle to num_parallel_chain_of_file
6、rename kag_extractor to schema_free_extractor, schema_base_extractor to schema_constraint_extractor
7、pre-define llm & vectorize_model and refer them in the yaml file

Issues to be resolved:
1、examples of event extract & spg extract
2、statistic of indexer, such as nums of nodes & edges extracted, ratio of llm invoke.
3、Exceptions such as Debt, account does not exist should be thrown in llm invoke.
4、conf of solver need to be re-examined.

* 1、fix bug in base_table_splitter

* 1、fix bug in base_table_splitter

* 1、fix bug in default_chain

* update outline splitter

* add main test

* add markdown report

* code refactor

* fix outline splitter

* fix outline prompt

* update markdown reader

* fix vectorizer num batch compute

* add retry for vectorize model call

* update markdown reader

* raise extractor failure

* rm parser

* run pipeline

* add config option of whether to perform llm config check, default to false

* fix

* recover pdf reader

* several components can be null for default chain

* 支持完整qa运行

* add if

* remove unused code

* 使用chunk兜底

* excluded source relation to choose

* add generate

* default recall 10

* add local memory

* 排除相似边

* 增加保护

* 修复并发问题

* add debug logger

* 支持topk参数化

* 支持chunk截断和调整spo select 的prompt

* 增加查询请求保护

* 增加force_chunk配置

* fix entity linker algorithm

* 增加sub query改写

* fix md reader dup in test

* fix

* merge knext to kag parallel

* fix package

* 修复指标下跌问题

* scanner update

* scanner update

* add doc and update example scripts

* fix

* add bridge to spg server

* add format

* fix bridge

* update conf for baike

* disable ckpt for spg server runner

* llm invoke error default raise exceptions

* chore(version): bump version to X.Y.Z

* update default response generation prompt

* add method getSummarizationMetrics

* fix(common): fix project conf empty error

* fix typo

* 增加上报信息

* 修改main solver

* postprocessor support spg server

* 修改solver支持名

* fix language

* 修改chunker接口,增加openapi

* rename vectorizer to vectorize_model in spg server config

* generate_random_string start with gen

* add knext llm vector checker

* add knext llm vector checker

* add knext llm vector checker

* solver移除默认值

* udpate yaml and register_name for baike

* udpate yaml and register_name for baike

* remove config key check

* 修复llmmodule

* fix knext project

* udpate yaml and register_name for examples

* udpate yaml and register_name for examples

* Revert "udpate yaml and register_name for examples"

This reverts commit b3fa5ca9ba749e501133ac67bd8746027ab839d9.

* update register name

* fix

* fix

* support multiple resigter names

* update component

* update reader register names (#183)

* fix markdown reader

* fix llm client for retry

* feat(common): add processed chunk id checkpoint (#185)

* update reader register names

* add processed chunk id checkpoint

* feat(example): add example config (#186)

* update reader register names

* add processed chunk id checkpoint

* add example config file

* add max_workers parameter for getSummarizationMetrics to make it faster

* add csqa data generation script generate_data.py

* commit generated csqa builder and solver data

* add csqa basic project files

* adjust split_length and num_threads_per_chain to match lightrag settings

* ignore ckpt dirs

* add csqa evaluation script eval.py

* save evaluation scripts summarization_metrics.py and factual_correctness.py

* save LightRAG output csqa_lightrag_answers.json

* ignore KAG output csqa_kag_answers.json

* add README.md for CSQA

* fix(solver): fix solver pipeline conf (#191)

* update reader register names

* add processed chunk id checkpoint

* add example config file

* update solver pipeline config

* fix project create

* update links and file paths

* reformat csqa kag_config.yaml

* reformat csqa python files

* reformat getSummarizationMetrics and compare_summarization_answers

* fix(solver): fix solver config (#192)

* update reader register names

* add processed chunk id checkpoint

* add example config file

* update solver pipeline config

* fix project create

* fix main solver conf

* add except

* fix typo in csqa README.md

* feat(conf): support reinitialize config for call from java side (#199)

* update reader register names

* add processed chunk id checkpoint

* add example config file

* update solver pipeline config

* fix project create

* fix main solver conf

* support reinitialize config for java call

* revert default response generation prompt

* update project list

* add README.md for the hotpotqa, 2wiki and musique examples

* 增加spo检索

* turn off kag config dump by default

* turn off knext schema dump by default

* add .gitignore and fix kag_config.yaml

* add README.md for the medicine example

* add README.md for the supplychain example

* bugfix for risk mining

* use exact out

* refactor(solver): format solver code (#205)

* update reader register names

* add processed chunk id checkpoint

* add example config file

* update solver pipeline config

* fix project create

* fix main solver conf

* support reinitialize config for java call

* black format

---------

Co-authored-by: peilong <peilong.zpl@antgroup.com>
Co-authored-by: 锦呈 <zhangxinhong.zxh@antgroup.com>
Co-authored-by: zhengke.gzk <zhengke.gzk@antgroup.com>
Co-authored-by: huaidong.xhd <huaidong.xhd@antgroup.com>
2025-01-03 17:10:51 +08:00

121 lines
3.7 KiB
Python

# -*- coding: utf-8 -*-
# Copyright 2023 OpenSPG Authors
#
# Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except
# in compliance with the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software distributed under the License
# is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express
# or implied.
from abc import ABC
from typing import List, Union, Type
import networkx as nx
from knext.common.base.runnable import Runnable
from knext.common.base.restable import RESTable
class Component(Runnable, RESTable, ABC):
"""
Base class for all component.
"""
@property
def id(self):
return str(id(self))
@property
def type(self):
return
@property
def label(self):
return
@property
def name(self):
return self.__class__.__name__
def to_dict(self):
return {"id": self.id, "name": self.name}
def __hash__(self):
return id(self)
def __eq__(self, other):
return hash(self) == hash(other)
def __rshift__(
self,
other: Union[
Type["Chain"],
List[Type["Chain"]],
Type["Component"],
List[Type["Component"]],
None,
],
):
"""
Implements the right shift operator to support chaining.
This method allows connecting components or chains (self) with other components or chains (other),
forming a new workflow graph. It supports connecting single or multiple components,
single or multiple chains.
Parameters:
other (Union[Type["Chain"], List[Type["Chain"]], Type["Component"], List[Type["Component"]], None]):
The component(s) or chain(s) to connect with.
Returns:
Chain: A new Chain instance representing the connected workflow.
"""
from knext.common.base.chain import Chain
if not other:
return self
if not isinstance(other, list):
other = [other]
dag_list = []
for o in other:
# If o is empty, create an empty directed graph and add self to it
if not o:
dag = nx.DiGraph()
self.last = True
dag.add_node(self)
dag_list.append(dag)
# If o is an instance of Component, create a directed graph and add edges between self and o
elif isinstance(o, Component):
dag = nx.DiGraph()
dag.add_node(self)
dag.add_node(o)
dag.add_edge(self, o)
dag_list.append(dag)
# If o is an instance of Chain, create a directed graph and combine it with o's graph
elif isinstance(o, Chain):
dag = nx.DiGraph()
dag.add_node(self)
end_nodes = [
node
for node, out_degree in dag.out_degree()
if out_degree == 0 or node.last
]
start_nodes = [
node for node, in_degree in o.dag.in_degree() if in_degree == 0
]
if len(end_nodes) > 0 and len(start_nodes) > 0:
for end_node in end_nodes:
for start_node in start_nodes:
dag.add_edge(end_node, start_node)
combined_dag = nx.compose(dag, o.dag)
dag_list.append(combined_dag)
# Combine all subgraphs into a final directed graph
final_dag = nx.compose_all(dag_list)
return Chain(dag=final_dag)