llamaindex 解决报错 Can‘t instantiate abstract class BaseNode with abstract methods get_content

简介

使用llamaindex 做RAG实验过程中，在利用BM25算相似度时，遇到了如下报错；
给出了该报错的解决办法；

报错信息如下

Traceback (most recent call last):
  File "C:\Users\js\Desktop\cache\llama_index_demo\recall\bm25_fix_bug.py", line 50, in <module>
    res = retriever._retrieve(QueryBundle(q))
  File "C:\Users\js\anaconda3\envs\llm\lib\site-packages\llama_index\legacy\retrievers\bm25_retriever.py", line 99, in _retrieve
    scored_nodes = self._get_scored_nodes(query_bundle.query_str)
  File "C:\Users\js\anaconda3\envs\llm\lib\site-packages\llama_index\legacy\retrievers\bm25_retriever.py", line 91, in _get_scored_nodes
    nodes.append(NodeWithScore(node=node, score=doc_scores[i]))
  File "C:\Users\js\anaconda3\envs\llm\lib\site-packages\pydantic\v1\main.py", line 341, in __init__
    raise validation_error
pydantic.v1.error_wrappers.ValidationError: 1 validation error for NodeWithScore
node
  Can't instantiate abstract class BaseNode with abstract methods get_content, get_metadata_str, get_type, hash, set_content (type=type_error)

fix bug

常规的数据加载

from llama_index.core import (
    SimpleDirectoryReader
)
from llama_index.core.node_parser import SentenceWindowNodeParser
from llama_index.legacy.retrievers import BM25Retriever
from llama_index.legacy.schema import NodeWithScore, QueryBundle, Node

# Load data
documents = SimpleDirectoryReader(
    input_files=["./data/paul_graham_essay.txt"]
).load_data()

# create the sentence window node parser w/ default settings
node_parser = SentenceWindowNodeParser.from_defaults(
    window_size=3,
    window_metadata_key="window",
    original_text_metadata_key="original_text",
)

# Extract nodes from documents
nodes = node_parser.get_nodes_from_documents(documents)

# by default, the node ids are set to random uuids. To ensure same id's per run, we manually set them.
for idx, node in enumerate(nodes):
    node.id_ = f"node_{idx}"

重写方法

根据报错信息所示，_get_scored_nodes 方法报错；
解决思路：使用一个子类继承BM25Retriever， 重写_get_scored_nodes方法

class JieRetriever(BM25Retriever):
    def _get_scored_nodes(self, query: str):
        tokenized_query = self._tokenizer(query)
        doc_scores = self.bm25.get_scores(tokenized_query)
        nodes = []
        for i, node in enumerate(self._nodes):
            node_new = Node.from_dict(node.to_dict())
            node_with_score = NodeWithScore(node=node_new, score=doc_scores[i])
            nodes.append(node_with_score)
        return nodes


if __name__ == '__main__':
    retriever = JieRetriever.from_defaults(
    # retriever = BM25Retriever.from_defaults(
        similarity_top_k=5,
        nodes=nodes,
    )

    q = 'Before college the two main things I worked on, outside of school, were writing and programming'
    from pprint import pprint

    res = retriever._retrieve(QueryBundle(q))
    for item in res:
        pprint([item.node.id_, item.score, item.node.get_content()])

关键代码如下，创建一个新的Node：

node_new = Node.from_dict(node.to_dict())
node_with_score = NodeWithScore(node=node_new, score=doc_scores[i])

参考资料

遇到该报错后，花了一段时间意识到这个是llamda-index包的bug，然后便直接在issues中检索该报错信息。果然发现有人和我一样遇到了该问题，然后参考了他人给出的解决办法。

• https://github.com/run-llama/llama_index/issues/10697