CSPON-TSC/aiAgent_gd/qwen_agent/memory/plan_memory.py

# from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores.faiss import FAISS
from langchain.schema import Document

import jsonlines
import json
import os

from qwen_agent.memory.SqlMemory import embeddings
from qwen_agent.utils.util import get_data_from_jsons
os.environ["TOKENIZERS_PARALLELISM"] = "false"


class PlanExampleRetrieval():
    def __init__(self, query_type='bidding') -> None:
        # self.EMBEDDING_MODEL = "text2vec" # embedding 模型，对应 embedding_model_dict
        # self.DEVICE = "cuda:2"

        # self.embeddings = HuggingFaceEmbeddings(model_name=embedding_model_dict[self.EMBEDDING_MODEL],)
        # self.embeddings.client = sentence_transformers.SentenceTransformer(self.embeddings.model_name,
        #                                                                 device=self.DEVICE)
        few_shot_docs = []
        # embeddings = OpenAIEmbeddings() LianqiaiAgent/qwen_agent/memory/data/ifbunitplan_examples.jsonl
        # 修改成自动读取下面的多个json,方便扩展
        data = get_data_from_jsons(os.path.join(os.path.abspath(os.path.dirname(__file__)), 'data'), 'plan_examples')

        for line in data:
            if line['query_type'] == query_type:
                few_shot_docs.append(Document(page_content=line['query'], metadata={'plan':line['plan']}))

        self.vector_db = FAISS.from_documents(few_shot_docs, embeddings)

    def get_relevant_documents(self,query,top_k=4):
        results=[]
        for r in self.vector_db.similarity_search(query, k=top_k):
            results.append((r.page_content, r.metadata['plan']))
        return results


if __name__=="__main__":
    print(os.path.abspath(os.path.dirname(__file__)))
    print(get_data_from_jsons(os.path.join(os.path.abspath(os.path.dirname(__file__)), 'data'), 'plans'))
    # print(os.path.join(os.path.abspath(os.path.dirname(__file__)),'data/sqls.jsonl'))