[<Agent component: framework|tool|llm|etc...>]

[kbzh2558]

Description

请问如何运营rag agent自主进行多轮对话，谢谢。

例如：PROMPT_TEMPLATE="""

我的问题或指令：

{question}

请根据下面的参考信息回答我的问题或回复我的指令，并遵循以下改进后的指南：
. . .

您的回复应该遵循以下改进后的结构：
. . .

---

参考信息：

{context}
"""

目前我参照的模板是rag example中的llamaindex_rag.ipynb

import logging
import sys
import os

logging.basicConfig(stream=sys.stdout, level=logging.INFO)
logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))

from llama_index.core import (
SimpleDirectoryReader,
VectorStoreIndex,
Settings
)
from modelscope import snapshot_download

Specify multiple document paths

document_paths = [
'D:\PPT&WORD\folder\knowledge'
# Add more document paths as needed
]

Load data from multiple documents

documents = []
for path in document_paths:
documents.extend(SimpleDirectoryReader(path).load_data())

embedding_name='damo/nlp_gte_sentence-embedding_chinese-base'
local_embedding = snapshot_download(embedding_name)
Settings.embed_model = "local:"+local_embedding

index = VectorStoreIndex.from_documents(documents)

os.environ['ZHIPU_API_KEY']='apikey'

from modelscope_agent.agents import RolePlay

role_template = '知识库查询小助手，可以优先通过查询本地知识库来回答用户的问题'
llm_config = {
'model': 'GLM-4',
'model_server': 'zhipu'
}
function_list = []

bot = RolePlay(function_list=function_list,llm=llm_config, instruction=role_template)

index_ret = index.as_retriever(similarity_top_k=3)
query = "西安交通大学图书馆有几部分组成？"
result = index_ret.retrieve(query)
print(result)

ref_doc = ' '.join([doc.text for doc in result])
response = bot.run("西安交通大学图书馆有几部分组成？", remote=False, print_info=True, ref_doc=ref_doc)
text = ''
for chunk in response:
text += chunk
print(text)

Link

No response

[suluyana]

可以参考以下代码：

`
from modelscope_agent.memory import MemoryWithRag
from modelscope_agent.agents import RolePlay

role_template = '知识库查询小助手，可以优先通过查询本地知识库来回答用户的问题'
llm_config = {
'model': 'GLM-4',
'model_server': 'zhipu'
}
function_list = []
file_paths = ['./tests/samples/常见QA.pdf']
bot = RolePlay(function_list=function_list,llm=llm_config, instruction=role_template)
memory = MemoryWithRag(urls=file_paths, use_knowledge_cache=False)
use_llm = True if len(function_list) else False

query = "高德天气API在哪申请"
ref_doc = memory.run(query, use_llm=use_llm)

response = bot.run(query, remote=False, print_info=True, ref_doc=ref_doc)
text = ''
for chunk in response:
text += chunk
print(text)
`