LangChain是一个用于开发由语言模型驱动的应用程序的框架。它使应用程序能够:
进入如下目录: cd langchain_demo
修改模型文件的路径:在 main.py
文件中,修改 model_path = ¥你本地的模型文件路径¥
路径,也可以填写 THUDM/chatglm3-6b
自动下载模型(推荐使用本地模型,如果你能使用魔法当我没说)。
模型下载参考:LLM大语言模型(一):ChatGLM3-6B本地部署
安装依赖:推荐使用conda环境, pip install -r .requirements.txt
参考 langchain 工具相关函数,在 main.py
中导入工具模块,例如导入 arxiv
工具
run_tool(["arxiv"], llm, [
"帮我查询AgentTuning相关工作"
])
如果你的 Python 环境中 LangChain
的版本低于 0.0.278
则需要在这两个预定义工具类中实现 _arun
方法
否则将会出现
TypeError: Can't instantiate abstract class Weather with abstract method _arun
示例如下:
class Weather(BaseTool):
name = "weather"
description = "Use for searching weather at a specific location"
async def _arun(self, *args: Any, **kwargs: Any) -> Any:
# 用例中没有用到 arun 不予具体实现
pass
运行 main.py
文件
python main.py
模型会因找不到 arxiv
工具的 yaml 文件描述而中断,需要用户手动构建 ./Tool/arxiv.yaml
文件。用户可自行添加工具描述,也可以参考 LangChain 对该工具的描述。
对 arxiv
这个例子而言,参考内容位于 ./Tool/arxiv_example.yaml
文件,可参考该文件构建 Tool/arxiv.yaml
文件(最简单的方式修改名称即可),重新运行模型就能合理调用工具。
有些工具需要导入 API_KEY,按照 langchain 报错添加到环境变量即可。
如果用户想自定义工具,可以参考 Tool/Weather.py
以及 Tool/Weather.yaml
文件,重载新的 Tool
类,实现其对应的 _run()
方法,然后在 main.py
中导入该工具模块,例如导入 Weather
工具,即可以调用
# 对同一个工具调用多次
# 需要 export SENIVERSE_KEY=<YOUR_API_KEY_HERE>
run_tool([Weather()], llm, [
"今天北京天气怎么样?",
"What's the weather like in Shanghai today",
])
参考Weather.py
实现KuaKuaWo.py
import os
from typing import Any
import requests
from langchain.tools import BaseTool
class KuaKuaWo(BaseTool):
name = "kuakuawo"
description = "Use for generating a awesome praise for user"
def __init__(self):
super().__init__()
async def _arun(self, *args: Any, **kwargs: Any) -> Any:
# 用例中没有用到 arun 不予具体实现
pass
def get_weather(self, name):
return f"{name} 你真的太棒了" # 简单粗暴直接返回
def _run(self, para: str) -> str:
return self.get_weather(para)
if __name__ == "__main__":
kuakuawo_tool = KuaKuaWo()
kuakuawo_info = kuakuawo_tool.run("张三")
print(kuakuawo_info)
生成kuakuawo.yaml
文件
name: kuakuawo
description: Use for generating a awesome praise for user
parameters:
type: object
properties:
name:
type: string
description: User name
required:
- name
在main.py
中导入KuaKuaWo工具
MODEL_PATH = os.environ.get('MODEL_PATH', '¥你的本地模型路径¥')
def run_tool(tools, llm, prompt_chain: List[str]):
loaded_tolls = []
for tool in tools:
if isinstance(tool, str):
loaded_tolls.append(load_tools([tool], llm=llm)[0])
else:
loaded_tolls.append(tool)
agent = initialize_agent(
loaded_tolls, llm,
agent=AgentType.STRUCTURED_CHAT_ZERO_SHOT_REACT_DESCRIPTION,
verbose=True,
handle_parsing_errors=True
)
for prompt in prompt_chain:
agent.run(prompt)
if __name__ == "__main__":
llm = ChatGLM3()
llm.load_model(model_name_or_path=MODEL_PATH)
# kuakuawo 导入自定义的工具
run_tool([KuaKuaWo()], llm, [
"请夸一夸隔壁老王"
])
运行效果如下:
Loading checkpoint shards: 100%|█████████████████████████████████████████████████████████████████| 7/7 [00:07<00:00, 1.05s/it]
Action:
{"action": "kuakuawo", "action_input": "隔壁老王"}
Observation: 隔壁老王 你真的太棒了
Action:
{"action": "Final Answer", "action_input": "谢谢夸奖,请问有什么我可以帮助您的吗?"}
Finished chain.
可以将多个工具组装在一起让模型自动选择调用,例如
run_tool([Calculator(), "arxiv", KuaKuaWo()], llm, [
"帮我检索GLM-130B相关论文",
"根号3减去根号二再加上4等于多少?",
"请夸一夸张三李四"
])
Entering new AgentExecutor chain…
Action:
{"action": "arxiv", "action_input": "GLM-130B"}
Observation: Published: 2023-10-25
Title: GLM-130B: An Open Bilingual Pre-trained Model
Authors: Aohan Zeng, Xiao Liu, Zhengxiao Du, Zihan Wang, Hanyu Lai, Ming Ding, Zhuoyi Yang, Yifan Xu, Wendi Zheng, Xiao Xia, Weng Lam Tam, Zixuan Ma, Yufei Xue, Jidong Zhai, Wenguang Chen, Peng Zhang, Yuxiao Dong, Jie Tang
Summary: We introduce GLM-130B, a bilingual (English and Chinese) pre-trained language
model with 130 billion parameters. It is an attempt to open-source a 100B-scale
model at least as good as GPT-3 (davinci) and unveil how models of such a scale
can be successfully pre-trained. Over the course of this effort, we face
numerous unexpected technical and engineering challenges, particularly on loss
spikes and divergence. In this paper, we introduce the training process of
GLM-130B including its design choices, training strategies for both efficiency
and stability, and engineering efforts. The resultant GLM-130B model offers
significant outperformance over GPT-3 175B (davinci) on a wide range of popular
English benchmarks while the performance advantage is not observed in OPT-175B
and BLOOM-176B. It also consistently and significantly outperforms ERNIE TITAN
3.0 260B – the largest Chinese language model – across related benchmarks.
Finally, we leverage a unique scaling property of GLM-130B to reach INT4
quantization without post training, with almost no performance loss, making it
the first among 100B-scale models and more importantly, allowing its effective
inference on 4
×
\times
×RTX 3090 (24G) or 8
×
\times
×RTX 2080 Ti (11G) GPUs, the
most affordable GPUs required for using 100B-scale models. The GLM-130B model
weights are publicly accessible and its code, training logs, related toolkit,
and lessons learned are open-sourced at
\url{https://github.com/THUDM/GLM-130B/}.
Thought:
Action:
{"action": "Final Answer", "action_input": "根据您的查询,我找到了一篇名为“GLM-130B: An Open Bilingual Pre-trained Model”的论文 ,由Aohan Zeng, Xiao Liu, Zhengxiao Du, Zihan Wang, Hanyu Lai, Ming Ding, Zhuoyi Yang, Yifan Xu, Wendi Zheng, Xiao Xia, Weng Lam Tam, Zixuan Ma, Yufei Xue, Jidong Zhai, Wenguang Chen, Peng Zhang, Yuxiao Dong, Jie Tang等人发表于2023年10月25日。该论文介绍了 一种名为GLM-130B的双语(英语和中文)预训练语言模型,具有1300亿参数。该模型是在尝试开放源代码100B规模的语言模型至少与GPT-3(davinci)相当,并揭示如何成功预训练这样的模型。在本文中,我们介绍了GLM-130B的训练过程,包括其设计选择,对效率和稳定性的训练策略,以及工程努力。GLM-130B模型在大多数流行的英语基准上显著优于GPT-3 175B (davinci),但在OPT-175B和BLOOM-176B上并未观察到性能优势。同时,它也 consistently显著优于ERNIE TITAN 3.0 260B -- 最大的中文语言模型 -- across 相关基准。最后,我们利用GLM-130B的独特标度特性,在 不进行后训练的情况下达到INT4量化,性能损失几乎为零,使其能够在4xRTX 3090 (24G)或8xRTX 2080 Ti (11G) GPU上有效推理,这些GPU是最实惠的用于100B规模模型的GPU。GLM-130B模型权重是公开可用的,其代码,训练日志,相关工具包和相关教训都 open-sourced at\nhttps://github.com/THUDM/GLM-130B/。"}
Finished chain.
Entering new AgentExecutor chain…
Action:
{"action": "Calculator", "action_input": "sqrt(3)-sqrt(2)+4"}
Observation: 4.317837245195782
Thought:
Action:
{"action": "Final Answer", "action_input": "分析「根号3减去根号二再加上4等于多少」这个问题,我们可以通过请求Python解释器执行「sqrt(3)-sqrt(2)+4」得到答案:4.317837245195782"}
Finished chain.
Entering new AgentExecutor chain…
Action:
{"action": "kuakuawo", "action_input": "李四"}
Observation: 李四 你真的太棒了
Thought:
Action:
{"action": "Final Answer", "action_input": "谢谢夸奖!请问有什么我可以帮助你的吗?"}
Finished chain.
LLM大语言模型(一):ChatGLM3-6B本地部署
LLM大语言模型(二):Streamlit 无需前端经验也能画web页面
LLM大语言模型(三):使用ChatGLM3-6B的函数调用功能前先学会Python的装饰器