Langchain学习日记


字数统计: 1.1k字   |   阅读时长≈ 6分

啥都不懂，随便学学

Day 1

用的是OpenRouter的Minimax 2.5，因为是免费的而且参数稍微多一点

一开始想直接上手搞个能调用工具的模型的，结果Documentation没看完，学艺不精，最后只写了一个半成品：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
from langchain.tools import tool, ToolRuntime
from langchain_openrouter import ChatOpenRouter
from dotenv import load_dotenv
from pydantic import BaseModel, Field
from base64 import b64decode

load_dotenv()

@tool
def decode_base64(encoded_str: str) -> str:
    """Decode a base64 encoded string."""
    try:
        decoded_bytes = b64decode(encoded_str)
        return decoded_bytes.decode('utf-8')
    except Exception as e:
        return f"String {encoded_str} seems not to be a valid base64 encoded string. Error: {str(e)}"

@tool
def decode_hex(encoded_str: str) -> str:
    """Decode a hex encoded string."""
    try:
        decoded_bytes = bytes.fromhex(encoded_str)
        return decoded_bytes.decode('utf-8')
    except Exception as e:
        return f"String {encoded_str} seems not to be a valid hex encoded string. Error: {str(e)}"

raw_model = ChatOpenRouter(
    model="minimax/minimax-m2.5:free",
    temperature=0.7,
    max_retries=5,
)

model = raw_model.bind_tools([decode_base64, decode_hex])

messages = [("system", "You are a helpful assistant that can decode base64 and hex encoded strings. If the user provides an encoded string, use the appropriate tool to decode it and provide the decoded result. If the string is not valid, return an error message."),
            ("user", "Decode the following string: NDE3NTcyNmY3MjYxN2I1NDY4Njk3MzVmMzE3MzVmNDA1Zjc0NjU3Mzc0NjY2YzYxNjc3ZA==")]
response = model.invoke(messages)
print(response)
print(response.content)

搞笑的是content是空的，盯着response原因找了很久都不清楚，后面问了Gemini才知道这一步模型只是提出了工具调用的策略，实际的调用和反馈还得自己搞，感觉自己糖丸了

后面根据文档改了一下代码：

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
from langchain.tools import tool, ToolRuntime
from langchain_openrouter import ChatOpenRouter
from langchain.agents import create_agent
from langchain.agents.middleware import wrap_tool_call
from langchain_core.messages import ToolMessage
from dotenv import load_dotenv
from pydantic import BaseModel, Field
from base64 import b64decode

load_dotenv()

@tool
def decode_base64(encoded_str: str) -> str:
    """Decode a base64 encoded string."""
    decoded_bytes = b64decode(encoded_str)
    return decoded_bytes.decode('utf-8')

@tool
def decode_hex(encoded_str: str) -> str:
    """Decode a hex encoded string."""
    decoded_bytes = bytes.fromhex(encoded_str)
    return decoded_bytes.decode('utf-8')
    
@wrap_tool_call
def error_handler(request, handler):
    try:
        return handler(request)
    except Exception as e:
        return ToolMessage(
            content=f"An error occurred while processing the request: {str(e)}",
            tool_call_id=request.tool_call["id"],
        )

raw_model = ChatOpenRouter(
    model="minimax/minimax-m2.5:free",
    temperature=0.7,
    max_retries=5,
)

agent = create_agent(
    model=raw_model,
    tools=[decode_base64, decode_hex],
    middleware=[error_handler],
    system_prompt="You are a helpful assistant that can decode base64 and hex encoded strings. If the user provides an encoded string, use the appropriate tool to decode it and provide the decoded result. If the string is not valid, return an error message."
)

messages = {"messages": [
    {"role": "user", "content": "Decode the following string: NDE3NTcyNmY3MjYxN2I1NDY4Njk3MzVmMzE3MzVmNDA1Zjc0NjU3Mzc0NjY2YzYxNjc3ZA=="},
]}

response = agent.invoke(messages)
print(response)

好消息是能跑了，坏消息是不太像平时遇到的大模型，而且最后一大坨State dict闹麻了，所以想着再跟着文档写个流式的Agent

其实从这里开始已经是Day 2了（笑）

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
from langchain.tools import tool, ToolRuntime
from langchain_openrouter import ChatOpenRouter
from langchain.agents import create_agent
from langchain.agents.middleware import wrap_tool_call
from langchain_core.messages import ToolMessage, AIMessage
from openrouter.errors import TooManyRequestsResponseError
from dotenv import load_dotenv
from pydantic import BaseModel, Field
from base64 import b64decode
import time

load_dotenv()

@tool
def decode_base64(encoded_str: str) -> str:
    """Decode a base64 encoded string."""
    decoded_bytes = b64decode(encoded_str)
    return decoded_bytes.decode('utf-8')

@tool
def decode_hex(encoded_str: str) -> str:
    """Decode a hex encoded string."""
    decoded_bytes = bytes.fromhex(encoded_str)
    return decoded_bytes.decode('utf-8')
    
@wrap_tool_call
def error_handler(request, handler):
    try:
        return handler(request)
    except Exception as e:
        return ToolMessage(
            content=f"An error occurred while processing the request: {str(e)}",
            tool_call_id=request.tool_call["id"],
        )

raw_model = ChatOpenRouter(
    model="minimax/minimax-m2.5:free",
    temperature=0.7,
    max_retries=50,
)
agent = create_agent(
    model=raw_model,
    tools=[decode_base64, decode_hex],
    middleware=[error_handler],
    system_prompt="You are a helpful assistant that can decode base64 and hex encoded strings. If the user provides an encoded string, use the appropriate tool to decode it and provide the decoded result. If the string is not valid, return an error message."
).with_retry(
    retry_if_exception_type=(Exception,TooManyRequestsResponseError,),
    wait_exponential_jitter=True,
    stop_after_attempt=50,
)

messages = {"messages": [
    {"role": "user", "content": "Decode the following string: NDE3NTcyNmY3MjYxN2I1NDY4Njk3MzVmMzE3MzVmNDA1Zjc0NjU3Mzc0NjY2YzYxNjc3ZA=="},
]}

# response = agent.invoke(messages)
for chunk in agent.stream(messages, stream_mode="values"):
    latest_message = chunk["messages"][-1]
    if isinstance(latest_message, AIMessage):
        if latest_message.additional_kwargs.get("reasoning_content") != '':
            print(f"AI reasoning: {latest_message.additional_kwargs['reasoning_content']}")
        if latest_message.tool_calls != []:
            print(f"AI called tool: {latest_message.tool_calls[0]['name']}")
        if latest_message.content != '':
            print(f"AI response: {latest_message.content}")
    time.sleep(5)

由于用的是免费的模型，所以爆429（Too Many Requests）是日常，因此被逼的把retry拉到50，还加了个time.sleep ╮(╯-╰)╭

好在最后跑出来了：

Day 1就先这样吧，Blast Rivals也马上要开了，该看比赛了

Day 2

看文档感觉有点乱，在完全不了解的情况下很难系统化学习LC，所以找了LC自己的教程LangChain Academy，感觉会好一些

刚好Day 1做的就是Module 1的前两部分，所以就简单过一下前面的视频和概念就行

  • Temperature：控制模型输出的随机度

  • Max Token：控制模型输出的长度

  • Max Retries和Timeout…就是字面意思

扫一扫，分享到微信

tag:

    缺失模块。
    1、请确保node版本大于6.2
    2、在博客根目录（注意不是yilia-plus根目录）执行以下命令：
    npm i hexo-generator-json-content --save

    3、在根目录_config.yml里添加配置： 

      jsonContent:
    meta: false
    pages: false
    posts:
      title: true
      date: true
      path: true
      text: false
      raw: false
      content: false
      slug: false
      updated: false
      comments: false
      link: false
      permalink: false
      excerpt: false
      categories: false
      tags: true

一个喜欢睡觉的fw~~~