Amazon SageMaker 是一个完全托管的服务，用于快速轻松地构建、训练和部署机器学习（ML）模型。
Amazon SageMaker Experiments 是 Amazon SageMaker 的一项功能，可帮助您组织、跟踪、比较和评估 ML 实验与模型版本。
本笔记展示如何使用 LangChain Callback 将提示和其他 LLM 超参数记录并跟踪到 SageMaker Experiments。我们使用三个场景来展示其功能：
  • 场景 1：单个 LLM - 使用单个 LLM 模型根据给定提示生成输出。
  • 场景 2：顺序链 - 使用两个 LLM 模型组成的顺序链。
  • 场景 3：带工具的代理（思维链） - 在 LLM 之外使用多个工具（搜索和数学）。
在本笔记中，我们将创建一个实验以记录每个场景的提示。

​
安装与设置

pip install -qU  sagemaker
pip install -qU  langchain-openai
pip install -qU  google-search-results
首先，设置所需的 API 密钥： 
import os

## Add your API keys below
os.environ["OPENAI_API_KEY"] = "<ADD-KEY-HERE>"
os.environ["SERPAPI_API_KEY"] = "<ADD-KEY-HERE>"

from langchain_community.callbacks.sagemaker_callback import SageMakerCallbackHandler

from langchain.agents import initialize_agent, load_tools
from langchain.chains import LLMChain, SimpleSequentialChain
from langchain_core.prompts import PromptTemplate
from langchain_openai import OpenAI
from sagemaker.analytics import ExperimentAnalytics
from sagemaker.experiments.run import Run
from sagemaker.session import Session

​
LLM 提示跟踪

# LLM Hyperparameters
HPARAMS = {
    "temperature": 0.1,
    "model_name": "gpt-3.5-turbo-instruct",
}

# Bucket used to save prompt logs (Use `None` is used to save the default bucket or otherwise change it)
BUCKET_NAME = None

# Experiment name
EXPERIMENT_NAME = "langchain-sagemaker-tracker"

# Create SageMaker Session with the given bucket
session = Session(default_bucket=BUCKET_NAME)

​
场景 1 - LLM

RUN_NAME = "run-scenario-1"
PROMPT_TEMPLATE = "tell me a joke about {topic}"
INPUT_VARIABLES = {"topic": "fish"}

with Run(
    experiment_name=EXPERIMENT_NAME, run_name=RUN_NAME, sagemaker_session=session
) as run:
    # 创建 SageMaker 回调
    sagemaker_callback = SageMakerCallbackHandler(run)

    # 定义带回调的 LLM 模型
    llm = OpenAI(callbacks=[sagemaker_callback], **HPARAMS)

    # 创建提示模板
    prompt = PromptTemplate.from_template(template=PROMPT_TEMPLATE)

    # 创建 LLM 链
    chain = LLMChain(llm=llm, prompt=prompt, callbacks=[sagemaker_callback])

    # 运行链
    chain.run(**INPUT_VARIABLES)

    # 重置回调
    sagemaker_callback.flush_tracker()

​
场景 2 - 顺序链

RUN_NAME = "run-scenario-2"

PROMPT_TEMPLATE_1 = """You are a playwright. Given the title of play, it is your job to write a synopsis for that title.
Title: {title}
Playwright: This is a synopsis for the above play:"""
PROMPT_TEMPLATE_2 = """You are a play critic from the New York Times. Given the synopsis of play, it is your job to write a review for that play.
Play Synopsis: {synopsis}
Review from a New York Times play critic of the above play:"""

INPUT_VARIABLES = {
    "input": "documentary about good video games that push the boundary of game design"
}

with Run(
    experiment_name=EXPERIMENT_NAME, run_name=RUN_NAME, sagemaker_session=session
) as run:
    # 创建 SageMaker 回调
    sagemaker_callback = SageMakerCallbackHandler(run)

    # 为链创建提示模板
    prompt_template1 = PromptTemplate.from_template(template=PROMPT_TEMPLATE_1)
    prompt_template2 = PromptTemplate.from_template(template=PROMPT_TEMPLATE_2)

    # 定义带回调的 LLM 模型
    llm = OpenAI(callbacks=[sagemaker_callback], **HPARAMS)

    # 创建 chain1
    chain1 = LLMChain(llm=llm, prompt=prompt_template1, callbacks=[sagemaker_callback])

    # 创建 chain2
    chain2 = LLMChain(llm=llm, prompt=prompt_template2, callbacks=[sagemaker_callback])

    # 创建顺序链
    overall_chain = SimpleSequentialChain(
        chains=[chain1, chain2], callbacks=[sagemaker_callback]
    )

    # 运行整个顺序链
    overall_chain.run(**INPUT_VARIABLES)

    # 重置回调
    sagemaker_callback.flush_tracker()

​
场景 3 - 带工具的代理

RUN_NAME = "run-scenario-3"
PROMPT_TEMPLATE = "Who is the oldest person alive? And what is their current age raised to the power of 1.51?"

with Run(
    experiment_name=EXPERIMENT_NAME, run_name=RUN_NAME, sagemaker_session=session
) as run:
    # 创建 SageMaker 回调
    sagemaker_callback = SageMakerCallbackHandler(run)

    # 定义带回调的 LLM 模型
    llm = OpenAI(callbacks=[sagemaker_callback], **HPARAMS)

    # 定义工具
    tools = load_tools(["serpapi", "llm-math"], llm=llm, callbacks=[sagemaker_callback])

    # 使用所有工具初始化代理
    agent = initialize_agent(
        tools, llm, agent="zero-shot-react-description", callbacks=[sagemaker_callback]
    )

    # 运行代理
    agent.run(input=PROMPT_TEMPLATE)

    # 重置回调
    sagemaker_callback.flush_tracker()

​
加载日志数据

提示被记录后，我们可以轻松地将其加载并转换为 Pandas DataFrame，如下所示。 
# Load
logs = ExperimentAnalytics(experiment_name=EXPERIMENT_NAME)

# Convert as pandas dataframe
df = logs.dataframe(force_refresh=True)

print(df.shape)
df.head()
如上所示，该实验中有三个运行（行），对应每个场景。每次运行都会将提示和相关 LLM 设置/超参数以 json 形式记录，并保存在 S3 存储桶中。您可以随意加载并探索各 json 路径的日志数据。
Edit the source of this page on GitHub.
Connect these docs programmatically to Claude, VSCode, and more via MCP for real-time answers.