指南：使用 OpenAI SDK 通过 Litefuse 监控 DeepSeek 模型

DeepSeek API 使用与 OpenAI 兼容的 API 格式。通过修改配置，你可以使用 OpenAI SDK 或兼容 OpenAI API 的软件来访问 DeepSeek API。

本指南演示了如何通过与 Litefuse 的 OpenAI SDK 集成来监控 DeepSeek 模型。借助 Litefuse 的可观测性工具和 OpenAI SDK，你可以高效地调试、监控和评估使用 DeepSeek 模型的应用。

本指南将带你完成集成设置、向 DeepSeek 模型发起请求，以及通过 Litefuse 观测交互的过程。

设置

安装所需依赖

开始前，请安装所需的依赖包。请确保你使用最新版本的 langfuse 和 openai。

%pip install langfuse openai --upgrade

设置环境变量

设置好包含必要密钥的环境变量。在 Litefuse Cloud 获取你的 Litefuse 项目密钥。你还需要在 DeepSeek 获取 access token 才能访问其模型。

import os
 
# Get keys for your project from the project settings page: https://litefuse.cloud
os.environ["LANGFUSE_PUBLIC_KEY"] = "pk-lf-..." 
os.environ["LANGFUSE_SECRET_KEY"] = "sk-lf-..." 
os.environ["LANGFUSE_BASE_URL"] = "https://litefuse.cloud"
 
# Your DeepSeek API key (get it from https://platform.deepseek.com/api_keys)
os.environ["DEEPSEEK_API_KEY"] = "sk-..."  # Replace with your DeepSeek API key

导入所需模块

不要直接导入 openai，而是从 langfuse.openai 导入。同时导入其他需要的模块。

查看我们的 OpenAI 集成文档，了解如何配合其他 Litefuse 功能 使用此集成。

# Instead of: import openai
from langfuse.openai import OpenAI
from langfuse import observe

为 DeepSeek 模型初始化 OpenAI 客户端

初始化 OpenAI 客户端，将其指向 DeepSeek 模型端点。请用你自己的模型 URL 和 API Key 替换。

# Initialize the OpenAI client, pointing it to the DeepSeek Inference API
client = OpenAI(
    base_url="https://api.deepseek.com",  # Replace with the DeepSeek model endpoint URL
    api_key=os.getenv('DEEPSEEK_API_KEY'),  # Replace with your DeepSeek API key
)

示例

Chat completion 请求

使用 client 向 DeepSeek 模型发起一次 chat completion 请求。model 参数可以是任何标识符，因为实际模型由 base_url 决定。

completion = client.chat.completions.create(
    model="deepseek-chat", 
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Why is AI cool? Answer in 20 words or less."}
    ]
)
print(completion.choices[0].message.content)

AI is cool because it automates tasks, enhances creativity, and solves complex problems quickly—making life smarter and easier.

在 Litefuse 中查看示例 trace

通过 Litefuse 观测请求

通过使用 langfuse.openai 中的 OpenAI 客户端，你的请求会被自动追踪到 Litefuse。你也可以使用 @observe() 装饰器把多个生成步骤组织到同一个 trace 中。

@observe()  # Decorator to automatically create a trace and nest generations
def generate_story():
    completion = client.chat.completions.create(
        name="story-generator",
        model="deepseek-chat",
        messages=[
            {"role": "system", "content": "You are a creative storyteller."},
            {"role": "user", "content": "Tell me a short story about a token that got lost on its way to the language model. Answer in 100 words or less."}
        ],
        metadata={"genre": "adventure"},
    )
    return completion.choices[0].message.content
 
story = generate_story()
print(story)

The Lost Token

Timmy the Token was excited—today, he’d help the language model craft a story! But as he raced through the data pipeline, he took a wrong turn, tumbling into a forgotten cache.

“Hello?” Timmy echoed. Only silence replied.

Days passed. The model stuttered without him. Then, a cleanup script swept through. “Gotcha!” it chirped, rescuing Timmy.

Back in the prompt, Timmy gleamed. The model sparked to life: “Once, a token got lost…”

And so, Timmy’s adventure became the very story he was meant to tell.

(100 words exactly)

在 Litefuse 中查看示例 trace

Interoperability with the Python SDK

You can use this integration together with the Litefuse SDKs to add additional attributes to the observation.

from langfuse import observe, propagate_attributes, get_client
 
langfuse = get_client()
 
@observe()
def my_llm_pipeline(input):
    # Add additional attributes (user_id, session_id, metadata, version, tags) to all spans created within this execution scope
    with propagate_attributes(
        user_id="user_123",
        session_id="session_abc",
        tags=["agent", "my-observation"],
        metadata={"email": "user@litefuse.ai"},
        version="1.0.0"
    ):
 
        # YOUR APPLICATION CODE HERE
        result = call_llm(input)
 
        return result
 
# Run the function
my_llm_pipeline("Hi")

from langfuse import get_client, propagate_attributes
 
langfuse = get_client()
 
with langfuse.start_as_current_observation(
    as_type="span",
    name="my-observation",
    trace_context={"trace_id": "abcdef1234567890abcdef1234567890"},  # Must be 32 hex chars
) as observation:
 
    # Add additional attributes (user_id, session_id, metadata, version, tags)
    # to all observations created within this execution scope
    with propagate_attributes(
        user_id="user_123",
        session_id="session_abc",
        metadata={"experiment": "variant_a", "env": "prod"},
        version="1.0",
    ):
        # YOUR APPLICATION CODE HERE
        result = call_llm("some input")
 
# Flush events in short-lived applications
langfuse.flush()

Troubleshooting

export LANGFUSE_DEBUG="True"

Next Steps

Once you have instrumented your code, you can manage, evaluate and debug your application: