DevOps、MLOps与LLMOps的核心差异解析
DevOps vs. MLOps vs. LLMOps, clearly explained:
清晰梳理了 LLM 工程落地的运维范式差异，特别是监控指标与评估闭环的非线性特征，对做 Agent 和 RAG 系统的同学很有参考价值。
DevOps vs. MLOps vs. LLMOps, clearly explained:
DevOps、MLOps 与 LLMOps，清晰解析：
Many teams are trying to apply DevOps practices to LLM apps.
许多团队正试图将 DevOps 实践应用于大语言模型（LLM）应用。
But DevOps, MLOps, and LLMOps solve fundamentally different problems.
但 DevOps、MLOps 和 LLMOps 解决的是根本不同的问题。
DevOps is software-centric. You write code, test it, and deploy it. The feedback loop is straightforward, i.e., does the code work or not?
DevOps 以软件为中心。你编写代码、测试并部署它。反馈循环很直接，即：代码是否有效？
MLOps is model-centric. Here, you're dealing with data drift, model decay, and continuous retraining. The code might be fine, but the model's performance can degrade over time because the world changes.
MLOps 以模型为中心。在这里，你需要处理数据漂移、模型衰退以及持续重新训练等问题。代码可能没问题，但随着世界变化，模型的性能可能会随时间推移而下降。
LLMOps is foundation-model-centric. Here, you're typically not training models from scratch. Instead, you're selecting foundation models and then optimizing through three common paths:
LLMOps 以基础模型为中心。在这里，你通常不是从头训练模型。相反，你是选择基础模型，然后通过三条常见路径进行优化：
- Prompt engineering - Context/RAG setup - Fine-tuning
- 提示词工程 - 上下文/RAG 设置 - 微调
But here's what really separates LLMOps: The monitoring is completely different.
但真正区分 LLMOps 的是：监控方式完全不同。
In MLOps, you track data drift, model decay, and accuracy.
在 MLOps 中，你跟踪数据漂移、模型衰退和准确率。
In LLMOps, you're watching for:
在 LLMOps 中，你关注的是：
- Hallucination detection - Bias and toxicity - Token usage and cost - Human feedback loops
- 幻觉检测 - 偏见与毒性 - Token 使用量与成本 - 人类反馈循环
This is because you can't just check if the output is "correct." You need to ensure it's safe, grounded, and cost-effective.
这是因为你不能仅仅检查输出是否“正确”。你需要确保它是安全的、有依据的且具备成本效益。
The evaluation loop in LLMOps also feeds back into all three optimization paths simultaneously. Failed evals might mean you need better prompts, richer context, OR fine-tuning.
LLMOps 中的评估循环也会同时反馈到所有三个优化路径中。失败的评估可能意味着你需要更好的提示词、更丰富的上下文，或者进行微调。
So it's not a linear pipeline anymore.
因此，它不再是一个线性的流水线。
One more thing: prompt versioning and RAG pipelines are now first-class citizens in LLMOps, just like data versioning became essential in MLOps.
还有一点：提示词版本控制和 RAG 管道现在已成为 LLMOps 中的一等公民，就像数据版本控制在 MLOps 中变得不可或缺一样。
And the ops layer you choose should match the system you're building.
而你选择的运维层应与你正在构建的系统相匹配。
If you want to go deeper into LLMOps, I wrote a full LLM engineering roadmap a while back.
如果你想深入了解 LLMOps，我之前写过一份完整的 LLM 工程路线图。
It walks through the eight pillars of building LLM systems, starting at prompt engineering and ending at observability and safety, with free and open-source resources attached to each one.
它详细介绍了构建 LLM 系统的八大支柱，从提示词工程开始，到可观测性与安全结束，并为每个部分附带了免费和开源的资源。
You can read it below.
你可以在下方阅读。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力