关于MLOps
MLOps 更快地交付机器学习模型
一系列设计、构建和管理可重现、可测试和可持续的基于 ML 的软件实践。对于大数据 / 机器学习团队,MLOps 包含了大多数 DataOps 的任务以及其他特定于 ML 的任务,例如模型版本控制、测试、验证和监控
From Big Data to Good Data
Takeaways: Data-centric AI
百度架构:
训练:kubeflow+airflow+自己的工作流系统
预测:CRD + istio
调度器:volcano
# 参考资料
A Chat with Andrew on MLOps: From Model-centric to Data-centric AI:[video] (opens new window) [slides] (opens new window)
Practical MLOPS: HOW TO GET READY FORPRODUCTION MODELS: [blog] (opens new window) [eBook] (opens new window)
What the Ops are you talking about? [Blog (English)] (opens new window) [Blog (中译版)] (opens new window)
MLOps:机器学习中的持续交付和自动化流水线 [Blog] (opens new window)
从小作坊到智能中枢: MLOps简介:https://zhuanlan.zhihu.com/p/357897337
人类早期驯服野生机器学习模型的珍贵资料:https://zhuanlan.zhihu.com/p/330577488
Full Stack Deep Learning:https://zhuanlan.zhihu.com/p/218468169
深入学习AI:https://stevenjokess.github.io/2bPM/chapter_AI_dive/MLOps.html
fullstackdeeplearning:https://github.com/full-stack-deep-learning/fsdl-text-recognizer-2021-labs
MLOps 简介:https://segmentfault.com/a/1190000039957405
使用Kubeflow和Volcano实现典型AI训练任务:
https://support.huaweicloud.com/bestpractice-cce/cce_bestpractice_0075.html
- 02
- README 美化05-20
- 03
- 常见 Tricks 代码片段05-12