Это видео недоступно.
Сожалеем об этом.
Effective Observability for MLOps Pipelines at Scale with Rishit Dagli
HTML-код
- Опубликовано: 14 авг 2024
- Join Rishit Dagli as he explores effective observability for ML pipelines at scale. Learn about the critical differences between observability and monitoring in ML applications, common challenges like distribution shifts, and feedback loops. Rishit demonstrates practical methods for logging and interpreting various metrics to maintain model performance and reliability.
Learn more about Civo Navigate here -►www.civo.com/n...
====================
Get free credit to try the world’s first K3s-powered, managed Kubernetes service.
Sign up to Civo -► www.civo.com
Get started with Kubernetes with Civo Academy: www.civo.com/a...
Subscribe to our RUclips Channel -► civo.io/subscribe
Follow Civo:
• Twitter -► / civocloud
• Github -► github.com/civo
• LinkedIn -► / civocloud
• Facebook -► / civocloud
0:00 - Introduction and Background
0:27 - Observability vs. Monitoring
0:55 - Importance of Observability in ML
1:17 - Challenges in ML Observability
3:00 - Distribution Shifts and Their Impact
4:03 - Handling Data and Feature Changes
5:06 - Issues with Feedback Loops
7:14 - Case Study: InstructPix2Pix
8:21 - Monitoring Inference Effects
10:10 - Problems with Traditional Metrics
11:12 - Logging System Metrics
12:00 - Logging Model-Related Metrics
13:38 - Practical Example: Setting Up Monitoring
17:00 - Implementing Prometheus and Grafana
19:20 - Visualizing Metrics in Grafana
21:08 - Conclusion and Q&A