Lightning Talk: Multiplayer Machine Learning with Metaflow, OpenAI Whisper and...
summary
Demonstrates using Metaflow with OpenAI Whisper on Kubernetes to transcribe three audio URLs in parallel, running tiny and small Whisper models per file and joining results; tiny runs with about 2 CPUs and 1 GB, small runs with about 4 CPUs and 8 GB, each step executing as a separate Kubernetes pod. The flow runs locally for development then on Kubernetes for scale, completes in minutes in the demo, and produces viewable transcripts in the Metaflow UI; can scale to many files, use GPUs for speed, and trigger summarization and Slack delivery.
AI-written summary · about
chapters (8)
- 0:00Introduction to multiplayer machine learning
- 1:03Metaflow capabilities
- 3:08Whisper model sizes
- 4:11Multi-model transcription flow
- 5:55Kubernetes resource allocation
- 6:57Demo results & UI overview
- 8:03Extending for real-world use
- 9:48Q&A
related talks
35mBuilding Robust Reactive ML Systems In A Multiplayer Setting
32mLesson learned from orchestrating large-scale GenAI, ML & data on k8s
36mHuman-Friendly, Production-Ready Data Science Stack with Metaflow & Kubernetes
32mHuman-Friendly, Production-Ready Data Science Stack With Metaflow & Kubernetes
35mTricks Learned from Scaling WhisperSpeech Models to 80k+ Hours of Speech
40mOptimizing AI/ML Workloads
22mBuild and Monitor Machine Learning Services in Kubernetes
27mApplied Machine Learning: a Netflix production