jono_irwin

3 ms·

1. 1.▲
   Reduce GVisor Cold Starts with GPU Snapshotting(cerebrium.ai)

   51 points by jono_irwin 3mo ago|17 comments
2. 2.▲
   by jono_irwin 7mo ago

   Yeah that’s fair. For weights specifically there often isn’t a huge dedupe win across versions since retraining tends to change most of them. That said, we generally don’t advocate including model weights in container images anyway. The mai
3. 3.▲
   by jono_irwin 7mo ago

   That approach works really well when you have a stable shared base image. Where it starts to get harder is when you have multiple base stacks (different CUDA versions, frameworks, etc.) or when you need to update them frequently. You end up
4. 4.▲
   by jono_irwin 7mo ago

   Good point, network dependency is a valid concern. In practice these systems typically fetch data over a local, highly available network and aggressively cache anything that gets read. If that network path becomes unavailable, it usually in
5. 5.▲
   by jono_irwin 7mo ago

   hey cosmotic, we're not really advocating for storing model weights in the container image. even the smaller nvidia images (like nvidia/cuda:13.1.1-cudnn-runtime-ubuntu24.04) are about 2Gb before adding any python deps and that is
6. 6.▲
   by jono_irwin 2y ago

   Thanks for the feedback! I like the sound of all of those: - clearer messaging - more tutorials - one-click deploys - clear & upfront costing We have plans to add other runtimes (like Typescript) in the future but Python is our focus fo
7. 7.▲
   by jono_irwin 2y ago

   There are definitely some parallels between Cerebrium and paperspace, but I don't think they are a direct competitor. The biggest difference being that paperspace doesn't have a serverless offering afaik. Cerebrium abstracts some