How hard is it to deploy pretrained models on GPUs without tons of YAML files and unmanaged instances?
Re: pricing. I think you're comparing Modal's GPU price (without cpu and mem) with AWS GPU+CPU+MEM price
Good point. I will update the calculations when I have more time :)
Thanks for writing this, it was a great read. What are some things you think Modal is still lacking?
Maybe fine-grained control over the container scheduling algorithm? There's an experimental feature for this: https://modal.com/docs/guide/concurrent-inputs
In regards to storage, there is support for persistent file systems but it's a bit different than S3 - so maybe object storage support as well?
Re: pricing. I think you're comparing Modal's GPU price (without cpu and mem) with AWS GPU+CPU+MEM price
Good point. I will update the calculations when I have more time :)
Thanks for writing this, it was a great read. What are some things you think Modal is still lacking?
Maybe fine-grained control over the container scheduling algorithm? There's an experimental feature for this: https://modal.com/docs/guide/concurrent-inputs
In regards to storage, there is support for persistent file systems but it's a bit different than S3 - so maybe object storage support as well?