12 Replies
i don't know if theres any limits on this, what do you need
the limitation is on the build time, i think its bit over an hour, we plan to optimize the time by next week and make it faster
We just connected a network volume for now. It seems faster
yes that would be faster if you were adding model into the container image, its a huge reduction to size
I do sometimes have an issue with delay time. What could be the cause? It is still downloading the models from the network volume? The second one is the ECR version ( we don't want to use this due to 12 hour refresh is needed)
You should check if file exists and don't re-download if it exist, in your code
Ecr Version? What do you mean
ECR Version (ECR image I mean my bad)
it is good but we have to refresh credentials and github integration to runpod is way easieeeeeer
I just need to get over the hump of the network volume and we golden
Oh yes I see.. You can use graphql api to automatically refresh your token
https://discord.com/channels/912829806415085598/1196527741974106223/1278056539210514557
yea! Although My team now like the network volume + github integration for now so we can just focus on developing
what are you putting in network volume? whats causing the delay time? make sure to have at least 3 or more workers; rolling releases won't impact your current workload so even with a new release, some workers will handle the workload using current image until new container image is loaded
It's okay now lol when we used 4090 (24g pro) the delay time shortened drastically
@nerdylive thank you also for the advice in the cron job we manage to automate the refresh 🔥