Artem Andreenko

We will provide deb/rpm/etc repositories in future. Thank you for advise. https://github.com/meshbird/meshbird/issues/39

Hello Joe! > This looks very smart and a new way to overcome NAT. > > About 5% of remote devices will not be able to do UDP based NAT...

It's an intriguing idea. We could close the loop by distilling a training dataset and making it dynamic, and then continuously refining the model with new observed data.

It would be great if we can visualize update waves inside models day to day. And maybe, incorporate how waves propogates into mdoel itself.

@pauldog Could you please talk a little bit more about your fine tuning pipeline and foundation model? What hardware are you using for it?

What other models have you used?

It looks it's possible to pack it to a AWS Lambda on ARM Gravitron + S3 weights offloading.

@neuhaus kswapd0 process is pretty active

@dalnk Do you have swap in the system?

@MarkSchmidty thank you for sharing your results. I believe my system swaped a lot due to limit size of RAM (4GB RAM, model size 4GB).