Skip to content

Comment on Infrastructure setup and open-source scripts to train 70B model from bare metal

Comments

once the model is trained, what happens to the hardware and infrastructure?

Voltage Park is a cloud provider. This is no different from renting barebone infra from AWS, GCP or Azure.

Except Voltage Park, being smaller, is probably more willing to provide some customized setup.

Indeed, they may even see it as a learning opportunity for when they rent similar setups to other customers.

Imbue was telling Voltage Park how to setup and wire their network and booting from bare metal, so it’s definitely lower level than what the big clouds provide access to.

It probably isn't the answer but should be - LAN party.

GPUs will be reused for mining Monero and exfiltrate money to the founders at the expense of the investors.

Oops, don't tell I told you.

EDIT: Sorry Dogecoin, thanks to the tip!

Monero uses a CPU-optimized consensus algorithm. Dogecoin is a better bet.

Are you suggesting training models is a cover for mining crypto? The hardware is dual-purpose...

Either training the next model or inference for the already trained one. In some cases, you might even offer it as a service.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.