Remix.run Logo
zozbot234 5 hours ago

The "training" chips will probably be quite usable for slower, higher-throughput inference at scale. I expect that to be quite popular eventually for non-time-sensitive uses.