There's no quick way from the CLI to answer the question "is this sampler checkpoint actually serving right now?" tinker checkpoint info shows you size, type, TTL, all the static stuff.
But nothing about liveness. The only way to find out today is to round-trip through the Python SDK, instantiate a SamplingClient, encode a prompt, and call sample(). That's fine when you're already in a notebook; it's a lot of overhead before kicking off an eval where the first thing you want to know is whether the checkpoint is reachable at all.
A tinker checkpoint probe <tinker://...> subcommand that fires one tiny sample request and prints ok / fail / timeout + latency_ms would close that gap cleanly. Same shape as info, just with a liveness check stapled on.
There's no quick way from the CLI to answer the question "is this sampler checkpoint actually serving right now?"
tinker checkpoint infoshows you size, type, TTL, all the static stuff.But nothing about liveness. The only way to find out today is to round-trip through the Python SDK, instantiate a
SamplingClient, encode a prompt, and callsample(). That's fine when you're already in a notebook; it's a lot of overhead before kicking off an eval where the first thing you want to know is whether the checkpoint is reachable at all.A
tinker checkpoint probe <tinker://...>subcommand that fires one tiny sample request and printsok / fail / timeout + latency_mswould close that gap cleanly. Same shape asinfo, just with a liveness check stapled on.