On: Open-R1: an open reproduction of DeepSeek-R1

[brianstrimp]

Only meaningful if code+data deterministically reproduce the weights.

At that point, the weights are just the cached output. Which has value since it's costly to produce from code+data.

[frabcus → brianstrimp]

I don't think it needs to be deterministic - and if it isn't, having the data and code becomes even more important!

Compilers generally aren't deterministic (see the reproduceable build movement), yet we still use their output binaries.