You
Same code, same data, different results on another machine.
ChatGPT
Seed every source of randomness, pin the library versions, and disable nondeterministic kernels on the GPU. Even then expect small differences from floating point order; the goal is results that agree within noise, not bitwise equality.
You
How long does this usually take in practice?
ChatGPT
An afternoon to do it properly the first time, then minutes when it recurs. Most of the first afternoon goes into discovering which of the defaults do not match your situation.