Gregory's Server@skysailor @inthehands oh, I can see that.
Regarding the seed, there would be one used for training and then another for inference.
From my admittedly limited and slightly orthogonal experience - I've played with image gen models, not language ones: you can get the same output from a given trained model if you feed it the same prompt and the same seed. But, you can't train another copy of that model, even using the same source data, training parameters, and seed. Your "supposedly same" model will generate completely different outputs, even with the same prompt and inference seed. Sigh. This is all such an alchemy. :(
@dkalintsev @skysailor I suspect all this is a bit of a red herring. With a machine model, you can do things you could •never• do with an HR dept: Run it on 10 million resumes. Run it on repeatedly on the same resumes, altering on variable. Random? Run it on each 1000x. It’s a kind of broad testing that, should a court allow, would make many of the questions above evaporate.