The borrow checker doesn't decide when things are dropped. It only checks reference uses and doesn't generate any code. This will work exactly the same as long as your program doesn't violate any borrowing rules.
But that fine-tuning is done only on those 100-200 good samples. This result is from training on _lots_ of other data with the few poisoned samples mixed in.
This doesn't really make sense to me. Most cats I've known react to such reflections without ever having seen a laser pointer in their life, for the same reason they react to laser pointers.
JSON allows you to store arbitrarily large integers/floats. It's only in JS this is a problem, not if you use JSON in languages that support larger (than 54-bit) integers.
Correct me if I'm wrong, but I think you still need to compute gradients of non-trained weights in order to compute the gradients of the LoRA weights. What you don't have to do is store and update the optimizer state for all those non-trained weights.
The angular separation would be only 0.00013 arcseconds, so very impressive indeed. Hubble has an angular resolution of around 0.05 arcseconds in comparison. Radio telescopes can get better angular resolution than that using interferometry between distantly separated telescopes. Perhaps that's how they can do it.
dvui?