Typically you need about 1GB graphics RAM for each billion parameters (i.e. one byte per parameter). This is a 405B parameter model. Ouch.
Edit: you can try quantizing it. This reduces the amount of memory required per parameter to 4 bits, 2 bits or even 1 bit. As you reduce the size, the performance of the model can suffer. So in the extreme case you might be able to run this in under 64GB of graphics RAM.
All junior devs should read OCs comment and really think about this.
The issue is whether
is_number()
is performing a semantic language matter or checking whether the text input can be converted by the program to a number type.The former case - the semantic language test - is useful for chat based interactions, analysis of text (and ancient text - I love the cuneiform btw) and similar. In this mode, some applications don’t even have to be able to convert the text into eg binary (a ‘gazillion’ of something is quantifying it, but vaguely)
The latter case (validating input) is useful where the input is controlled and users are supposed to enter numbers using a limited part of a standard keyboard. Clay tablets and triangular sticks are strictly excluded from this interface.
Another example might be
is_address()
. Which of these are addresses? ‘10 Downing Street, London’, ‘193.168.1.1’, ‘Gettysberg’, ‘Sir/Madam’.To me this highlights that code is a lot less reusable between different projects/apps than it at first appears.