Temporarily increased compile time limit of #GPUs to 120. (#121076)
Fixes #115331.
This is a temporary fix to increase the compile time number of GPUs to 120 until #119639 can be merged. Changing the parameter to 128 leads to annoying errors, as some checks would be tautological (`int8_t` is always < 128).
Pull Request resolved: https://github.com/pytorch/pytorch/pull/121076
Approved by: https://github.com/albanD