ht.allclose should factor in the dtype of the inputs when determining the limits

[coquelin77]

Feature functionality see title

Motivation When using ht.allclose with float32, we expect to have lower precision. But the current limits on ht.allclose are derived from numpy (https://numpy.org/doc/stable/reference/generated/numpy.allclose.html) which uses 64 bit numbers as default.

[NamanKothari5]

Hi, I m Naman Kothari a sophomore in Information Technology. I wish to work on this issue can you help me out in getting started

[ClaudiaComito]

This is still open.

[ClaudiaComito]

We won't address this issue. The background is this: atol and rtol would have to be set manually for many operations anyway, the numpy defaults aren't particularly related to 64-bit precision (otherwise they would be much lower), and the numpy defaults are identical to the torch defaults.

ht.allclose will stick to the numpy defaults, deviations will have to be addressed in single test functions like we have been doing so far.

Thanks @Sai-Suraj-27 for looking into this.