[WIP] Fix cholesky TF32 tests [v2]

[xwang233]

test_linalg.test_cholesky_cuda is failing on A100 machine.

The test_cholesky was probably introduced in #46083.
Also see #45492 for the fix [v1].

[xwang233]

cc @ptrblck

[codecov]

Codecov Report

Merging #48288 (c70c837) into master (c0723a0) will decrease coverage by 0.15%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master   #48288      +/-   ##
==========================================
- Coverage   80.96%   80.81%   -0.16%     
==========================================
  Files        1844     1844              
  Lines      199443   199443              
==========================================
- Hits       161471   161170     -301     
- Misses      37972    38273     +301     

[ngimel]

tf32 settings should not affect linear algebra functions, and there are no matrix multiplies in the test itself, as far as I can see, so that should not be a problem for tolerances either. Can you please take a deeper look into what causes worse tolerance when tf32 is on, and fix that preferably? That is, tf32 setting should not affect cholesky result.

[pytorchbot]

Looks like this PR hasn't been updated in a while so we're going to go ahead and mark this as Stale.
Feel free to remove the Stale label if you feel this was a mistake.
If you are unable to remove the Stale label please contact a maintainer in order to do so.
Stale pull requests will automatically be closed 30 days after being marked Stale