Closed Faisal-Alsrheed closed 11 months ago
Patch coverage: 100.00%
and project coverage change: +0.06%
:tada:
Comparison is base (
ac8a204
) 76.45% compared to head (a7b92b4
) 76.51%.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Bug Fixe:
Corrected the
dtype_size
function by fixing the comparison logic. Previously, if thedtype size
of the current tensor was greater than thehighest_float
, thehighest_float
value was not being updated correctly. This has been addressed by explicitly comparing the dtype sizes.Test Coverage:
A new file
dtype_utils_test.py
has been added, ensuring rigorous testing of the utility functions.The tests cover:
is_float
function's accuracy in recognizing float dtypes.cast_to_common_dtype
function, ensuring tensors are casted to the most precise commondtype
, especially covering edge cases involvingfloat16
,bfloat16
, and other mixed dtypes.