mojolearn.numeric_profile_measured#
- mojolearn.numeric_profile_measured(profile, vendor=None)#
Measured speed of a numeric profile relative to
fp32_v1.- Parameters:
profile (str) – A registered profile name (see
mojolearn.numeric_profiles()).vendor ({“metal”, “cuda”, “hip”} or None, default None) – One vendor, as
mojolearn.vendor()names it; None returns every measured vendor.
- Returns:
dict or None – With
vendor=None, a dict from vendor to its row. With a vendor, that vendor’s row, or None when it was not measured. A row’soveris the range of the profile’s time divided byfp32_v1’s on the same shapes (above 1 means slower);whatsays which operation was timed.- Raises:
ValueError – If
profileis not a registered profile.
Examples
>>> import mojolearn >>> rows = mojolearn.numeric_profile_measured("fixed15_v1")