mojolearn.numeric_profiles

mojolearn.numeric_profiles()

Every registered numeric profile, the baseline first.

Returns:

tuple of dict – One record per profile with its name; default and training_default flags; inference and training availability; products (the arithmetic of each product family); computes (the model families it applies to, None for all); default_for; quality (measured relative change in held-out perplexity against fp32_v1); a note; and measured speed rows per vendor (see mojolearn.numeric_profile_measured()).

Examples

>>> import mojolearn
>>> [p["name"] for p in mojolearn.numeric_profiles()]
['fp32_v1', 'fixed15_v1']