Add new Floating Point Formats for Submission

mlcommons / training_policies

Issues related to MLPerf™ training policies, including rules and suggested changes

https://mlcommons.org/en/groups/training

Apache License 2.0

92 stars 66 forks source link

Closed bitfort closed 4 years ago

bitfort commented 4 years ago

We would like to add the new formats:

tf32, [u]int4, [u]int8

There was no objection made during Special Topics for these additions.

bitfort commented 4 years ago

SWG:

Rules to include tf32, [u]int4, [u]int8