bigscience-workshop / bigscience-workshop/Megatron-DeepSpeed

Double counts in parameter count

Open
#99 2 comments 0 reactions 2 assignees Claimed by @TevenLeScao View on GitHub
Dominant language
Python
Stars
1.4k
Forks
226
PR merge metrics
No merged PRs in 30d

Description

Currently, parameter counts in `utils.get_parameters_in_billions` are inaccurate when PP > 1. Tied variables, in particular embedding layers, exist in several copies in the first and last PP stage, which causes double counts. For now the codebase only uses the count without embedding layers which is accurate, but it would be good for the count with embedding layers to also function, mostly for operation-counting.

For background see: https://github.com/bigscience-workshop/Megatron-DeepSpeed/pull/40

Contributor guide

No contributing guide indexed for this repository

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.