Skip to content
This repository has been archived by the owner on Oct 16, 2023. It is now read-only.

Optimize Linear and GEGLU #152

Draft
wants to merge 1 commit into
base: main
Choose a base branch
from
Draft

Optimize Linear and GEGLU #152

wants to merge 1 commit into from

Conversation

mejai1206
Copy link
Contributor

@mejai1206 mejai1206 commented Sep 27, 2023

๐Ÿ™ Describe the pull request

Linear์™€ GEGLU๋ฅผ ์ตœ์ ํ™”ํ•˜์˜€์Šต๋‹ˆ๋‹ค.

๐Ÿ’ฌ Additional context

Batch๊ฐ€ ์žˆ๊ณ , m์ด ์ž‘์€ ๊ฒฝ์šฐ์— ์„ฑ๋Šฅํ–ฅ์ƒ์„ ๋ณด์ž…๋‹ˆ๋‹ค.

Linear (FP16, Batch=16, M=1, K=N)
image
image

GEGLU (FP16, Batch=16, M=1, K=N)
image
image

โœ… Checklist

  • Code follows the project's coding conventions and style.
  • Tests have been added or updated to cover the changes.
  • Documentation has been updated, if necessary.

trident/function/function.py Outdated Show resolved Hide resolved
trident/function/function.py Outdated Show resolved Hide resolved
trident/operation/linear.py Outdated Show resolved Hide resolved
trident/kernel/linear.py Outdated Show resolved Hide resolved
trident/kernel/linear.py Outdated Show resolved Hide resolved
trident/function/function.py Outdated Show resolved Hide resolved
@mejai1206 mejai1206 force-pushed the optimize_geglu branch 2 times, most recently from c6908ec to 19230ac Compare September 27, 2023 02:28
@mejai1206 mejai1206 marked this pull request as draft September 27, 2023 02:38
@mejai1206 mejai1206 force-pushed the optimize_geglu branch 2 times, most recently from 8260887 to 67d4eb0 Compare October 4, 2023 04:53
@mejai1206 mejai1206 marked this pull request as ready for review October 4, 2023 05:32
Copy link
Contributor

@kakao-steve-ai kakao-steve-ai left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@mejai1206 mejai1206 marked this pull request as draft October 5, 2023 07:29
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants