DeepSpeed
Transformer/fix layer norm
#1404
Open

Transformer/fix layer norm #1404

RezaYazdaniAminabadi wants to merge 5 commits into master from transformer/fix-layer-norm
RezaYazdaniAminabadi
fix the workspace allocation for the transformer kernel
4017a7c9
Modify normalize kernels to save variance in floating-point for backward
f26900ec
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from awan-10 awan-10 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from cli99 cli99 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from conglongli conglongli 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from eltonzheng eltonzheng 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from jeffra jeffra 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from minjiaz minjiaz 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from niumanar niumanar 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from samyam samyam 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from ShadenSmith ShadenSmith 4 years ago
RezaYazdaniAminabadi RezaYazdaniAminabadi requested a review from tjruwase tjruwase 4 years ago
remove unused grad keys
5eae60f0
jeffra Merge branch 'master' into transformer/fix-layer-norm
af34cc47
hyunwoongko
hyunwoongko commented on 2021-10-01
fix some issue with saving the variance for normalize forward kernel
135ffe30
rocm-mici
FarzanT
jeffra jeffra requested a review from mrwyattii mrwyattii 2 years ago
jeffra jeffra requested a review from cmikeh2 cmikeh2 2 years ago
jeffra jeffra requested a review from arashb arashb 2 years ago
molly-smith molly-smith assigned molly-smith molly-smith 2 years ago
molly-smith molly-smith assigned RezaYazdaniAminabadi RezaYazdaniAminabadi 2 years ago
molly-smith molly-smith unassigned molly-smith molly-smith 2 years ago
loadams loadams closed this 2 years ago
loadams loadams reopened this 2 years ago

Login to write a write a comment.

Login via GitHub