PR #2782 Fix gpt-Neox rotary embedding implementation

Fix gpt-Neox rotary embedding implementation #2782

jeffra merged 16 commits into master from fix-neox-rope

Reset KV-cache at the beginning of text-generation

8e91537e

Merge branch 'master' of github.com:microsoft/DeepSpeed

963cc9a5

Merge branch 'master' of github.com:microsoft/DeepSpeed

28be9bba

Merge branch 'master' of github.com:microsoft/DeepSpeed

ae91f324

fix the seq_id used for RoPE implementation

d48eb478

RezaYazdaniAminabadi requested a review from

jeffra 2 years ago

RezaYazdaniAminabadi requested a review from

mrwyattii 2 years ago

RezaYazdaniAminabadi requested a review from

awan-10 2 years ago

RezaYazdaniAminabadi requested a review from

cmikeh2 2 years ago

RezaYazdaniAminabadi requested a review from

arashb 2 years ago

remove reset-cache

e877e1df

lekurile approved these changes on 2023-02-02

fix formatting

e1c82659

RezaYazdaniAminabadi enabled auto-merge (squash) 2 years ago