llama.cpp
Add attention and final logit soft-capping, update scaling factor to Gemma2
#8197
Merged
Merged
Add attention and final logit soft-capping, update scaling factor to Gemma2 #8197
abetlen
merged
10 commits into
master
from
add-gemma2-soft-capping
Add attention and final logit softcapping.
4d3f17b4
fix
d3d3c4eb
Add custom add_ functions
d1137c20
Disable flash attention for Gemma2
f4424c15
slaren
commented
on 2024-06-28
Update src/llama.cpp
3a247181
slaren
approved these changes
on 2024-06-28
ngxson
commented
on 2024-06-28
Merge branch 'master' of github.com:ggerganov/llama.cpp into add-gemm…
8edf73a7
Add default value for attention and final logit softcap value
bb715992
Merge branch 'add-gemma2-soft-capping' of github.com:ggerganov/llama.…
6f2464e3
ngxson
approved these changes
on 2024-06-29
ggerganov
approved these changes
on 2024-06-29
Add custom kq scaling from Gemma2Attention
a8942790
abetlen
changed the title 
Remove custom pre attention scaling and use computed value instead.
51f0bd50
abetlen
changed the title
slaren
commented
on 2024-06-30
eran-medan
commented
on 2024-07-02
Reviewers
Assignees
No one assigned
Labels
python
Review Complexity : Medium
Milestone
No milestone
Login to write a write a comment.
Login via GitHub