 Hamid Shojanazeri
|
f6152893d8
update prompts
|
1 year ago |
|
Hamid Shojanazeri
|
75f291fe1c
resolved conflicts
|
1 year ago |
 lchu
|
41ffbcab52
code cleanup to remove all unused imports
|
1 year ago |
|
lchu
|
1cc9df19e6
remove unused import
|
1 year ago |
 Geeta Chauhan
|
597e63429c
fix some typos. (#93)
|
1 year ago |
 Chengyu Ma
|
e741dde675
fix some typos
|
1 year ago |
|
Hamid Shojanazeri
|
bf4e0a134b
Add BT after peft module load
|
1 year ago |
|
Hamid Shojanazeri
|
b2a55022cb
clean up
|
1 year ago |
|
Hamid Shojanazeri
|
0941a9ecf7
clean up
|
1 year ago |
|
Geeta Chauhan
|
1803c35f3c
Fix filename typo (#94)
|
1 year ago |
|
Hamid Shojanazeri
|
8601ae14ea
extend word list
|
1 year ago |
|
Hamid Shojanazeri
|
44ef280d31
adding flash attention and xformer memory efficient through PT SDPA
|
1 year ago |
 Philipp Parzer
|
e6329a1fac
Fix filename typo
|
1 year ago |
 luoyifan
|
79d0d4fc4e
fix some typos.
|
1 year ago |
|
lchu
|
c453b668fa
add doc example about using low_cpu_fsdp
|
1 year ago |
|
Geeta Chauhan
|
dd57dc5cf9
Update spellcheck.sh (#88)
|
1 year ago |
|
lchu
|
80a4c36707
further fix #90
|
1 year ago |
|
lchu
|
0c51b47262
fix #90
|
1 year ago |
|
lchu
|
c19c5c69aa
fix fsdp construction on low_cpu_fsdp
|
1 year ago |
|
lchu
|
e216c6f1f3
address #87
|
1 year ago |
 sekyonda
|
59b59afe88
Update spellcheck.sh
|
1 year ago |
|
lchu
|
895dfcea30
add nightly check for using low_cpu_fsdp mode
|
1 year ago |
|
Hamid Shojanazeri
|
9b0eae4056
add max_pad length as an arg
|
1 year ago |
|
Hamid Shojanazeri
|
6beadcd8dd
adding padding to inference script with embedding size changing
|
1 year ago |
|
lchu
|
1e64fc98d9
switch to simpler param_init_fn and meta device init
|
1 year ago |
|
lchu
|
101391f46a
Revert "replace init_empty_weights with torch.device(meta)"
|
1 year ago |
|
lchu
|
c8d4f38d23
replace init_empty_weights with torch.device(meta)
|
1 year ago |
|
lchu
|
d8a81bb531
save cpu mem by leveraging FSDP rank0 broadcasting
|
1 year ago |
|
Geeta Chauhan
|
1387b76e11
fixing the full state path in checkpoint handler+loss report calculation (#51)
|
1 year ago |
|
Hamid Shojanazeri
|
88d3e1febc
fix the save_train_param condition
|
1 year ago |
