 lchu
|
c453b668fa
add doc example about using low_cpu_fsdp
|
hai 1 ano |
 Geeta Chauhan
|
dd57dc5cf9
Update spellcheck.sh (#88)
|
hai 1 ano |
|
lchu
|
80a4c36707
further fix #90
|
hai 1 ano |
|
lchu
|
0c51b47262
fix #90
|
hai 1 ano |
|
lchu
|
c19c5c69aa
fix fsdp construction on low_cpu_fsdp
|
hai 1 ano |
|
lchu
|
e216c6f1f3
address #87
|
hai 1 ano |
 sekyonda
|
59b59afe88
Update spellcheck.sh
|
hai 1 ano |
|
lchu
|
895dfcea30
add nightly check for using low_cpu_fsdp mode
|
hai 1 ano |
 Hamid Shojanazeri
|
9b0eae4056
add max_pad length as an arg
|
hai 1 ano |
|
Hamid Shojanazeri
|
6beadcd8dd
adding padding to inference script with embedding size changing
|
hai 1 ano |
|
lchu
|
1e64fc98d9
switch to simpler param_init_fn and meta device init
|
hai 1 ano |
|
lchu
|
101391f46a
Revert "replace init_empty_weights with torch.device(meta)"
|
hai 1 ano |
|
lchu
|
c8d4f38d23
replace init_empty_weights with torch.device(meta)
|
hai 1 ano |
|
lchu
|
d8a81bb531
save cpu mem by leveraging FSDP rank0 broadcasting
|
hai 1 ano |
|
Geeta Chauhan
|
1387b76e11
fixing the full state path in checkpoint handler+loss report calculation (#51)
|
hai 1 ano |
|
Hamid Shojanazeri
|
88d3e1febc
fix the save_train_param condition
|
hai 1 ano |
|
Hamid Shojanazeri
|
b56028c98d
fixing the word list/spell check
|
hai 1 ano |
|
Hamid Shojanazeri
|
62be60355a
resolving conflicts
|
hai 1 ano |
|
Geeta Chauhan
|
174b856591
update README: python 3.9 rec + fix formatting (#63)
|
hai 1 ano |
|
Geeta Chauhan
|
0cd5694a14
Fsdp inference checkpoints (#39)
|
hai 1 ano |
|
Hamid Shojanazeri
|
c4e96af6ee
clean up
|
hai 1 ano |
 Christian Miller
|
7c1884c690
recommend python 3.9
|
hai 1 ano |
|
Hamid Shojanazeri
|
7d2e06821e
fixing the path to script
|
hai 1 ano |
|
Hamid Shojanazeri
|
5f97db8f0c
fix spell check word list
|
hai 1 ano |
|
Hamid Shojanazeri
|
017cadd04b
Merge branch 'checkpoint_handler_path_fix' of https://github.com/facebookresearch/llama-recipes into checkpoint_handler_path_fix
|
hai 1 ano |
|
Hamid Shojanazeri
|
4f70348b94
remove the redundant lr step
|
hai 1 ano |
|
Hamid Shojanazeri
|
9c95ed4bbe
clean up
|
hai 1 ano |
|
Hamid Shojanazeri
|
311a5c1eec
add notes for train_param.yaml
|
hai 1 ano |
|
Hamid Shojanazeri
|
5b916114eb
merge main branch
|
hai 1 ano |
|
Hamid Shojanazeri
|
668c364f6b
add rank to save_train_params
|
hai 1 ano |
