Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

merge from magma #20

Merged
merged 10 commits into from
Jun 13, 2023
Merged

merge from magma #20

Show file tree
Hide file tree
Changes from 5 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
299f546
Add support to HFTokenzer
May 31, 2023
f1810aa
Script to convert GPT Neox Checkpoint for magma
May 31, 2023
eb4edda
Merge pull request #10 from floatingsnake/neox_checkpoint_loading
floatingbigcat May 31, 2023
bef649e
add 70mOpenclipH model config for summit
Jun 2, 2023
acbc2b2
rm eod token for hfgpt2tokenizer
Jun 7, 2023
cb1e342
add mask for pad token
Jun 7, 2023
48ceea8
Merge pull request #14 from floatingsnake/fix-maskpad
kshitijkg Jun 7, 2023
29697ad
add pad token
Jun 9, 2023
5795012
Merge branch 'magma' into fix-tokenizer
floatingbigcat Jun 10, 2023
922a26e
Merge pull request #9 from floatingsnake/fix-tokenizer
kshitijkg Jun 12, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion configs/summit-70m-openclipH.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -96,5 +96,5 @@
"wall_clock_breakdown": true,

"tokenizer-type": "HFTokenizer"
# "tokenizer-type": "HFGPT2Tokenizer"

}
2 changes: 1 addition & 1 deletion configs/summit_setup.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,9 +4,9 @@
"valid-data-paths": "/gpfs/alpine/csc499/proj-shared/LAION-400m-webdataset/data/{40001..41000}.tar",
"test-data-paths": "/gpfs/alpine/csc499/proj-shared/LAION-400m-webdataset/data/{41000..41455}.tar",

# we use tokenzier from huggingface, don't need vocal or merge file

"vocab-file": "./data/20B_tokenizer.json",


"save": "/gpfs/alpine/scratch/lfsm/csc499/checkpoints",
"load": "/gpfs/alpine/scratch/lfsm/csc499/checkpoints",
Expand Down
2 changes: 2 additions & 0 deletions megatron/training.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -284,6 +284,7 @@ def _get_batch(neox_args, tokenizer, keys, data, datatype):
# Get the masks and position ids.
attention_mask, loss_mask, position_ids = get_ltor_masks_and_position_ids(
data=tokens,
pad_token=neox_args.tokenizer.pad_id,
eod_token=neox_args.tokenizer.eod,
eod_mask_loss=neox_args.eod_mask_loss,
)
Expand Down Expand Up @@ -360,6 +361,7 @@ def get_batch_pipe_image_text(input,neox_args):

attention_mask, loss_mask, position_ids = get_ltor_masks_and_position_ids(
data=captions,
pad_token=neox_args.tokenizer.pad_id,
eod_token=neox_args.tokenizer.eod,
eod_mask_loss=neox_args.eod_mask_loss,
)
Expand Down
2 changes: 2 additions & 0 deletions megatron/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -78,6 +78,7 @@ def get_attn_mask(seq_length, device):

def get_ltor_masks_and_position_ids(
data,
pad_token,
eod_token,
eod_mask_loss=False,
):
Expand All @@ -94,6 +95,7 @@ def get_ltor_masks_and_position_ids(

# Loss mask.
loss_mask = torch.ones(data.size(), dtype=torch.float, device=data.device)
loss_mask[data == pad_token] = 0.0
if eod_mask_loss:
loss_mask[data == eod_token] = 0.0

Expand Down