Skip to content

Update modernbert-base-pretrain.yaml with a few comments #213

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

Open
wants to merge 1 commit into
base: pretraining_documentation
Choose a base branch
from
Open

Update modernbert-base-pretrain.yaml with a few comments #213

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
9 changes: 6 additions & 3 deletions yamls/modernbert/modernbert-base-pretrain.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,7 +12,7 @@ run_name: modernbert-base-pretrain
# Model
model:
name: flex_bert
pretrained_model_name: bert-base-uncased # has to be set to bert-base-uncased legacy reasons
pretrained_model_name: bert-base-uncased # has to be set to bert-base-uncased to inherit a few configs
tokenizer_name: ${tokenizer_name}
disable_train_metrics: true # save some time by not computing metrics on the training set
model_config:
Expand Down Expand Up @@ -73,7 +73,7 @@ train_loader:
streaming: false
drop_last: true
num_workers: 6
sequence_packing: true
sequence_packing: true # only works with non-streaming dataset
batch_size_warmup_min_size: ${device_train_microbatch_size}
batch_size_warmup_tokens: 50_000_000_000tok

Expand Down Expand Up @@ -133,6 +133,8 @@ callbacks:
window_size: 100
lr_monitor: {}
scheduled_gc: {}
dataloader_speed: {}
runtime_estimator: {}
log_grad_norm:
batch_log_interval: 10
packing_efficiency:
Expand All @@ -150,4 +152,5 @@ save_num_checkpoints_to_keep: -1 # Important, this cleans up checkpoints saved
save_folder: checkpoints/{run_name}

# Load from local filesystem or remote object store to
# load_path: null
# load_path: null
# auto_resume: true # uncomment this to load latest one from checkpoint folder