Skip to content

Make data shuffling in run_clm_flax.py respect global seed #13410

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

Merged
merged 2 commits into from
Dec 14, 2021
Merged

Make data shuffling in run_clm_flax.py respect global seed #13410

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
14db665
use jax and jnp instead of numpy in data_loader
bminixhofer Sep 3, 2021
3279d45
return batches as np.ndarray
bminixhofer Sep 6, 2021
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
4 changes: 2 additions & 2 deletions examples/flax/language-modeling/run_clm_flax.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -187,9 +187,9 @@ def data_loader(rng: jax.random.PRNGKey, dataset: Dataset, batch_size: int, shuf
steps_per_epoch = len(dataset) // batch_size

if shuffle:
Copy link
Contributor

@patrickvonplaten patrickvonplaten Sep 6, 2021

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@patil-suraj - I think here we could actually switch to jax.random.permutation as this is called only once per epoch. I'm pretty sure that it wouldn't slow down the training

batch_idx = np.random.permutation(len(dataset))
batch_idx = jax.random.permutation(rng, len(dataset))
else:
batch_idx = np.arange(len(dataset))
batch_idx = jnp.arange(len(dataset))

batch_idx = batch_idx[: steps_per_epoch * batch_size] # Skip incomplete batch.
batch_idx = batch_idx.reshape((steps_per_epoch, batch_size))
Expand Down