Add multi-worker support for JAX training. #18654
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## master #18654 +/- ##
==========================================
- Coverage 78.57% 78.51% -0.06%
==========================================
Files 335 335
Lines 32979 33020 +41
Branches 6455 6467 +12
==========================================
+ Hits 25913 25927 +14
- Misses 5510 5532 +22
- Partials 1556 1561 +5
Flags with carried forward coverage won't be shown. Click here to find out more.
☔ View full report in Codecov by Sentry. |
fchollet
reviewed
Oct 20, 2023
| @@ -7,7 +7,7 @@ | |||
| class TFDatasetAdapter(DataAdapter): | |||
| """Adapter that handles `tf.data.Dataset`.""" | |||
|
|
|||
| def __init__(self, dataset, class_weight=None): | |||
| def __init__(self, dataset, class_weight=None, distribution=None): | |||
There was a problem hiding this comment.
Please add a Args: section to document the type of each argument (to avoid confusion with tf.distribute)
| ): | ||
| raise ValueError( | ||
| "Only `tf.data.Dataset` is supported for multi worker " | ||
| f"distribution, received input types is {type(x)}" |
There was a problem hiding this comment.
When using multi-worker distribution, the data must be provided
as a `tf.data.Dataset` instance. Received: type(x)={type(x)}
| inputs: `jax.Array` that is already sharded to a local process size. | ||
| layout: `TensorLayout` for the distribution information, or a | ||
| `jax.sharding.Sharding` instance. | ||
| Returns: |
| ).prefetch(tf.data.AUTOTUNE) | ||
| batch_size = tf_data_distribute.compute_batch_size(dataset) | ||
| if batch_size.numpy() < 0: | ||
| raise ValueError( |
There was a problem hiding this comment.
In what cases does this happen? Unbatched dataset? The error message should make explicit what user action is required (e.g. calling .batch()).
There was a problem hiding this comment.
yea, most likely due to the dataset is not batched. Update the error message.
google-ml-butler
bot
added
kokoro:force-run
ready to pull
google-ml-butler
bot
removed
awaiting review
ready to pull
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I have tried to add some backend agnostic tests to dataset builder, and the JAX specific multi-worker test will be hosted internally.