[JAX] Flax with compute dtype inferred from input dtype. #1485
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
ptrendx
reviewed
Feb 14, 2025
| axes=lora_b_kernel_axes, | ||
| ) | ||
| lora_b_kernel = lora_b_kernel.astype(self.dtype) | ||
| if not FP8Helper.is_fp8_enabled(): | ||
| lora_b_kernel = lora_b_kernel.astype(input_dtype) |
There was a problem hiding this comment.
are Lora weights also applied in FP8? It's possible they are not and then this cast would be required.
ptrendx
reviewed
Feb 14, 2025
ptrendx
reviewed
Feb 14, 2025
ptrendx
reviewed
Feb 14, 2025
ptrendx
reviewed
Feb 14, 2025
| @@ -265,8 +265,8 @@ def test_forward( | |||
| """Test only the forward""" | |||
| inputs, (ref_masks, test_masks) = self.generate_inputs(data_shape, dtype) | |||
|
|
|||
| ref_layer_cls = partial(self.reference_layer, dtype=dtype, **self.attrs) | |||
| layer_cls = partial(TransformerLayer, layer_type=self.layer_type, dtype=dtype, **self.attrs) | |||
| ref_layer_cls = partial(self.reference_layer, **self.attrs) | |||
There was a problem hiding this comment.
Is reference_layer following the correct logic?
There was a problem hiding this comment.
Updated reference layer in utils.py
|
/te-ci jax L1 |
3 similar comments
|
/te-ci jax L1 |
|
/te-ci jax L1 |
|
/te-ci jax L1 |
Signed-off-by: Phuong Nguyen <[email protected]>
phu0ngng
force-pushed
the
flax_with_input_dtype
branch
from
77f2edb to
d41d6f8
Compare
ptrendx
approved these changes
Feb 18, 2025
ptrendx
pushed a commit
that referenced
this pull request
Feb 19, 2025
flax module with compute dtype inferred from the inputs Signed-off-by: Phuong Nguyen <[email protected]>
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Outdated suggestions cannot be applied.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.
Description
Flax modules should use
dtypeto initialize their parameters while the compute type should depend on the input data type.This PR adds the capability to infer the compute type from the input type.
Type of change
Checklist: