Skip to content

Fix Bug in Flax-Speech-Encoder-Decoder Test #16041

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
Mar 10, 2022
Merged

Fix Bug in Flax-Speech-Encoder-Decoder Test #16041

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
fc292cb
Fix Bug in Flax-Speech-Encoder-Decoder Test
Mar 10, 2022
9d04137
change thresholds for CPU precision
Mar 10, 2022
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
12 changes: 4 additions & 8 deletions tests/speech_encoder_decoder/test_modeling_flax_speech_encoder_decoder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -303,14 +303,12 @@ def compute_loss(
inputs,
attention_mask,
decoder_input_ids,
decoder_attention_mask,
freeze_feature_encoder: bool = False,
):
outputs_enc_dec = enc_dec_model(
inputs=inputs,
attention_mask=attention_mask,
decoder_input_ids=decoder_input_ids,
decoder_attention_mask=decoder_attention_mask,
freeze_feature_encoder=freeze_feature_encoder,
params=params,
)
Expand All @@ -323,13 +321,11 @@ def compute_loss(
grad_fn = jax.value_and_grad(compute_loss)

# compute the loss and gradients for the unfrozen model
loss, grads = grad_fn(
params, inputs, attention_mask, decoder_input_ids, decoder_attention_mask, freeze_feature_encoder=False
)
loss, grads = grad_fn(params, inputs, attention_mask, decoder_input_ids, freeze_feature_encoder=False)

# compare to the loss and gradients for the frozen model
loss_frozen, grads_frozen = grad_fn(
params, inputs, attention_mask, decoder_input_ids, decoder_attention_mask, freeze_feature_encoder=True
params, inputs, attention_mask, decoder_input_ids, freeze_feature_encoder=True
)

self.assert_almost_equals(loss, loss_frozen, 1e-5)
Expand All @@ -348,14 +344,14 @@ def compute_loss(
feature_extractor_grads, feature_extractor_grads_frozen
):
self.assertTrue((feature_extractor_grad_frozen == 0.0).all())
self.assert_difference(feature_extractor_grad, feature_extractor_grad_frozen, 1e-8)
self.assert_difference(feature_extractor_grad, feature_extractor_grad_frozen, 1e-10)
Copy link
Contributor

@patrickvonplaten patrickvonplaten Mar 10, 2022

Choose a reason for hiding this comment

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

No need to make it that aggressive btw :-) We usually are happy with 1e-4

Copy link
Contributor Author

@sanchit-gandhi sanchit-gandhi Mar 10, 2022

Choose a reason for hiding this comment

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

Sure! I just set it in-tandem with the assert_almost_equals threshold that is run for the unfrozen gradients.


# ensure that the gradients of all unfrozen layers remain equal, i.e. all layers excluding the frozen 'feature_extractor'
grads = tuple(grads[k] for k in grads if "feature_extractor" not in k)
grads_frozen = tuple(grads_frozen[k] for k in grads_frozen if "feature_extractor" not in k)

for grad, grad_frozen in zip(grads, grads_frozen):
self.assert_almost_equals(grad, grad_frozen, 1e-8)
self.assert_almost_equals(grad, grad_frozen, 1e-10)

def check_pt_flax_equivalence(self, pt_model, fx_model, inputs_dict):

Expand Down