saymrwulf/transformers
1
0
Fork 0
mirror of https://github.com/saymrwulf/transformers.git synced 2026-05-14 20:58:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[FlaxSpeechEncoderDecoderModel] Ensure Input and Output Word Embeddings Are **Not** Tied (#16444)

Browse source 
* [FlaxSpeechEncoderDecoderModel] Ensure Input and Output Word Embeddings Are **Not** Tied

* rebase
This commit is contained in:
Sanchit Gandhi 2022-03-28 14:14:10 +02:00 committed by GitHub
parent e0ac72b7bd
commit 7ca4633555
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 7 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/transformers/models/speech_encoder_decoder/modeling_flax_speech_encoder_decoder.py
View file

@ -347,6 +347,8 @@ class FlaxSpeechEncoderDecoderModel(FlaxPreTrainedModel):
f"and {config.encoder.hidden_size} for `config.encoder.hidden_size`."
)
# make sure input & output embeddings are not tied
config.tie_word_embeddings = False
module = self.module_class(config=config, dtype=dtype, **kwargs)
if input_shape is None:
@ -890,6 +892,9 @@ class FlaxSpeechEncoderDecoderModel(FlaxPreTrainedModel):
dtype = kwargs.pop("dtype", jnp.float32)
config = SpeechEncoderDecoderConfig.from_encoder_decoder_configs(encoder.config, decoder.config, **kwargs)
# make sure input & output word embeddings are not tied
config.tie_word_embeddings = False
# init model
model = cls(config, dtype=dtype)
model.params["encoder"] = encoder.params

1
tests/speech_encoder_decoder/test_modeling_flax_speech_encoder_decoder.py
View file

@ -79,6 +79,7 @@ class FlaxEncoderDecoderMixin:
enc_dec_model = FlaxSpeechEncoderDecoderModel(encoder_decoder_config)
self.assertTrue(enc_dec_model.config.is_encoder_decoder)
self.assertFalse(enc_dec_model.config.tie_word_embeddings)
outputs_encoder_decoder = enc_dec_model(
inputs=inputs,

1
tests/speech_encoder_decoder/test_modeling_speech_encoder_decoder.py
View file

@ -72,6 +72,7 @@ class EncoderDecoderMixin:
enc_dec_model.eval()
self.assertTrue(enc_dec_model.config.is_encoder_decoder)
self.assertFalse(enc_dec_model.config.tie_word_embeddings)
outputs_encoder_decoder = enc_dec_model(
input_values=input_values,