def create_and_check_causal_lm_decoder(
self,
config,
input_ids,
decoder_input_ids,
attention_mask,
decoder_attention_mask,
lm_labels,
):
model = ProphetNetForCausalLM(config=config).to(torch_device).eval()
outputs = model(
input_ids=decoder_input_ids,
attention_mask=decoder_attention_mask,
labels=lm_labels,
)
self.parent.assertEqual(len(outputs), 4)
self.parent.assertEqual(outputs["logits"].size(), (self.batch_size, self.decoder_seq_length, self.vocab_size))
self.parent.assertEqual(outputs["loss"].size(), ())
def get_encoder_decoder_model(self, config, decoder_config):
encoder_model = BertModel(config)
decoder_model = ProphetNetForCausalLM(decoder_config)
return encoder_model, decoder_model