def get_pooled_out(self, summary_type, use_summ_proj=True):
"""
Args:
summary_type: str, "last", "first", "mean", or "attn". The method
to pool the input to get a vector representation.
use_summ_proj: bool, whether to use a linear projection during pooling.
Returns:
float32 Tensor in shape [bsz, d_model], the pooled representation.
"""
xlnet_config = self.xlnet_config
run_config = self.run_config
with tf.variable_scope("model", reuse=tf.AUTO_REUSE):
summary = modeling.summarize_sequence(
summary_type=summary_type,
hidden=self.output,
d_model=xlnet_config.d_model,
n_head=xlnet_config.n_head,
d_head=xlnet_config.d_head,
dropout=run_config.dropout,
dropatt=run_config.dropatt,
is_training=run_config.is_training,
input_mask=self.input_mask,
initializer=self.initializer,
use_proj=use_summ_proj)
return summary
def get_pooled_out(self, summary_type, use_summ_proj=True):
"""
Args:
summary_type: str, "last", "first", "mean", or "attn". The method
to pool the input to get a vector representation.
use_summ_proj: bool, whether to use a linear projection during pooling.
Returns:
float32 Tensor in shape [bsz, d_model], the pooled representation.
"""
summary = modeling.summarize_sequence(
summary_type=summary_type,
hidden=self.output,
d_model=self._d_model,
n_head=self._n_head,
d_head=self._d_head,
dropout=self._dropout,
dropatt=self._dropatt,
input_mask=self.input_mask,
initializer=self._param_initializer,
use_proj=use_summ_proj,
name='model_sequnece_summary')
return summary
