modeling

class ErnieForGeneration(cfg, name=None)

Bases: paddlenlp.transformers.ernie_gen.modeling.ErnieModel

Ernie Model for sequence to sequence generation.

forward(*args, **kwargs)

Args

tgt_labels(Variable of shape [batch_size, seqlen] or [batch, seqlen, vocab_size]):

ground trouth target sequence id (hard label) or distribution (soft label)

tgt_pos(Variable of shape [n_targets, 2]):

index of tgt_labels in src_ids, can be obtained from fluid.layers.where(src_ids==mask_id)

encoder_only(Bool):

if set, will not return loss, logits_2d

Returns

cross entropy loss mean over every target label. if encode_only, returns None. logits(Variable of shape [n_targets, vocab_size]):

logits for every targets. if encode_only, returns None.

info(Dictionary): see ErnieModel

Return type

loss(Variable of shape [])