sinat_24395003的博客

from torch import nn from torch.nn import Identity from typing import Callable,Optional import torch PretrainedConfig = None get_activation = None class SequenceSummary(nn.Module): r""" Compute a single vector summary of a sequence hidden states. .

from torch import nn import torch class Conv1D(nn.Module): """ 1D-convolutional layer as defined by Radford et al. for OpenAI GPT (and also used in GPT-2). Basically works like a linear layer but the weights are transposed. Args: .

""" transformer的generation_beam_search.py中beam_search过程 当decoder的输入是[N,1],N为batch_size,设置beams=k,将输入转化为[N*k,1] 输入放入到decoder中生成了logits,形状为[N*k,T]，T为总的token数 logits和历史beam_score相加成为新的beam_score,进行topk排序，获取next_beam_scores、next_beam_index、next_beam_tokens be.

#transformer.generation_logits_process NoRepeatNGramLogitsProcessor的_calc_banned_ngram_tokens目的是生成不重复的ngram import torch from typing import List, Iterable def _get_ngrams(ngram_size: int, prev_input_ids: torch.Tensor, num_hypos: int): generated_ngra..