比较与torch.nn.TransformerDecoderLayer的功能差异

torch.nn.TransformerDecoderLayer(d_model, nhead, dim_feedforward=2048, dropout=0.1, activation=<function relu>, layer_norm_eps=1e-05, batch_first=False, norm_first=False, device=None, dtype=None)

class mindspore.nn.transformer.TransformerDecoderLayer(hidden_size, ffn_hidden_size, num_heads, batch_size, src_seq_length, tgt_seq_length, attention_dropout_rate=0.1, hidden_dropout_rate=0.1, post_layernorm_residual=False, use_past=False, layernorm_compute_type=mstype.float32, softmax_compute_type=mstype.float32, param_init_type=mstype.float32, hidden_act="gelu", moe_config=default_moe_config, parallel_config=default_dpmp_config)(
    hidden_stats, decoder_mask, encoder_output=None,
    memory_mask=None, init_reset=True, batch_valid_length=None
)

import numpy as np
from mindspore import dtype as mstype
from mindspore.nn.transformer import TransformerDecoderLayer
from mindspore import Tensor
model = TransformerDecoderLayer(batch_size=32, hidden_size=512, ffn_hidden_size=2048,
                                num_heads=8, src_seq_length=10, tgt_seq_length=20)
encoder_input_value = Tensor(np.ones((32, 10, 512)), mstype.float32)
decoder_input_value = Tensor(np.ones((32, 20, 512)), mstype.float32)
decoder_input_mask = Tensor(np.ones((32, 20, 20)), mstype.float16)
memory_mask = Tensor(np.ones((32, 20, 10)), mstype.float16)
output, past = model(decoder_input_value, decoder_input_mask, encoder_input_value, memory_mask)
print(output.shape)
# output:
# (32, 20, 512)

import torch
decoder_layer = torch.nn.TransformerDecoderLayer(d_model=512, nhead=8)
memory = torch.rand(10, 32, 512)
tgt = torch.rand(20, 32, 512)
output = decoder_layer(tgt, memory)
print(output.shape)
# output:
# torch.Size([20, 32, 512])