Differences with torch.nn.GRU

torch.nn.GRU

class torch.nn.GRU(
    input_size,
    hidden_size,
    num_layers=1,
    bias=True,
    batch_first=False,
    dropout=0,
    bidirectional=False)(input, h_0) -> Tensor

For more information, see torch.nn.GRU.

mindspore.nn.GRU

class mindspore.nn.GRU(
    input_size,
    hidden_size,
    num_layers=1,
    has_bias=True,
    batch_first=False,
    dropout=0.0,
    bidirectional=False)(x, hx, seq_length) -> Tensor

For more information, see mindspore.nn.GRU.

Differences

PyTorch: Calculate the output sequence and the final state based on the output sequence and the given initial state.

MindSpore: Consistent function. One more interface input seq_length, indicating the length of each sequence in the input batch.

Categories	Subcategories	PyTorch	MindSpore	Difference
Parameters	Parameter 1	input_size	input_size	-
	Parameter 2	hidden_size	hidden_size	-
	Parameter 3	num_layers	num_layers	-
	Parameter 4	bias	has_bias	Same function, different parameter names
	Parameter 5	batch_first	batch_first	-
	Parameter 6	dropout	dropout	-
	Parameter 7	bidirectional	bidirectional	-
Input	Input 1	input	x	Same function, different parameter names
	Input 2	h_0	hx	Same function, different parameter names
	Input 3	-	seq_length	The length of each sequence in input batch

Code Example

The two APIs achieve the same function and have the same usage.

# PyTorch
import torch
import torch.nn as nn
import numpy as np

rnn = nn.GRU(10, 16, 2, batch_first=True)
input = torch.ones([3, 5, 10], dtype=torch.float32)
h0 = torch.ones([1 * 2, 3, 16], dtype=torch.float32)
output, hn = rnn(input, h0)
output = output.detach().numpy()
print(output.shape)
# (3, 5, 16)

# MindSpore
import mindspore
from mindspore import Tensor, nn

net = nn.GRU(10, 16, 2, batch_first=True)
x = Tensor(np.ones([3, 5, 10]), mindspore.float32)
h0 = Tensor(np.ones([1 * 2, 3, 16]), mindspore.float32)
output, hn = net(x, h0)
print(output.shape)
# (3, 5, 16)