PyTorch torch.split_with_sizes Function
Pytorch torch Reference Manual
torch.split_with_sizesIt is a function in PyTorch used to split a tensor according to specified sizes.
Function Definition
torch.split_with_sizes(input, split_sizes, dim=0)
Usage Example
Example
import torch
x = torch.arange(10)
print("Original tensor:")
print(x)
# Split by sizes [2, 3, 5]
result = torch.split_with_sizes(x, [2, 3, 5])
print("Split results:")
for i, t in enumerate(result):
print(f" Chunk {i}: {t}")
# Split a 2D tensor by rows
y = torch.arange(12).reshape(4, 3)
print("nOriginal 2D tensor:")
print(y)
result = torch.split_with_sizes(y, [1, 2, 1], dim=0)
print("Split by rows [1, 2, 1]:")
for i, t in enumerate(result):
print(f" Chunk {i}:n{t}")
x = torch.arange(10)
print("Original tensor:")
print(x)
# Split by sizes [2, 3, 5]
result = torch.split_with_sizes(x, [2, 3, 5])
print("Split results:")
for i, t in enumerate(result):
print(f" Chunk {i}: {t}")
# Split a 2D tensor by rows
y = torch.arange(12).reshape(4, 3)
print("nOriginal 2D tensor:")
print(y)
result = torch.split_with_sizes(y, [1, 2, 1], dim=0)
print("Split by rows [1, 2, 1]:")
for i, t in enumerate(result):
print(f" Chunk {i}:n{t}")
The output is:
原始张量:
tensor([0, 1, 2, 3, 4, 5, 6, 7, 8, 9])
分割结果:
块 0: tensor([0, 1])
Chunk 1: tensor([2, 3, 4])
Chunk 2: tensor([5, 6, 7, 8, 9])
原始二维张量:
tensor([[ 0, 1, 2],
[ 3, 4, 5],
[ 6, 7, 8],
[ 9, 10, 11]])
按行分割 [1, 2, 1]:
块 0:
tensor([[0, 1, 2]])
Chunk 1:
tensor([[3, 4, 5],
[6, 7, 8]])
Chunk 2:
tensor([[ 9, 10, 11]])
Other Extensions