bmm和matmul的区别

一句话，bmm()进行批量矩阵乘法，输入的张量只能是3维的。matmul()强大的多借助广播机制，1维、2维、3维、4维的张量都行。

当bmm()输入的张量不是3维的时候：RuntimeError: batch1 must be a 3D tensor

matmul()

1维，向量点积

>>> # vector x vector
>>> tensor1 = torch.randn(3)
>>> tensor2 = torch.randn(3)
>>> torch.matmul(tensor1, tensor2).size()
torch.Size([])

矩阵x向量

>>> # matrix x vector    
>>> tensor1 = torch.randn(3, 4)
>>> tensor2 = torch.randn(4)
>>> torch.matmul(tensor1, tensor2).size()    # 相当于这样形状3x4 4x1的矩阵相乘
torch.Size([3])

>>> # vector x matrix
>>> tensor1 = torch.randn(4)
>>> tensor2 = torch.randn(4, 3)
>>> torch.matmul(tensor1, tensor2).size()    # 相当于这样形状1x4 4x3的矩阵相乘
torch.Size([3])

>>> # batched matrix x broadcasted vector
>>> tensor1 = torch.randn(10, 3, 4)
>>> tensor2 = torch.randn(4)
>>> torch.matmul(tensor1, tensor2).size()
torch.Size([10, 3])

>>> # batched matrix x batched matrix
>>> tensor1 = torch.randn(10, 3, 4)
>>> tensor2 = torch.randn(10, 4, 5)
>>> torch.matmul(tensor1, tensor2).size()
torch.Size([10, 3, 5])

>>> # batched matrix x broadcasted matrix
>>> tensor1 = torch.randn(10, 3, 4)
>>> tensor2 = torch.randn(4, 5)
>>> torch.matmul(tensor1, tensor2).size()
torch.Size([10, 3, 5])

维度更高的

t2 = torch.rand((1,16,2,3))
t3 = torch.rand((1,16,3,2)) 
print(torch.matmul(t2, t3).shape)
# torch.Size([1, 16, 2, 2])

t8 = torch.rand((1,16,15,2,3))
t9 = torch.rand((1,16,15,3,2))
print(torch.matmul(t8, t9).shape)
# torch.Size([1, 16, 15, 2, 2])