Why outputs are replicated after model:forward ondifferent tensors ?

[sri phani krishna karri]

In corresponding code

require 'nn'

torch.manualSeed(27)

model = nn.Sequential()

model:add(nn.SpatialConvolution(1,1,5,5))

gg = model:forward(torch.rand(1,1,16,16))

print(gg)

gg2 = model:forward(torch.ones(1,1,16,16))

print(gg2)

print(gg2-gg)

gg and gg2 are expected to take different values but after 2nd model forward ie gg2 the gg values are also being changed.

Any help?

[Vislab]

You are overriding the buffers in this case. Just do gg = model:forward(torch.rand(1,1,16,16)):clone() and gg2 = model:forward(torch.ones(1,1,16,16)):clone() and you'll get different values.

[sri phani krishna karri]

Thanks, it worked out