Python torch.nn 模块，html() 实例源码

def loss(self, x, samples):
        _, proposal_output = self.forward(x)
        batch_size = len(samples)
        log_weights = torch.log(proposal_output + util.epsilon)
        l = 0
        for b in range(batch_size):
            value = Variable(samples[b].value, requires_grad=False) # value is one-hot
            l -= torch.sum(log_weights[b] * value) # Should we average this over dimensions? See http://pytorch.org/docs/nn.html#torch.nn.KLDivLoss
        return l

def forward(self, x):
        x = self.block1(x)
        x = self.block2(x)
        x = self.block3(x)
        return x

########################################################################
# The code does not need to be changed in CPU-mode.
#
# The documentation for DataParallel is
# `here <http://pytorch.org/docs/nn.html#torch.nn.DataParallel>`_.
#
# **Primitives on which DataParallel is implemented upon:**
#
#
# In general, pytorch’s `nn.parallel` primitives can be used independently.
# We have implemented simple MPI-like primitives:
#
# - replicate: replicate a Module on multiple devices
# - scatter: distribute the input in the first-dimension
# - gather: gather and concatenate the input in the first-dimension
# - parallel\_apply: apply a set of already-distributed inputs to a set of
#   already-distributed models.
#
# To give a better clarity, here function ``data_parallel`` composed using
# these collectives

def forward(self, x):
        x = self.block1(x)
        x = self.block2(x)
        x = self.block3(x)
        return x

########################################################################
# The code does not need to be changed in CPU-mode.
#
# The documentation for DataParallel is
# `here <http://pytorch.org/docs/nn.html#torch.nn.DataParallel>`_.
#
# **Primitives on which DataParallel is implemented upon:**
#
#
# In general, pytorch’s `nn.parallel` primitives can be used independently.
# We have implemented simple MPI-like primitives:
#
# - replicate: replicate a Module on multiple devices
# - scatter: distribute the input in the first-dimension
# - gather: gather and concatenate the input in the first-dimension
# - parallel\_apply: apply a set of already-distributed inputs to a set of
#   already-distributed models.
#
# To give a better clarity, here function ``data_parallel`` composed using
# these collectives