colossalai.nn.loss.loss_3d

class colossalai.nn.loss.loss_3d.CrossEntropyLoss3D(reduction=True, *args, **kwargs)

Cross entropy loss for 3D parallelism

Parameters
  • reduction (bool, optional) – whether to average the loss, defaults to True

  • args – Args for loss function

  • kwargs – Kwargs for loss function

forward(logits, targets)

Calculate loss between logits and targets

Parameters
  • logits – Output logits of model

  • targets – True targets from data

class colossalai.nn.loss.loss_3d.VocabParallelCrossEntropyLoss3D(reduction=True)

Vocab parallel cross entropy loss for 2D parallelism

Parameters

reduction (bool, optional) – whether to average the loss, defaults to True

forward(logits, targets)

Calculate loss between logits and targets

Parameters
  • logits – Output logits of model

  • targets – True targets from data