loss.py File Reference

Namespaces
namespace	loss

Functions
	loss.base_loss_components (pred_wdl, target_wdl, wdl_weight=1.0)
	loss.base_loss (pred_wdl, target_wdl, wdl_weight=1.0)
	loss.killer_loss (pred_wdl, target_wdl, wdl_weight=1.0)
	loss.survivor_loss (pred_wdl, target_wdl, wdl_weight=1.0)
	loss.gater_loss (pred_gates, target_gates)
	loss.moe_loss (pred_wdl, target_wdl, expert_type='BASE')

Variables
int	loss.B = 8
	loss.pred_wdl = F.softmax(torch.randn(B, 3), dim=1)
	loss.target_wdl = F.softmax(torch.randn(B, 3), dim=1)
	loss.loss = moe_loss(pred_wdl, target_wdl, expert_type)
	loss.pred_gates = torch.sigmoid(torch.randn(B, 2))
	loss.target_gates = torch.rand(B, 2)
	loss.g_loss = gater_loss(pred_gates, target_gates)