Commit 1668312

committed

Ver 0.4

1 parent ff8cee2 commit 1668312Copy full SHA for 1668312

File tree

-4

lines changed

-4

lines changed

Lines changed: 5 additions & 1 deletion

Original file line number	Diff line number	Diff line change
`@@ -42,4 +42,8 @@ import torchbnn`
`42`	`42`
`43`	`43`	`### Version 0.3`
`44`	`44`	`* bayesian_kl_loss/BKLLoss returns tensor.Tensor([0]) as default`
`45`		`- * In the previous version, bayesian_kl_loss/BKLLoss returns 0 of int type if there is no Bayesian layers. However, considering all torch loss returns tensor and .item() is used to make them to int type, they are changed to return tensor.Tensor([0]) if there is no Bayesian layers.`
	`45`	`+ * In the previous version, bayesian_kl_loss/BKLLoss returns 0 of int type if there is no Bayesian layers. However, considering all torch loss returns tensor and .item() is used to make them to int type, they are changed to return tensor.Tensor([0]) if there is no Bayesian layers.`
	`46`	`+`
	`47`	`+### Version 0.4`
	`48`	`+* bayesian_kl_loss/BKLLoss is modified`
	`49`	`+ * In some cases, the device problem(cuda/cpu) has occurred. Thus, losses are initialized with tensor.Tensor([0]) on the device on which the model is.`

Lines changed: 3 additions & 3 deletions

Original file line number	Diff line number	Diff line change
`@@ -32,9 +32,9 @@ def bayesian_kl_loss(model, reduction='mean', last_layer_only=False) :`
`32`	`32`	`last_layer_only (Bool): True for return only the last layer's KL divergence.`
`33`	`33`
`34`	`34`	`"""`
`35`		`-`
`36`		`- kl = torch.Tensor([0])`
`37`		`- kl_sum = torch.Tensor([0])`
	`35`	`+ device=torch.device("cuda"ifnext(model.parameters()).is_cudaelse"cpu")`
	`36`	`+ kl = torch.Tensor([0]).to(device)`
	`37`	`+ kl_sum = torch.Tensor([0]).to(device)`
`38`	`38`	`n = 0`
`39`	`39`
`40`	`40`	`for m in model.modules() :`

Comments

(0)