mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-12-11 06:55:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

Memory fix in WithLoss

Browse Source
This commit is contained in:
Daniel Povey 2022-12-11 11:15:56 +08:00
parent 0edaf4d25c
commit 2d0fe7637c
1 changed files with 5 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
egs/librispeech/ASR/pruned_transducer_stateless7/scaling.py
View File

@ -911,17 +911,14 @@ class Whiten(nn.Module):
class WithLoss(torch.autograd.Function): class WithLoss(torch.autograd.Function):
@staticmethod @staticmethod
def forward(ctx, x: Tensor, y: Tensor, name: str): def forward(ctx, x: Tensor, y: Tensor, name: str):
ctx.name = name ctx.y_shape = y.shape
ctx.save_for_backward(y) # just for printing the name, and the shape if random.random() < 0.002 and name is not None:
loss_sum = y.sum().item()
logging.info(f"WithLoss: name={ctx.name}, loss-sum={loss_sum:.3e}")
return x return x
@staticmethod @staticmethod
def backward(ctx, ans_grad: Tensor): def backward(ctx, ans_grad: Tensor):
y, = ctx.saved_tensors return ans_grad, torch.ones(ctx.y_shape,
if random.random() < 0.002 and ctx.name is not None:
loss_sum = y.sum().item()
logging.info(f"WithLoss: name={ctx.name}, loss-sum={loss_sum:.3e}")
return ans_grad, torch.ones(y.shape,
dtype=ans_grad.dtype, dtype=ans_grad.dtype,
device=ans_grad.device), None device=ans_grad.device), None
def with_loss(x, y, name): def with_loss(x, y, name):