mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-09-19 05:54:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

Improve speed by messing with configuration, removing asserts.

Browse Source
This commit is contained in:
Daniel Povey 2022-06-13 17:27:34 +08:00
parent a41c4b6c9b
commit d1e96afce2
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
egs/librispeech/ASR/pruned_transducer_stateless7/optim.py
View File

@ -265,8 +265,8 @@ class NeutralGradient(Optimizer):
scalar_exp_avg_sq = state["scalar_exp_avg_sq"] scalar_exp_avg_sq = state["scalar_exp_avg_sq"]
grad_sq = (grad**2).mean() grad_sq = (grad**2).mean()
conditioned_grad_sq = (conditioned_grad**2).mean() conditioned_grad_sq = (conditioned_grad**2).mean()
assert grad_sq - grad_sq == 0 #assert grad_sq - grad_sq == 0
assert conditioned_grad_sq - conditioned_grad_sq == 0 #assert conditioned_grad_sq - conditioned_grad_sq == 0
scalar_exp_avg_sq.mul_(beta2).add_(grad_sq, alpha=(1-beta2)) scalar_exp_avg_sq.mul_(beta2).add_(grad_sq, alpha=(1-beta2))
bias_correction2 = 1 - beta2 ** (step + 1) bias_correction2 = 1 - beta2 ** (step + 1)
avg_grad_sq = scalar_exp_avg_sq / bias_correction2 avg_grad_sq = scalar_exp_avg_sq / bias_correction2
@ -285,7 +285,7 @@ class NeutralGradient(Optimizer):
delta.add_(this_delta, alpha=alpha) delta.add_(this_delta, alpha=alpha)
if random.random() < 0.005: if random.random() < 0.005:
print(f"Delta rms = {(delta**2).mean().item()}, shape = {delta.shape}") print(f"Delta rms = {(delta**2).mean().item()}, shape = {delta.shape}")
assert delta.abs().max() < 10.0 #assert delta.abs().max() < 10.0
p.add_(delta) p.add_(delta)
if step % 10 == 0: if step % 10 == 0:
p.clamp_(min=-param_max, max=param_max) p.clamp_(min=-param_max, max=param_max)