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

Renaming in optim.py; remove step() from scan_pessimistic_batches_for_oom in train.py

Browse Source
This commit is contained in:
Daniel Povey 2022-10-22 17:44:21 +08:00
parent 525e87a82d
commit 146626bb85
2 changed files with 5 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
egs/librispeech/ASR/pruned_transducer_stateless7/optim.py
View File

@ -562,10 +562,10 @@ class LRScheduler(object):
self.verbose = verbose
for group in optimizer.param_groups:
group.setdefault("initial_lr", group["lr"])
group.setdefault("base_lr", group["lr"])
self.base_lrs = [
group["initial_lr"] for group in optimizer.param_groups
group["base_lr"] for group in optimizer.param_groups
]
self.epoch = 0
@ -647,13 +647,13 @@ class Eden(LRScheduler):
"""
Eden scheduler.
The basic formula (before warmup) is:
lr = initial_lr * (((batch**2 + lr_batches**2) / lr_batches**2) ** -0.25 *
(((epoch**2 + lr_epochs**2) / lr_epochs**2) ** -0.25)) * warmup
lr = base_lr * (((batch**2 + lr_batches**2) / lr_batches**2) ** -0.25 *
(((epoch**2 + lr_epochs**2) / lr_epochs**2) ** -0.25)) * warmup
where `warmup` increases from linearly 0.5 to 1 over `warmup_batches` batches
and then stays constant at 1.
E.g. suggest initial-lr = 0.04 (passed to optimizer) if used with ScaledAdam
E.g. suggest base_lr = 0.04 (passed to optimizer) if used with ScaledAdam
Args:
optimizer: the optimizer to change the learning rates on

1
egs/librispeech/ASR/pruned_transducer_stateless7/train.py
View File

@ -1156,7 +1156,6 @@ def scan_pessimistic_batches_for_oom(
is_training=True,
)
loss.backward()
optimizer.step()
optimizer.zero_grad()
except Exception as e:
if "CUDA out of memory" in str(e):