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

delete comments

Browse Source
This commit is contained in:
yaozengwei 2022-06-18 20:07:59 +08:00
parent d7f4920206
commit b2a4c33ddd
4 changed files with 2 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
egs/librispeech/ASR/conv_emformer_transducer_stateless/train.py
View File

@ -1090,9 +1090,6 @@ def scan_pessimistic_batches_for_oom(
for criterion, cuts in batches.items():
batch = train_dl.dataset[cuts]
try:
# warmup = 0.0 is so that the derivs for the pruned loss stay zero
# (i.e. are not remembered by the decaying-average in adam), because
# we want to avoid these params being subject to shrinkage in adam.
with torch.cuda.amp.autocast(enabled=params.use_fp16):
loss, _ = compute_loss(
params=params,

3
egs/librispeech/ASR/pruned_transducer_stateless4/train.py
View File

@ -1004,9 +1004,6 @@ def scan_pessimistic_batches_for_oom(
for criterion, cuts in batches.items():
batch = train_dl.dataset[cuts]
try:
# warmup = 0.0 is so that the derivs for the pruned loss stay zero
# (i.e. are not remembered by the decaying-average in adam), because
# we want to avoid these params being subject to shrinkage in adam.
with torch.cuda.amp.autocast(enabled=params.use_fp16):
loss, _ = compute_loss(
params=params,

3
egs/librispeech/ASR/pruned_transducer_stateless5/train.py
View File

@ -1084,9 +1084,6 @@ def scan_pessimistic_batches_for_oom(
for criterion, cuts in batches.items():
batch = train_dl.dataset[cuts]
try:
# warmup = 0.0 is so that the derivs for the pruned loss stay zero
# (i.e. are not remembered by the decaying-average in adam), because
# we want to avoid these params being subject to shrinkage in adam.
with torch.cuda.amp.autocast(enabled=params.use_fp16):
loss, _ = compute_loss(
params=params,

7
egs/librispeech/ASR/pruned_transducer_stateless6/train.py
View File

@ -74,9 +74,9 @@ from conformer import Conformer
from decoder import Decoder
from joiner import Joiner
from lhotse.cut import Cut, MonoCut
from lhotse.dataset.collation import collate_custom_field
from lhotse.dataset.sampling.base import CutSampler
from lhotse.utils import fix_random_seed
from lhotse.dataset.collation import collate_custom_field
from model import Transducer
from optim import Eden, Eve
from torch import Tensor
@ -376,7 +376,7 @@ def get_params() -> AttributeDict:
"distillation_layer": 5, # 0-based index
# Since output rate of hubert is 50, while that of encoder is 8,
# two successive codebook_index are concatenated together.
# Detailed in function Transducer::concat_sucessive_codebook_indexes.
# Detailed in function Transducer::concat_sucessive_codebook_indexes
"num_codebooks": 16, # used to construct distillation loss
}
)
@ -1060,9 +1060,6 @@ def scan_pessimistic_batches_for_oom(
for criterion, cuts in batches.items():
batch = train_dl.dataset[cuts]
try:
# warmup = 0.0 is so that the derivs for the pruned loss stay zero
# (i.e. are not remembered by the decaying-average in adam), because
# we want to avoid these params being subject to shrinkage in adam.
with torch.cuda.amp.autocast(enabled=params.use_fp16):
loss, _ = compute_loss(
params=params,