Skip to content
GitLab
Menu
Projects
Groups
Snippets
Loading...
Help
Help
Support
Community forum
Keyboard shortcuts
?
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
Menu
Open sidebar
chenpangpang
transformers
Commits
889d3bfd
Unverified
Commit
889d3bfd
authored
Feb 20, 2020
by
srush
Committed by
GitHub
Feb 20, 2020
Browse files
default arg fix (#2937)
parent
ea8eba35
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
10 additions
and
4 deletions
+10
-4
examples/ner/transformer_base.py
examples/ner/transformer_base.py
+10
-4
No files found.
examples/ner/transformer_base.py
View file @
889d3bfd
...
@@ -248,16 +248,22 @@ def generic_train(model, args):
...
@@ -248,16 +248,22 @@ def generic_train(model, args):
filepath
=
args
.
output_dir
,
prefix
=
"checkpoint"
,
monitor
=
"val_loss"
,
mode
=
"min"
,
save_top_k
=
5
filepath
=
args
.
output_dir
,
prefix
=
"checkpoint"
,
monitor
=
"val_loss"
,
mode
=
"min"
,
save_top_k
=
5
)
)
train
er
=
pl
.
Trainer
(
train
_params
=
dict
(
accumulate_grad_batches
=
args
.
gradient_accumulation_steps
,
accumulate_grad_batches
=
args
.
gradient_accumulation_steps
,
gpus
=
args
.
n_gpu
,
gpus
=
args
.
n_gpu
,
max_epochs
=
args
.
num_train_epochs
,
max_epochs
=
args
.
num_train_epochs
,
use_amp
=
args
.
fp16
,
amp_level
=
args
.
fp16_opt_level
,
distributed_backend
=
"ddp"
,
gradient_clip_val
=
args
.
max_grad_norm
,
gradient_clip_val
=
args
.
max_grad_norm
,
checkpoint_callback
=
checkpoint_callback
,
checkpoint_callback
=
checkpoint_callback
,
)
)
if
args
.
fp16
:
train_params
[
"use_amp"
]
=
args
.
fp16
train_params
[
"amp_level"
]
=
args
.
fp16_opt_level
if
args
.
n_gpu
>
1
:
train_params
[
"distributed_backend"
]
=
"ddp"
trainer
=
pl
.
Trainer
(
**
train_params
)
if
args
.
do_train
:
if
args
.
do_train
:
trainer
.
fit
(
model
)
trainer
.
fit
(
model
)
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
.
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment