riaz.somc/transformers
1
0
Fork 0
mirror of https://github.com/huggingface/transformers.git synced 2025-07-04 21:30:07 +06:00
Code Issues Actions 25 Packages Projects Releases Wiki Activity

clean up apex integration

Browse Source
This commit is contained in:
thomwolf 2018-12-13 13:02:17 +01:00
parent 4946c2c500
commit 52c53f39d0
3 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
pytorch_pretrained_bert/convert_tf_checkpoint_to_pytorch.py
View File

@ -59,9 +59,9 @@ def convert_tf_checkpoint_to_pytorch(tf_checkpoint_path, bert_config_file, pytor
l = re.split(r'_(\d+)', m_name) l = re.split(r'_(\d+)', m_name)
else: else:
l = [m_name] l = [m_name]
if l[0] == 'kernel': if l[0] == 'kernel' or l[0] == 'gamma':
pointer = getattr(pointer, 'weight') pointer = getattr(pointer, 'weight')
elif l[0] == 'output_bias': elif l[0] == 'output_bias' or l[0] == 'beta':
pointer = getattr(pointer, 'bias') pointer = getattr(pointer, 'bias')
elif l[0] == 'output_weights': elif l[0] == 'output_weights':
pointer = getattr(pointer, 'weight') pointer = getattr(pointer, 'weight')

2
tests/optimization_test.py
View File

@ -35,7 +35,7 @@ class OptimizationTest(unittest.TestCase):
criterion = torch.nn.MSELoss() criterion = torch.nn.MSELoss()
# No warmup, constant schedule, no gradient clipping # No warmup, constant schedule, no gradient clipping
optimizer = BertAdam(params=[w], lr=2e-1, optimizer = BertAdam(params=[w], lr=2e-1,
weight_decay=0.0, weight_decay_rate=0.0,
max_grad_norm=-1) max_grad_norm=-1)
for _ in range(100): for _ in range(100):
loss = criterion(w, target) loss = criterion(w, target)