提交 272fda2b 编写于 作者: Y Yang Nie

update config for 8 cards

上级 e61f3925
......@@ -15,7 +15,7 @@ Global:
save_inference_dir: ./inference
# training model under @to_static
to_static: False
update_freq: 8
update_freq: 8 # for 8 cards
# model ema
EMA:
......@@ -50,6 +50,7 @@ Optimizer:
weight_decay: 0.05
one_dim_param_no_weight_decay: True
lr:
# for 8 cards
name: Cosine
learning_rate: 4e-3 # lr 4e-3 for total_batch_size 4096
eta_min: 1e-6
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册