From b2e09f9f58e495cff503359a9e2d13c792e1b62d Mon Sep 17 00:00:00 2001 From: Eren Golge Date: Mon, 14 May 2018 19:04:29 -0700 Subject: [PATCH] larger lr for stop token --- train.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/train.py b/train.py index 791c5d4f..ba712ecc 100644 --- a/train.py +++ b/train.py @@ -88,12 +88,13 @@ def train(model, criterion, criterion_st, data_loader, optimizer, optimizer_st, # setup lr current_lr = lr_decay(c.lr, current_step, c.warmup_steps) + current_lr_st = lr_decay(0.01, current_step, c.warmup_steps) for params_group in optimizer.param_groups: params_group['lr'] = current_lr for params_group in optimizer_st.param_groups: - params_group['lr'] = current_lr + params_group['lr'] = current_lr_st optimizer.zero_grad() optimizer_st.zero_grad()