Adjust learning rate, don't print out infinite ppltrans_lm

1 files changed, 7 insertions, 3 deletions

diff --git a/stanza/models/constituency/trans_lm.py b/stanza/models/constituency/trans_lm.py
index 4b1de1ba..6e8de99d 100644
--- a/stanza/models/constituency/trans_lm.py
+++ b/stanza/models/constituency/trans_lm.py
@@ -240,10 +240,14 @@ def train(optimizer, scheduler, epoch, device, train_data, model: nn.Module) ->
             lr = scheduler.get_last_lr()[0]
             ms_per_batch = (time.time() - start_time) * 1000 / log_interval
             cur_loss = total_loss / log_interval
-            ppl = math.exp(cur_loss)
+            try:
+                ppl = math.exp(cur_loss)
+                ppl = f"{ppl:8.2f}"
+            except OverflowError:
+                ppl = "inf"
             print(f'| epoch {epoch:3d} | {batch_idx:5d}/{num_batches:5d} batches | '
                   f'lr {lr:02.2f} | ms/batch {ms_per_batch:5.2f} | '
-                  f'loss {cur_loss:5.2f} | ppl {ppl:8.2f}')
+                  f'loss {cur_loss:5.2f} | ppl {ppl}')
             total_loss = 0
             start_time = time.time()
 
@@ -286,7 +290,7 @@ DEFAULT_FILES = {
 def parse_args(args):
     parser = argparse.ArgumentParser()
     parser.add_argument('--epochs', default=5, type=int, help='Num epochs to run')
-    parser.add_argument('--learning_rate', default=5.0, type=float, help='Initial learning rate')
+    parser.add_argument('--learning_rate', default=0.2, type=float, help='Initial learning rate')
 
     parser.add_argument('--data_dir', default="data/trans_lm", help='Where to find the data')
     parser.add_argument('--lang', default='vi', help='Which language to train - sets defaults for the data files, for example')