Create README.md

README.md

@@ -0,0 +1,22 @@
+training_args = TrainingArguments(
+    output_dir='t5-base-newsela-biendata-with-domain-adaptation',
+    num_train_epochs=20,
+    warmup_steps=250,
+    per_device_train_batch_size=BATCH_SIZE,
+    weight_decay=0.01,
+    learning_rate=2e-4,
+    # fp16=True,
+    optim="adafactor",
+)
+
+Step	Training Loss
+500	31.865400
+1000	37.290400
+1500	18.697100
+2000	4.517400
+2500	2.976500
+3000	2.558100
+3500	2.305200
+4000	2.187100
+4500	2.125000
+TrainOutput(global_step=4640, training_loss=11.326794861102925, metrics={'train_runtime': 5055.1704, 'train_samples_per_second': 58.688, 'train_steps_per_second': 0.918, 'total_flos': 0.0, 'train_loss': 11.326794861102925, 'epoch': 20.0})