Update train.py

train.py

@@ -8,8 +8,8 @@ from datasets import load_dataset, DatasetDict, Dataset
 from tokenizers import ByteLevelBPETokenizer
 
 BATCH_SIZE = 96
-EPOCHS = 2
-LEARNING_RATE = 2e-5
+EPOCHS = 10
+LEARNING_RATE = 2e-4
 FACTOR = 512
 MAX_SEQ_LENGTH = 128
 VOCAB_SIZE = 32000
@@ -24,7 +24,7 @@ PUSH_TO_HUB = True
 
 def load_data():
     pretrain = load_dataset(INPUT_DATASET, "cosmopedia-v2", split="train", streaming=True)
-    pretrain = Dataset.from_generator(lambda: pretrain.take(int(3e+6)))
+    pretrain = Dataset.from_generator(lambda: pretrain.take(int(3e+5)))
     instruct = load_dataset(INSTRUCT_DATASET, split="train", streaming=True)
     instruct = Dataset.from_generator(lambda: instruct.take(int(5e+5)))
     dataset_dict = DatasetDict({