Update train.py

train.py

@@ -7,26 +7,26 @@ from transformers import AutoTokenizer, LlamaConfig, LlamaForCausalLM, TrainingA
 from datasets import load_dataset, DatasetDict, Dataset
 from tokenizers import ByteLevelBPETokenizer
 
-MAX_SEQ_LENGTH = 512
-BATCH_SIZE = 128
+MAX_SEQ_LENGTH = 128
+BATCH_SIZE = 256
 EPOCHS = 2
-LEARNING_RATE = 2e-2
+LEARNING_RATE = 2e-3
 FACTOR = 256
 VOCAB_SIZE = 32000
 INPUT_DATASET = "HuggingFaceTB/smollm-corpus"
 INSTRUCT_DATASET = "nroggendorff/openhermes"
 OUTPUT_REPO = "smallama"
-FP16 = True
+FP16 = False
 WARMUP_STEPS = 20
-DECAY = 0.01
+DECAY = 1e-3
 GRADIENT_ACCUMULATION_STEPS = 1
 CLIPPING = 1.0
 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(200000))
-    instruct = load_dataset(INSTRUCT_DATASET, split="train").select(range(200000))
+    pretrain = Dataset.from_generator(lambda: pretrain.take(int(2e+6)))
+    instruct = load_dataset(INSTRUCT_DATASET, split="train").select(range(int(2e+5)))
     dataset_dict = DatasetDict({
         'pretrain': pretrain,
         'instruct': instruct