Spaces:

nroggendorff
/

train-llama

Starting on L40S

nroggendorff commited on Jul 31, 2024

Commit

d1dd4d3

verified ·

1 Parent(s): 74bf403

Update train.py

Files changed (1) hide show

train.py CHANGED Viewed

@@ -8,7 +8,7 @@ from datasets import load_dataset
 from tokenizers import ByteLevelBPETokenizer
 MAX_SEQ_LENGTH = 512
-BATCH_SIZE = 256
 EPOCHS = 3
 LEARNING_RATE = 2e-4
 FACTOR = 2
@@ -23,7 +23,7 @@ CLIPPING = 1.0
 PUSH_TO_HUB = True
 def load_data():
-    dataset = load_dataset(INPUT_DATASET, split="train")#.select(range(int(2e+4)))
     return dataset
 def create_tokenizer(training_corpus):

 from tokenizers import ByteLevelBPETokenizer
 MAX_SEQ_LENGTH = 512
+BATCH_SIZE = 128
 EPOCHS = 3
 LEARNING_RATE = 2e-4
 FACTOR = 2
 PUSH_TO_HUB = True
 def load_data():
+    dataset = load_dataset(INPUT_DATASET, split="train").select(range(int(2e+6)))
     return dataset
 def create_tokenizer(training_corpus):