Spaces:

nroggendorff
/

train-llama

Starting on L40S

nroggendorff commited on Aug 4, 2024

Commit

2c73579

verified ·

1 Parent(s): cbb8349

Update train.py

Files changed (1) hide show

train.py CHANGED Viewed

@@ -10,7 +10,7 @@ from tokenizers import ByteLevelBPETokenizer
 MAX_SEQ_LENGTH = 2048
 BATCH_SIZE = 4
 EPOCHS = 2
-LEARNING_RATE = 2e-2
 FACTOR = 1024
 VOCAB_SIZE = 3200
 INPUT_DATASET = "HuggingFaceTB/smollm-corpus"

 MAX_SEQ_LENGTH = 2048
 BATCH_SIZE = 4
 EPOCHS = 2
+LEARNING_RATE = 1e-4
 FACTOR = 1024
 VOCAB_SIZE = 3200
 INPUT_DATASET = "HuggingFaceTB/smollm-corpus"