Upload dyve_tts/eval/math/modeling/clean_merges.py with huggingface_hub
Browse files
dyve_tts/eval/math/modeling/clean_merges.py
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
|
| 3 |
+
Clean GPT-2 merges file, removing all tokens from the tokenizer that have
|
| 4 |
+
digits, other than the "0" - "9" tokens.
|
| 5 |
+
|
| 6 |
+
"""
|
| 7 |
+
|
| 8 |
+
merges_fname = "merges_gpt2.txt"
|
| 9 |
+
new_merges_fname = "merges_gpt2_single_digit_numbers.txt"
|
| 10 |
+
|
| 11 |
+
def hasNumbers(inputString):
|
| 12 |
+
return any(char.isdigit() for char in inputString)
|
| 13 |
+
|
| 14 |
+
with open(new_merges_fname, 'w') as f_new:
|
| 15 |
+
with open(merges_fname, 'r') as f:
|
| 16 |
+
lines = f.read().split("\n")
|
| 17 |
+
for l in lines:
|
| 18 |
+
if len(l) < 1:
|
| 19 |
+
break
|
| 20 |
+
|
| 21 |
+
left, right = l.split(" ")
|
| 22 |
+
if hasNumbers(left) or hasNumbers(right):
|
| 23 |
+
print(left, right)
|
| 24 |
+
else:
|
| 25 |
+
f_new.write(l + "\n")
|
| 26 |
+
|
| 27 |
+
|