Datasets:

hac541309
/

multilingual_tokenizers

Modalities:

Text

License:

Dataset card Data Studio Files Files and versions

xet

Community

multilingual_tokenizers / scripts /byte_exemplars.py

hac541309

Create scripts/byte_exemplars.py

ab60a05 about 3 years ago

Raw History Blame Contribute Delete

1.22 kB

import json
import tokenizers

# unicode exemplar json
with open("lang_exemplars.json") as file_handle:
    unicode_exemplars = json.load(file_handle)

# (17411, 'a')
print(len(unicode_exemplars),unicode_exemplars[0])

# ksx-1001 that includes 2350 korean "exemplar" characters
with open("ksx-1001.txt") as file_handle:
    kor_exemplar = file_handle.readlines()[0].split()

# (2350, '가')
len(kor_exemplar), kor_exemplar[0]

# initialize bytelevel tokenizer
ByteLevel = tokenizers.pre_tokenizers.ByteLevel(False,False)

# make new bytelevel exemplar json
byte_exemplars = []
for exemplar in unicode_exemplars:
    byte_exemplars.append(ByteLevel.pre_tokenize_str(exemplar)[0][0])
byte_exemplars = sorted(byte_exemplars,key=lambda x: [x,len(x)])
len(byte_exemplars), byte_exemplars[0]

# make new korean bytelevel exemplar json
kb_exemplar = []
for exemplar in kor_exemplar:
    kb_exemplar.append(ByteLevel.pre_tokenize_str(exemplar)[0][0])
kb_exemplar = sorted(kb_exemplar,key=lambda x: [x,len(x)])
len(kb_exemplar), kb_exemplar[0]

# write new byte_exemplar json. ensure_ascii=False is important
with open("byte_exemplars.json","w") as out_handle:
    json.dump(byte_exemplars,out_handle,ensure_ascii=False, indent=2)

	import json
	import tokenizers

	# unicode exemplar json
	with open("lang_exemplars.json") as file_handle:
	unicode_exemplars = json.load(file_handle)

	# (17411, 'a')
	print(len(unicode_exemplars),unicode_exemplars[0])

	# ksx-1001 that includes 2350 korean "exemplar" characters
	with open("ksx-1001.txt") as file_handle:
	kor_exemplar = file_handle.readlines()[0].split()

	# (2350, '가')
	len(kor_exemplar), kor_exemplar[0]

	# initialize bytelevel tokenizer
	ByteLevel = tokenizers.pre_tokenizers.ByteLevel(False,False)

	# make new bytelevel exemplar json
	byte_exemplars = []
	for exemplar in unicode_exemplars:
	byte_exemplars.append(ByteLevel.pre_tokenize_str(exemplar)[0][0])
	byte_exemplars = sorted(byte_exemplars,key=lambda x: [x,len(x)])
	len(byte_exemplars), byte_exemplars[0]

	# make new korean bytelevel exemplar json
	kb_exemplar = []
	for exemplar in kor_exemplar:
	kb_exemplar.append(ByteLevel.pre_tokenize_str(exemplar)[0][0])
	kb_exemplar = sorted(kb_exemplar,key=lambda x: [x,len(x)])
	len(kb_exemplar), kb_exemplar[0]

	# write new byte_exemplar json. ensure_ascii=False is important
	with open("byte_exemplars.json","w") as out_handle:
	json.dump(byte_exemplars,out_handle,ensure_ascii=False, indent=2)