Created
February 23, 2024 19:06
-
-
Save almugabo/43212e321968547709b0dac38f3ab92c to your computer and use it in GitHub Desktop.
compare tokenizers of gemma and llama
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import os | |
import pandas as pd | |
import random | |
from transformers import AutoTokenizer | |
xFld = '....' # folder with a bunch of text files | |
xtok_tl = AutoTokenizer.from_pretrained('unsloth/tinyllama-bnb-4bit') | |
xtok_ge = AutoTokenizer.from_pretrained('unsloth/gemma-2b-bnb-4bit') | |
print('vocab_size llama:', xtok_tl.vocab_size) | |
print('vocab_size gemma:', xtok_ge.vocab_size) | |
def len_tl (xtext): | |
return len(xtok_tl.tokenize(xtext)) | |
def len_ge (xtext): | |
return len(xtok_ge.tokenize(xtext)) | |
xstr = 'everybody is happier when they see the loved ones' | |
xfiles = [xFld + x for x in random.sample(os.listdir(xFld), 1000) ] | |
xlst_res = [] | |
for xfile in xfiles: | |
with open(xfile, 'r') as xff: | |
xtext = xff.read() | |
xdata={} | |
xdata['tllama'] = len_tl (xtext) | |
xdata['gemma'] = len_ge (xtext) | |
xlst_res.append(xdata) | |
t1 = pd.DataFrame(xlst_res) | |
t1.tllama.hist() | |
t1.gemma.hist() | |
t1.describe() | |
''' | |
tllama gemma | |
count 1000.00000 1000.000000 | |
mean 571.06400 496.238000 | |
std 223.80022 192.722971 | |
min 76.00000 64.000000 | |
25% 417.00000 363.750000 | |
50% 555.00000 487.500000 | |
75% 690.00000 604.000000 | |
max 1947.00000 1595.000000 | |
''' |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment