eaglelandsonce
commited on
Commit
•
d4877ec
1
Parent(s):
b9f95ed
Create 11_Tokenizer_Detokenizer.py
Browse files
pages/11_Tokenizer_Detokenizer.py
ADDED
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
+
|
4 |
+
# Load the tokenizer
|
5 |
+
tokenizer = AutoTokenizer.from_pretrained('gpt2')
|
6 |
+
|
7 |
+
# Streamlit app title
|
8 |
+
st.title("Tokenizer and Detokenizer using GPT-2 for 2D Canvas")
|
9 |
+
st.write("Example: cr8 lg cnvs html js hlds 9 wbs becomes 060980002300300000700026900077142592771144804002890033500082008600026601443")
|
10 |
+
|
11 |
+
# Tokenization section
|
12 |
+
st.header("Tokenization")
|
13 |
+
sentence = st.text_input("Enter a sentence to tokenize:", "cr8 lg cnvs html js hlds 9 wbs")
|
14 |
+
|
15 |
+
def format_token_ids(token_ids):
|
16 |
+
formatted_ids = [str(token_id).zfill(5) for token_id in token_ids]
|
17 |
+
return ''.join(formatted_ids)
|
18 |
+
|
19 |
+
if st.button("Tokenize"):
|
20 |
+
input_ids = tokenizer(sentence, return_tensors='pt').input_ids
|
21 |
+
token_ids_list = input_ids[0].tolist()
|
22 |
+
formatted_token_ids = format_token_ids(token_ids_list)
|
23 |
+
st.write("Tokenized input IDs (formatted):")
|
24 |
+
st.write(formatted_token_ids)
|
25 |
+
|
26 |
+
# Detokenization section
|
27 |
+
st.header("Detokenization")
|
28 |
+
token_ids = st.text_input("Enter token IDs (concatenated without spaces):", "619710116000284001536")
|
29 |
+
|
30 |
+
def split_token_ids(concatenated_ids, length=5):
|
31 |
+
return [concatenated_ids[i:i+length] for i in range(0, len(concatenated_ids), length)]
|
32 |
+
|
33 |
+
def remove_leading_zeros(grouped_ids):
|
34 |
+
return [id.lstrip('0') for id in grouped_ids]
|
35 |
+
|
36 |
+
if st.button("Detokenize"):
|
37 |
+
split_ids = split_token_ids(token_ids)
|
38 |
+
cleaned_ids = remove_leading_zeros(split_ids)
|
39 |
+
cleaned_token_ids_str = ' '.join(cleaned_ids)
|
40 |
+
token_id_list = [int(id) for id in cleaned_ids if id.isdigit()]
|
41 |
+
|
42 |
+
detokenized_sentence = tokenizer.decode(token_id_list)
|
43 |
+
|
44 |
+
st.write("Grouped and cleaned token IDs:")
|
45 |
+
st.write(cleaned_token_ids_str)
|
46 |
+
st.write("Detokenized sentence:")
|
47 |
+
st.write(detokenized_sentence)
|
48 |
+
|
49 |
+
# Load the model
|
50 |
+
gpt2 = AutoModelForCausalLM.from_pretrained('gpt2')
|
51 |
+
|
52 |
+
# Display help for the GPT-2 model
|
53 |
+
st.write("Help GPT2")
|
54 |
+
st.write(help(gpt2))
|