Nadil Karunarathna
commited on
Commit
·
6baca04
1
Parent(s):
03ce3fe
threads
Browse files
app.py
CHANGED
@@ -1,65 +1,60 @@
|
|
1 |
import gradio as gr
|
2 |
-
|
3 |
-
|
4 |
|
5 |
-
|
6 |
-
|
7 |
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
# global model, tokenizer
|
13 |
|
14 |
-
|
15 |
|
16 |
-
|
17 |
-
# torch.set_num_threads(2)
|
18 |
-
# tokenizer = T5TokenizerFast.from_pretrained("google/mt5-base")
|
19 |
-
# tokenizer.add_special_tokens({'additional_special_tokens': ['<ZWJ>']})
|
20 |
|
|
|
|
|
|
|
|
|
21 |
|
22 |
-
# def correct(text):
|
23 |
|
24 |
-
|
25 |
|
26 |
-
|
27 |
-
# inputs = tokenizer(
|
28 |
-
# text,
|
29 |
-
# return_tensors='pt',
|
30 |
-
# padding='do_not_pad',
|
31 |
-
# max_length=1024
|
32 |
-
# )
|
33 |
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
# )
|
42 |
-
# prediction = outputs[0]
|
43 |
|
44 |
-
|
45 |
-
|
46 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
47 |
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
# if token == special_token_id_to_keep or token not in all_special_ids
|
52 |
-
# ]
|
53 |
|
54 |
-
|
|
|
|
|
|
|
|
|
55 |
|
56 |
-
|
57 |
|
58 |
-
|
59 |
-
import os
|
60 |
-
return f"vCPUs: {os.cpu_count()}"
|
61 |
|
62 |
-
|
63 |
-
|
64 |
-
demo = gr.Interface(fn=test, inputs="text", outputs="text")
|
65 |
demo.launch()
|
|
|
1 |
import gradio as gr
|
2 |
+
import torch
|
3 |
+
import re
|
4 |
|
5 |
+
model = None
|
6 |
+
tokenizer = None
|
7 |
|
8 |
+
def init():
|
9 |
+
from transformers import MT5ForConditionalGeneration, T5TokenizerFast
|
10 |
+
import os
|
|
|
|
|
11 |
|
12 |
+
global model, tokenizer
|
13 |
|
14 |
+
hf_token = os.environ.get("HF_TOKEN")
|
|
|
|
|
|
|
15 |
|
16 |
+
model = MT5ForConditionalGeneration.from_pretrained("lm-spell/mt5-base-ft-ssc", token=hf_token)
|
17 |
+
torch.set_num_threads(16)
|
18 |
+
tokenizer = T5TokenizerFast.from_pretrained("google/mt5-base")
|
19 |
+
tokenizer.add_special_tokens({'additional_special_tokens': ['<ZWJ>']})
|
20 |
|
|
|
21 |
|
22 |
+
def correct(text):
|
23 |
|
24 |
+
model.eval()
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
|
26 |
+
text = re.sub(r'\u200d', '<ZWJ>', text)
|
27 |
+
inputs = tokenizer(
|
28 |
+
text,
|
29 |
+
return_tensors='pt',
|
30 |
+
padding='do_not_pad',
|
31 |
+
max_length=1024
|
32 |
+
)
|
|
|
|
|
33 |
|
34 |
+
with torch.inference_mode():
|
35 |
+
outputs = model.generate(
|
36 |
+
input_ids=inputs["input_ids"],
|
37 |
+
attention_mask=inputs["attention_mask"],
|
38 |
+
max_length=1024,
|
39 |
+
num_beams=1,
|
40 |
+
do_sample=False,
|
41 |
+
)
|
42 |
+
prediction = outputs[0]
|
43 |
|
44 |
+
special_token_id_to_keep = tokenizer.convert_tokens_to_ids('<ZWJ>')
|
45 |
+
all_special_ids = set(tokenizer.all_special_ids)
|
46 |
+
pred_tokens = prediction.cpu()
|
|
|
|
|
47 |
|
48 |
+
tokens_list = pred_tokens.tolist()
|
49 |
+
filtered_tokens = [
|
50 |
+
token for token in tokens_list
|
51 |
+
if token == special_token_id_to_keep or token not in all_special_ids
|
52 |
+
]
|
53 |
|
54 |
+
prediction_decoded = tokenizer.decode(filtered_tokens, skip_special_tokens=False).replace('\n', '').strip()
|
55 |
|
56 |
+
return re.sub(r'<ZWJ>\s?', '\u200d', prediction_decoded)
|
|
|
|
|
57 |
|
58 |
+
init()
|
59 |
+
demo = gr.Interface(fn=correct, inputs="text", outputs="text")
|
|
|
60 |
demo.launch()
|