Upload app.py
Browse files
app.py
ADDED
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
from transformers import (T5ForConditionalGeneration,T5Tokenizer)
|
3 |
+
import gradio as gr
|
4 |
+
|
5 |
+
def set_seed(seed):
|
6 |
+
torch.manual_seed(seed)
|
7 |
+
set_seed(42)
|
8 |
+
|
9 |
+
best_model_path = "aditi2222/t5-paraphrase"
|
10 |
+
model = T5ForConditionalGeneration.from_pretrained(best_model_path)
|
11 |
+
tokenizer = T5Tokenizer.from_pretrained('aditi2222/t5-paraphrase')
|
12 |
+
|
13 |
+
def tokenize_data(text):
|
14 |
+
# Tokenize the review body
|
15 |
+
#input_ = "paraphrase: "+ str(text) + ' </s>'
|
16 |
+
|
17 |
+
max_len = 64
|
18 |
+
# tokenize inputs
|
19 |
+
tokenized_inputs = tokenizer(input_, padding='max_length', truncation=True, max_length=max_len, return_attention_mask=True, return_tensors='pt')
|
20 |
+
|
21 |
+
inputs={"input_ids": tokenized_inputs['input_ids'],
|
22 |
+
"attention_mask": tokenized_inputs['attention_mask']
|
23 |
+
}
|
24 |
+
|
25 |
+
return inputs
|
26 |
+
|
27 |
+
def generate_answers(text):
|
28 |
+
inputs = tokenize_data(text)
|
29 |
+
results= model.generate(input_ids= inputs['input_ids'].to(device), attention_mask=inputs['attention_mask'].to(device), do_sample=True,
|
30 |
+
max_length=64,
|
31 |
+
top_k=120,
|
32 |
+
top_p=0.98,
|
33 |
+
early_stopping=True,
|
34 |
+
num_return_sequences=1)
|
35 |
+
answer = tokenizer.decode(results[0], skip_special_tokens=True)
|
36 |
+
return answer
|
37 |
+
|
38 |
+
iface = gr.Interface(fn=generate_answers, inputs=['text'],outputs=["text"])
|
39 |
+
iface.launch(share=True)
|