Spaces:
Runtime error
Runtime error
DDingcheol
commited on
Commit
ยท
7063ff1
1
Parent(s):
7172545
Update app.py
Browse files
app.py
CHANGED
@@ -1,78 +1,43 @@
|
|
1 |
#ํ๊น
ํ์ด์ค์์ ๋์๊ฐ ์ ์๋๋ก ๋ฐ๊พธ์ด ๋ณด์์
|
2 |
|
|
|
3 |
import torch
|
4 |
-
from transformers import
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
41 |
-
# ๋ชจ๋ธ ๊ฐ์ ธ์ค๊ธฐ
|
42 |
-
from transformers import AutoModelForSequenceClassification, TrainingArguments, Trainer
|
43 |
-
|
44 |
-
model_name = 'microsoft/git-base-vqav2'
|
45 |
-
model = AutoModelForSequenceClassification.from_pretrained(model_name)
|
46 |
-
|
47 |
-
# Trainer๋ฅผ ์ฌ์ฉํ์ฌ ๋ชจ๋ธ ํ์ต
|
48 |
-
tokenizer = BertTokenizerFast.from_pretrained('bert-base-multilingual-cased')
|
49 |
-
|
50 |
-
def preprocess_function(examples):
|
51 |
-
tokenized_inputs = tokenizer(examples['question'], truncation=True, padding=True)
|
52 |
-
return {
|
53 |
-
'input_ids': tokenized_inputs['input_ids'],
|
54 |
-
'attention_mask': tokenized_inputs['attention_mask'],
|
55 |
-
'pixel_values': [(4, 3, 244, 244)] * len(tokenized_inputs['input_ids']),
|
56 |
-
'pixel_mask': [1] * len(tokenized_inputs['input_ids']),
|
57 |
-
'labels': [[label] for label in examples['answers']]
|
58 |
-
}
|
59 |
-
|
60 |
-
dataset = load_dataset("Multimodal-Fatima/OK-VQA_train")['train'].select(range(300))
|
61 |
-
ok_vqa_dataset = dataset.map(preprocess_function, batched=True)
|
62 |
-
ok_vqa_dataset.set_format(type='torch', columns=['input_ids', 'attention_mask', 'pixel_values', 'pixel_mask', 'labels'])
|
63 |
-
|
64 |
-
training_args = TrainingArguments(
|
65 |
-
output_dir='./results',
|
66 |
-
num_train_epochs=20,
|
67 |
-
per_device_train_batch_size=4,
|
68 |
-
logging_steps=500,
|
69 |
-
)
|
70 |
-
|
71 |
-
trainer = Trainer(
|
72 |
-
model=model,
|
73 |
-
args=training_args,
|
74 |
-
train_dataset=ok_vqa_dataset
|
75 |
)
|
76 |
|
77 |
-
|
78 |
-
trainer.train()
|
|
|
1 |
#ํ๊น
ํ์ด์ค์์ ๋์๊ฐ ์ ์๋๋ก ๋ฐ๊พธ์ด ๋ณด์์
|
2 |
|
3 |
+
import gradio as gr
|
4 |
import torch
|
5 |
+
from transformers import BertTokenizer, BertForSequenceClassification
|
6 |
+
|
7 |
+
# ๋ชจ๋ธ ์ด๊ธฐํ ๋ฐ ๊ฐ์ค์น ๋ถ๋ฌ์ค๊ธฐ
|
8 |
+
model_name = 'microsoft/git-base-vqav2' # ์ฌ์ฉํ ๋ชจ๋ธ์ ์ด๋ฆ
|
9 |
+
model = BertForSequenceClassification.from_pretrained(model_name)
|
10 |
+
tokenizer = BertTokenizer.from_pretrained(model_name)
|
11 |
+
|
12 |
+
# ์์ธก ํจ์ ์ ์
|
13 |
+
def predict_answer(image, question):
|
14 |
+
inputs = tokenizer(question, return_tensors='pt')
|
15 |
+
input_ids = inputs['input_ids']
|
16 |
+
attention_mask = inputs['attention_mask']
|
17 |
+
|
18 |
+
# ์ด๋ฏธ์ง์ ๊ด๋ จ๋ ์ฒ๋ฆฌ ์ํ
|
19 |
+
# ์ด๋ฏธ์ง ์ฒ๋ฆฌ ์ฝ๋๋ฅผ ์ฌ๊ธฐ์ ์ถ๊ฐํด์ผ ํฉ๋๋ค (์
๋ ฅ๋ ์ด๋ฏธ์ง์ ๋ํ ์ ์ฒ๋ฆฌ ๋ฑ)
|
20 |
+
|
21 |
+
# ๋ชจ๋ธ์ ์
๋ ฅ ๋ฐ์ดํฐ๋ฅผ ์ ๋ฌํ์ฌ ์์ธก ์ํ
|
22 |
+
with torch.no_grad():
|
23 |
+
outputs = model(input_ids=input_ids, attention_mask=attention_mask)
|
24 |
+
|
25 |
+
# ์์ธก ๊ฒฐ๊ณผ์์ ๊ฐ์ฅ ๋์ ํ๋ฅ ์ ๊ฐ์ง ๋ ์ด๋ธ ID ๊ฐ์ ธ์ค๊ธฐ
|
26 |
+
predicted_label_id = torch.argmax(outputs.logits).item()
|
27 |
+
predicted_label = id_to_label_fn(predicted_label_id)
|
28 |
+
|
29 |
+
return predicted_label
|
30 |
+
|
31 |
+
iface = gr.Interface(
|
32 |
+
fn=predict_answer,
|
33 |
+
inputs=["image", "text"],
|
34 |
+
outputs="text",
|
35 |
+
title="Visual Question Answering",
|
36 |
+
description="Input an image and a question to get the model's answer.",
|
37 |
+
example=[
|
38 |
+
"https://your_image_url.jpg",
|
39 |
+
"What is shown in the image?"
|
40 |
+
]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
41 |
)
|
42 |
|
43 |
+
iface.launch()
|
|