hyunwoo3235
commited on
Commit
โข
33e0a37
1
Parent(s):
fdbe804
Update README.md
Browse files
README.md
CHANGED
@@ -1,9 +1,49 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
language:
|
4 |
-
- ko
|
5 |
pipeline_tag: image-to-text
|
6 |
tags:
|
7 |
-
- trocr
|
8 |
-
- vision-encoder-decoder
|
9 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
language:
|
4 |
+
- ko
|
5 |
pipeline_tag: image-to-text
|
6 |
tags:
|
7 |
+
- trocr
|
8 |
+
- vision-encoder-decoder
|
9 |
+
---
|
10 |
+
|
11 |
+
# trocr-small-korean
|
12 |
+
|
13 |
+
## Model Details
|
14 |
+
|
15 |
+
TrOCR์ Encoder-Decoder ๋ชจ๋ธ๋ก, ์ด๋ฏธ์ง ํธ๋์คํฌ๋จธ ์ธ์ฝ๋์ ํ
์คํธ ํธ๋์คํฌ๋จธ ๋์ฝ๋๋ก ์ด๋ฃจ์ด์ ธ ์์ต๋๋ค.
|
16 |
+
์ด๋ฏธ์ง ์ธ์ฝ๋๋ DeiT ๊ฐ์ค์น๋ก ์ด๊ธฐํ๋์๊ณ , ํ
์คํธ ๋์ฝ๋๋ ์์ฒด์ ์ผ๋ก ํ์ตํ RoBERTa ๊ฐ์ค์น๋ก ์ด๊ธฐํ๋์์ต๋๋ค.
|
17 |
+
|
18 |
+
์ด ์ฐ๊ตฌ๋ ๊ตฌ๊ธ์ TPU Research Cloud(TRC)๋ฅผ ํตํด ์ง์๋ฐ์ Cloud TPU๋ก ํ์ต๋์์ต๋๋ค.
|
19 |
+
|
20 |
+
## How to Get Started with the Model
|
21 |
+
|
22 |
+
```python
|
23 |
+
import torch
|
24 |
+
|
25 |
+
from transformers import VisionEncoderDecoderModel
|
26 |
+
|
27 |
+
model = VisionEncoderDecoderModel.from_pretrained("team-lucid/trocr-small-korean")
|
28 |
+
|
29 |
+
pixel_values = torch.rand(1, 3, 384, 384)
|
30 |
+
generated_ids = model.generate(pixel_values)
|
31 |
+
```
|
32 |
+
|
33 |
+
## Training Details
|
34 |
+
### Training Data
|
35 |
+
|
36 |
+
ํด๋น ๋ชจ๋ธ์ [synthtiger](https://github.com/clovaai/synthtiger)๋ก ํฉ์ฑ๋ 6M๊ฐ์ ์ด๋ฏธ์ง๋ก ํ์ต๋์์ต๋๋ค
|
37 |
+
|
38 |
+
### Training Hyperparameters
|
39 |
+
|
40 |
+
| Hyperparameter | Small |
|
41 |
+
|:--------------------|--------:|
|
42 |
+
| Warmup Steps | 4,000 |
|
43 |
+
| Learning Rates | 1e-4 |
|
44 |
+
| Batch Size | 512 |
|
45 |
+
| Weight Decay | 0.01 |
|
46 |
+
| Max Steps | 500,000 |
|
47 |
+
| Learning Rate Decay | 0.1 |
|
48 |
+
| \\(Adam\beta_1\\) | 0.9 |
|
49 |
+
| \\(Adam\beta_2\\) | 0.98 |
|