KoBART_with_LED / README.md
noahkim's picture
Update README.md
e7a7cf1
|
raw
history blame
1.14 kB
metadata
language: ko
license: mit
tags:
  - summarization
  - bart
  - longformer

<<20220904 Commit>>

๊ฐœ์ธ ์Šคํ„ฐ๋””์šฉ์œผ๋กœ ๊ธด ๋ฌธ์žฅ์˜ ์š”์•ฝ ๋ชจ๋ธ ํŠนํ™”๋œ ๋ชจ๋ธ์„ ๋งŒ๋“ค๊ธฐ ์œ„ํ•ด KoBART์˜ Input Size๋ฅผ 4096์œผ๋กœ ๋ณ€๊ฒฝํ•˜์—ฌ Customizing ์ง„ํ–‰ํ–ˆ์Šต๋‹ˆ๋‹ค. KoBART ๋ชจ๋ธ์„ Longformer Encoder Decoder๋กœ ๋ณ€ํ™˜ํ•œ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค. ๊ธฐ์กด์˜ transformers์˜ Model Load๋กœ๋Š” ๋ชจ๋ธ์„ ์“ฐ๊ธฐ๊ฐ€ ์–ด๋ ต์Šต๋‹ˆ๋‹ค. ๊ฐ™์ด ์˜ฌ๋ ค๋†“์€ load_model.py๋ฅผ ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌํ™”ํ•˜์—ฌ ๋ชจ๋ธ์„ ๋กœ๋“œํ•˜์‹œ๋ฉด ์‚ฌ์šฉ ๊ฐ€๋Šฅํ•ฉ๋‹ˆ๋‹ค.

KoBART ๋ชจ๋ธ์€

์ถ”ํ›„ ์ง€์†์ ์ธ ์—…๋ฐ์ดํŠธ๋กœ ๋ณ„๋„์˜ ๋ถˆํŽธํ•จ ์—†์ด ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋„๋ก ๋ฐ”๊พธ๊ฒ ์Šต๋‹ˆ๋‹ค.

Python Code


from transformers import PreTrainedTokenizerFas
from load_model import LongformerBartForConditionalGeneration

tokenizer = PreTrainedTokenizerFast.from_pretrained("hohovisual/KoBART_with_LED")

model = LongformerBartForConditionalGeneration.from_pretrained("hohovisual/KoBART_with_LED")