t5-large-korean-text-summary

์ด ๋ชจ๋ธ์€ lcw99 / t5-large-korean-text-summary์„ klue-ynat์œผ๋กœ ํ›ˆ๋ จ์‹œ์ผœ ๋งŒ๋“  ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.
Input = ['IT๊ณผํ•™','๊ฒฝ์ œ','์‚ฌํšŒ','์ƒํ™œ๋ฌธํ™”','์„ธ๊ณ„','์Šคํฌ์ธ ','์ •์น˜']
OUTPUT = ๊ฐ label์— ๋งž๋Š” ๋‰ด์Šค ๊ธฐ์‚ฌ ์ œ๋ชฉ์„ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค.
๋ฐฐ์น˜๋‹จ์œ„๋กœ ์ถ”๋ก ํ•˜๊ณ ์‹ถ๋‹ค๋ฉด batch_encode_plus๋ฅผ ์‚ฌ์šฉํ•˜์‹œ๋ฉด ๋ฉ๋‹ˆ๋‹ค.
git : https://github.com/taemin6697

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model_dir = "kfkas/t5-large-korean-news-title-klue-ynat"
tokenizer = AutoTokenizer.from_pretrained(model_dir)
model = AutoModelForSeq2SeqLM.from_pretrained(model_dir)
model.to(device)

label_list = ['IT๊ณผํ•™','๊ฒฝ์ œ','์‚ฌํšŒ','์ƒํ™œ๋ฌธํ™”','์„ธ๊ณ„','์Šคํฌ์ธ ','์ •์น˜']
text = "IT๊ณผํ•™"

input_ids = tokenizer.encode(text,return_tensors="pt").to(device)
with torch.no_grad():
  output = model.generate(
    input_ids,
    do_sample=True, #์ƒ˜ํ”Œ๋ง ์ „๋žต ์‚ฌ์šฉ
    max_length=128, # ์ตœ๋Œ€ ๋””์ฝ”๋”ฉ ๊ธธ์ด๋Š” 50
    top_k=50, # ํ™•๋ฅ  ์ˆœ์œ„๊ฐ€ 50์œ„ ๋ฐ–์ธ ํ† ํฐ์€ ์ƒ˜ํ”Œ๋ง์—์„œ ์ œ์™ธ
    top_p=0.95, # ๋ˆ„์  ํ™•๋ฅ ์ด 95%์ธ ํ›„๋ณด์ง‘ํ•ฉ์—์„œ๋งŒ ์ƒ์„ฑ
)
decoded_output = tokenizer.batch_decode(output, skip_special_tokens=True)[0]
print(decoded_output)#SKํ…”๋ ˆ์ฝค ์Šค๋งˆํŠธ ๋ชจ๋ฐ”์ผ ์š”๊ธˆ์ œ ์‹œ์ฆŒ1 ์ถœ์‹œ

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • optimizer: None
  • training_precision: float16

Training results

Framework versions

  • Transformers 4.22.1
  • TensorFlow 2.10.0
  • Datasets 2.5.1
  • Tokenizers 0.12.1
Downloads last month
11
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Collection including kfkas/t5-large-korean-news-title-klue-ynat