Edit model card

mt5-small-indonesian-sum

This model is a fine-tuned version of google/mt5-small on an csebuetnlp/xlsum dataset.

Training results

Train Loss Validation Loss Train Rouge1 Train Rouge2 Train Rougel Train Rougelsum Train Gen Len
4.6665 2.6526 19.7876 6.6344 16.7880 16.8459 18.92

Training hyperparameters

Optimizer VALUE
learning_rate 2e-05
decay 0.0
beta_1 0.9
beta_2 0.999
epsilon 1e-07
amsgrad False
weight_decay_rate 0.01
  • training_precision: float32

Using this model in transformers (tested on 4.19.2)

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import re

news = """
Anggota Unit Perlindungan Rakyat Kurdi di kota Rabia, pada perbatasan Irak-Suriah. Pasukan Kurdi Irak dilaporkan sudah menguasai kembali kota Rabia meskipun banyak korban jatuh. Pejabat senior Kurdi Irak mengatakan pasukan Kurdi Peshmerga mencatat kemajuan lewat serangan dini hari di Rabia. Sementara itu, milisi ISIS berusaha memukul mundur pasukan Kurdi Suriah di bagian lain perbatasan. Hal ini terjadi saat koalisi pimpinan Amerika terus melanjutkan serangan udara terhadap sasaran ISIS di Suriah dan Irak. Hari Selasa (30 September) dilaporkan juga terjadi serangkaian serangan bom di ibu kota Irak, Baghdad dan kota suci Syiah, Karbala. Dalam perkembangan terpisah, sejumlah tank Turki berada di bukit di sepanjang perbatasan dekat kota Kobane, Suriah setelah sejumlah bom mengenai wilayah Turki saat terjadi bentrokan dengan milisi ISIS dan pejuang Kurdi. Pemerintah Turki diperkirakan akan menyampaikan mosi ke parlemen, agar menyetujui aksi militer terhadap ISIS di Irak dan Suriah.
"""

tokenizer = AutoTokenizer.from_pretrained("assamim/mt5-small-indonesian-sum")
model = AutoModelForSeq2SeqLM.from_pretrained("assamim/mt5-small-indonesian-sum", from_tf=True)


WHITESPACE_HANDLER = lambda k: re.sub('\s+', ' ', re.sub('\n+', ' ', k.strip()))

input_ids = tokenizer.encode(WHITESPACE_HANDLER(news1), return_tensors='pt')
summary_ids = model.generate(input_ids,
            min_length=20,
            max_length=200,
            num_beams=7,
            repetition_penalty=2.5,
            length_penalty=1.0,
            early_stopping=True,
            no_repeat_ngram_size=2,
            use_cache=True,
            do_sample = True,
            temperature = 0.8,
            top_k = 50,
            top_p = 0.95)

summary_text = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
print(summary_text)

Framework versions

  • Transformers 4.19.2
  • TensorFlow 2.8.0
  • Datasets 2.2.2
  • Tokenizers 0.12.1
Downloads last month
56
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.

Dataset used to train assamim/mt5-small-indonesian