aisingapore / sealion-bert-base

huggingface.co
Total runs: 223
24-hour runs: 3
7-day runs: 22
30-day runs: -243
Model's Last Updated: January 08 2024
fill-mask

Introduction of sealion-bert-base

Model Details of sealion-bert-base

SEA-LION-BERT

SEA-LION stands for Southeast Asian Languages In One Network .

This is the card for the SEA-LION-BERT base model.

How To Use
from transformers import AutoModelForMaskedLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained('aisingapore/sealion-bert-base', trust_remote_code=True)
model = AutoModelForMaskedLM.from_pretrained('aisingapore/sealion-bert-base', trust_remote_code=True)

# prepare input
text = "Give me a <|mask|>!!!"
encoded_input = tokenizer(text, return_tensors='pt')
Model Details
Model Description

The SEA-LION-BERT model is built on the MosaicBERT architecture and has a vocabulary size of 256K.

For tokenization, the model employs our custom SEABPETokenizer, which is specially tailored for SEA languages, ensuring optimal model performance.

The training data for SEA-LION-BERT encompasses 790B tokens.

  • Developed by: Products Pillar, AI Singapore
  • Funded by: Singapore NRF
  • Model type: Encoder
  • Languages: English, Chinese, Indonesian, Malay, Thai, Vietnamese, Filipino, Tamil, Burmese, Khmer, Lao
  • License: MIT License
Training Details
Data

SEA-LION was trained on 790B tokens of the following data:

Data Source Tokens Percentage
RefinedWeb - English 571.3B 72.26%
mC4 - Chinese 91.2B 11.54%
mC4 - Indonesian 14.7B 1.86%
mC4 - Malay 2.9B 0.36%
mC4 - Filipino 5.3B 0.67%
mC4 - Burmese 4.9B 0.61%
mC4 - Vietnamese 63.4B 8.02%
mC4 - Thai 21.6B 2.74%
mC4 - Lao 1.1B 0.14%
mC4 - Khmer 3.9B 0.50%
mC4 - Tamil 10.2B 1.29%
Infrastructure

SEA-LION was trained using MosaicML Composer on the following hardware:

Training Details SEA-LION-BERT
Nvidia A100 40GB GPU 4
Training Duration 14 days
Configuration
HyperParameter SEA-LION-BERT
Precision bfloat16
Optimizer decoupled_adamw
Scheduler linear_decay_with_warmup
Learning Rate 5e-4
Global Batch Size 448
Micro Batch Size 56
Technical Specifications
Model Architecture and Objective

SEA-LION-BERT is an encoder model using the MosaicBERT architecture.

Parameter SEA-LION-BERT
Layers 12
d_model 768
head_dim 12
Vocabulary 256000
Sequence Length 128
Tokenizer Details

We sample 20M lines from the training data to train the tokenizer.
The framework for training is SentencePiece .
The tokenizer type is Byte-Pair Encoding (BPE).

The Team

Montalan Jann Railey
Nguyen Thanh Ngan
Rengarajan Hamsawardhini
Teo Eng Sipp Leslie
Tjhi William

Acknowledgements

AI Singapore is a national programme supported by the National Research Foundation, Singapore and hosted by the National University of Singapore. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not reflect the views of National Research Foundation, Singapore.

Contact

For more info, please contact us using this SEA-LION Inquiry Form

Runs of aisingapore sealion-bert-base on huggingface.co

223
Total runs
3
24-hour runs
12
3-day runs
22
7-day runs
-243
30-day runs

More Information About sealion-bert-base huggingface.co Model

More sealion-bert-base license Visit here:

https://choosealicense.com/licenses/mit

sealion-bert-base huggingface.co

sealion-bert-base huggingface.co is an AI model on huggingface.co that provides sealion-bert-base's model effect (), which can be used instantly with this aisingapore sealion-bert-base model. huggingface.co supports a free trial of the sealion-bert-base model, and also provides paid use of the sealion-bert-base. Support call sealion-bert-base model through api, including Node.js, Python, http.

sealion-bert-base huggingface.co Url

https://huggingface.co/aisingapore/sealion-bert-base

aisingapore sealion-bert-base online free

sealion-bert-base huggingface.co is an online trial and call api platform, which integrates sealion-bert-base's modeling effects, including api services, and provides a free online trial of sealion-bert-base, you can try sealion-bert-base online for free by clicking the link below.

aisingapore sealion-bert-base online free url in huggingface.co:

https://huggingface.co/aisingapore/sealion-bert-base

sealion-bert-base install

sealion-bert-base is an open source model from GitHub that offers a free installation service, and any user can find sealion-bert-base on GitHub to install. At the same time, huggingface.co provides the effect of sealion-bert-base install, users can directly use sealion-bert-base installed effect in huggingface.co for debugging and trial. It also supports api for free installation.

sealion-bert-base install url in huggingface.co:

https://huggingface.co/aisingapore/sealion-bert-base

Url of sealion-bert-base

sealion-bert-base huggingface.co Url

Provider of sealion-bert-base huggingface.co

aisingapore
ORGANIZATIONS

Other API from aisingapore

huggingface.co

Total runs: 13
Run Growth: 3
Growth Rate: 23.08%
Updated:March 02 2023