---
license: apache-2.0
---

# Model Card for SEA LION

<!-- Provide a quick summary of what the model is/does. -->
SEA LION is a collection of LLMs which has been pretrained and instruct-tuned for the Southeast Asia region.
The models range from 3 billion to 7 billion parameters.
This is the repository for the 3B pretrained model.

## Model Details

### Model Description

<!-- Provide a longer summary of what this model is. -->
The SEA LION model is a significant leap forward in the field of natural language processing and understanding,
specifically trained to understand South-East Asia (SEA) regional context.
SEA LION stands for SouthEast Asian Languages In One Network.
The SEA LION model comes in two variants, one with 3 billion parameters and another with 7 billion parameters.
Both variants are built on the robust MPT architecture and utilize a vocabulary size of 256K.
The model employs our proprietary SEABPETokenizer for tokenization.
Our SEABPETokenizer is specially tailored for SEA languages, ensuring optimal model performance.
The training data for SEA LION is encompasses 1 trillion tokens.


- **Developed by:** Products Pillar, AI Singapore
- **Funded by [optional]:** Singapore NRF
- **Shared by [optional]:** N/A
- **Model type:** Decoder
- **Language(s) (NLP):** English, Chinese, Indonesian, Malay, Thai, Vietnamese, Filipino/Tagalog, Tamil, Burnese, Khmer, Lao
- **License:** Apache 2.0
- **Finetuned from model [optional]:** N/A

### Model Sources [optional]

<!-- Provide the basic links for the model. -->

- **Repository:** _Coming soon_
- **Paper [optional]:** _Coming soon_
- **Demo [optional]:** _Coming soon_

## Uses

<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->

### Direct Use

<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->

[More Information Needed]

### Downstream Use [optional]

<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->

[More Information Needed]

### Out-of-Scope Use

<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->

[More Information Needed]

## Bias, Risks, and Limitations

<!-- This section is meant to convey both technical and sociotechnical limitations. -->

[More Information Needed]

### Recommendations

<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.

## How to Get Started with the Model

Use the code below to get started with the model.

[ Todo: Insert Code Here ]

## Training Details

### Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

SEA LION 3B was trained on 980B tokens of RefinedWeb (English) and mC4 (Chinese, Indonesian, Malay, Filipino/Tagalog, Burmese, Vietnamese, Thai, Lao, Khmer, Tamil).

| Data Source            | Tokens | Percentage |
|------------------------|--------|------------|
| RefinedWeb - English   | 571.3B |     62.80% |
| mC4 - Chinese          |  91.2B |     10.03% |
| mC4 - Indonesian       |   3.6B |      0.40% |
| mC4 - Malay            |   0.7B |      0.08% |
| mC4 - Filipino/Tagalog |   1.3B |      0.15% |
| mC4 - Burmese          |   1.2B |      0.13% |
| mC4 - Vietnamese       |  63.4B |      6.97% |
| mC4 - Thai             |  10.8B |      1.19% |
| mC4 - Lao              |   0.3B |      0.03% |
| mC4 - Khmer            |   0.9B |      0.11% |
| mC4 - Tamil            |   2.5B |      0.28% |
| Python                 |  20.9B |      2.30% |
| Javascript             |  55.6B |      6.11% |
| Shell                  |   1.3B |      0.14% |
| SQL                    |   6.4B |      0.70% |
| Markdown               |  26.6B |      2.91% |
| StackExchange          |  21.2B |      2.33% |
| ArXiv                  |  30.6B |      3.35% |

### Training Procedure 

<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->

SEA LION 3B was trained on 256 A100 40GB GPUs, using MosaicML Composer.

#### Preprocessing [optional]

N/A

#### Training Hyperparameters

| Hyperparameter    | Value             |
|-------------------|-------------------|
| Precision         | bfloat16          |
| Optimizer         | decoupled_adamw   |
| Scheduler         | cosin_with_warmup |
| Learning Rate     | 1.6e-4            |
| Global Batch Size | 1200              |
| Micro Batch Size  | 5                 |

#### Speeds, Sizes, Times [optional]

<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->

The training took 14 days to complete.

## Evaluation

<!-- This section describes the evaluation protocols and provides the results. -->

### Testing Data, Factors & Metrics

#### Testing Data

<!-- This should link to a Dataset Card if possible. -->

_Coming soon_

#### Factors

<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->

_Coming soon_

#### Metrics

<!-- These are the evaluation metrics being used, ideally with a description of why. -->

_Coming soon_

### Results

_Coming soon_

#### Summary



## Model Examination [optional]

<!-- Relevant interpretability work for the model goes here -->

[More Information Needed]

## Environmental Impact

<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->

Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).

- **Hardware Type:** [More Information Needed]
- **Hours used:** [More Information Needed]
- **Cloud Provider:** [More Information Needed]
- **Compute Region:** [More Information Needed]
- **Carbon Emitted:** [More Information Needed]

## Technical Specifications [optional]

### Model Architecture and Objective

SEA LION 3B is a decoder model using the MPT architecture.

| Parameter       | Value  |
|-----------------|--------|
| Layers          | 40     |
| d_model         | ?      |
| head_dim        | ?      |
| Vocabulary      | 256000 |
| Sequence Length | 2048   |

### Compute Infrastructure

#### Hardware

SEA LION 3B was trained on AWS EC2 cluster comprising 32 p4d.24xlarge instances, using a total of 256 A100 40GB GPUs.

#### Software

SEA LION 3B was trained using MosaicML Composer using PyTorch FullyShardedDataParallelism (FSDP).

## Citation [optional]

<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->

**BibTeX:**

N/A

**APA:**

N/A

## Glossary [optional]

<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->

N/A

## More Information [optional]

N/A

## The Team

Hamsawardhini Rengarajan<br>
Holy Lovenia<br>
Lam Clarence<br>
Leong Weiqi<br>
Li Yier<br>
Ng Raymond<br>
Ngui Jian Gang<br>
Railey Montalan<br>
Tai Ngee Chia<br>
Tan Choon Meng<br>
Thanh Ngan Nguyen<br>
Teo Jin Howe<br>
Teo Wei Yi<br>
Yeo Yeow Tong<br>
Yong Xianbin<br>
Yosephine<br>
William Tjhi<br>
David Ong Tat-Wee<br>
Darius Liu<br>
Leslie Teo<br>

## Model Card Contact

[ Todo: Get AISG Contact ]