|
--- |
|
base_model: |
|
- aisingapore/gemma2-9b-cpt-sea-lionv3-base |
|
language: |
|
- en |
|
- id |
|
- jv |
|
- su |
|
license: gemma |
|
--- |
|
# Gemma2 9B CPT Sahabat AI v1 |
|
|
|
Sahabat AI v1 is a collection of Large Language Models (LLMs) which has been pretrained and instruct-tuned for Indonesian languages. |
|
This is the card for the Gemma2 9B CPT Sahabat AI v1 base model which has undergone continued pre-training from the base [Gemma2 9B CPT SEA-LIONv3 base](https://huggingface.co/aisingapore/gemma2-9b-cpt-sea-lionv3-base) model. |
|
|
|
Sahabat is Indonesian for "Close Friends." |
|
|
|
|
|
## Model Details |
|
|
|
### Model Description |
|
|
|
The continued pre-training data for Gemma2 9B CPT Sahabat AI v1 base model encompasses approximately 50B tokens. |
|
|
|
- **Developed by:** PT GoTo Gojek Tokopedia Tbk, AI Singapore |
|
- **Funded by:** PT GoTo Gojek Tokopedia Tbk, AI Singapore |
|
- **Model type:** Decoder |
|
- **Languages:** English, Indonesian, Javanese, Sundanese |
|
- **License:** [Gemma Community License](https://ai.google.dev/gemma/terms) |
|
|
|
For tokenisation, the model employs the default tokenizer used in Gemma-2-9B. The model has a context length of 8192. |
|
|
|
### Benchmark Performance |
|
We evaluated Gemma2 9B CPT Sahabat AI v1 base model on general language capabilities. |
|
|
|
#### General Language Capabilities |
|
For the evaluation of general language capabilities, we employed the |
|
- [SEA HELM (also known as BHASA) evaluation benchmark](https://arxiv.org/abs/2309.06085v2) across a variety of tasks. |
|
- These tasks include Question Answering (QA), Sentiment Analysis (Sentiment), Toxicity Detection (Toxicity), Translation in both directions (Eng>Lang & Lang>Eng), Abstractive Summarization (Summ), Causal Reasoning (Causal) and Natural Language Inference (NLI). |
|
- We also added support for Javanese and Sundanese for the BHASA tasks whenever applicable |
|
- [IndoMMLU](https://arxiv.org/pdf/2310.04928) |
|
- These tasks include examination questions on Humanities, Indonesian language, Local languages and cultures, Social science and STEM across primary, middle, and high school levels. |
|
- and the well known [English MMLU](https://arxiv.org/pdf/2009.03300) |
|
|
|
Note: SEA HELM is implemented using prompts to elicit answers in a strict format. For all tasks, the model is expected to provide an answer tag from which the answer is automatically extracted. For tasks where options are provided, the answer should comprise one of the pre-defined options. The scores for each task is normalised to account for baseline performance due to random chance. |
|
|
|
The evaluation was done **five-shot** with native prompts on a sample of 100-1000 instances for each dataset. |
|
|
|
|
|
## Training Details |
|
|
|
### Data |
|
|
|
Gemma2 9B CPT Sahabat AI v1 base model was continued pre-trained on 50B tokens of the following data: |
|
|
|
| Data Source | Unique Tokens (B) | Multiplier | Total Tokens (B) | Percentage (%)| |
|
|---------------------------------------|:-----------------:|:----------:|:----------------:|:-------------:| |
|
| Dolma Refined Web | 9.5 | 1 | 9.5 | 17.36 | |
|
| Dolma arXiv | 0.6 | 1 | 0.6 | 1.10 | |
|
| Dolma Star Coder | 5.5 | 1 | 5.5 | 10.05 | |
|
| Dolma Semantic Scholar | 1.2 | 1 | 1.2 | 2.19 | |
|
| Dolma Reddit | 1.7 | 1 | 1.7 | 3.11 | |
|
| Dolma C4 | 1.5 | 1 | 1.4 | 2.56 | |
|
| Wiki* + News* - Indonesian | 1.3 | 4 | 5.2 | 9.50 | |
|
| SEA-LION Pile - Indonesian | 27.0 | 1 | 27.0 | 49.34 | |
|
| SEA-LION Pile - Javanese | 0.5 | 1.5 | 0.75 | 1.37 | |
|
| CC 100 - Javanese | 0.05 | 1.5 | 0.075 | 0.14 | |
|
| HPLT - Javanese | 0.3 | 1.5 | 0.45 | 0.82 | |
|
| SEA-LION Pile - Sundanese | 0.2 | 3.6 | 0.75 | 1.37 | |
|
| CC 100 - Sundanese | 0.02 | 3.6 | 0.075 | 0.14 | |
|
| HPLT - Sundanese | 0.16 | 3.6 | 0.45 | 0.82 | |
|
| Others (Javanese, Sundanese) | 0.034 | 2.2 | 0.076 | 0.14 | |
|
|
|
Note: |
|
- All token counts are counted using Gemma2 tokenizer |
|
- Wiki* sources includes Wikipedia, Wiki Books, Wiki Source, Wiki Voyage and Fandom Wiki |
|
- News* sources includes VOA, Global Voices, MediaCorp |
|
|
|
### Infrastructure |
|
|
|
Gemma2 9B CPT Sahabat AI v1 was trained using [MosaicML Composer](https://github.com/mosaicml/composer) |
|
on the following hardware: |
|
|
|
| Training Details | Gemma2 9B CPT Sahabat AI v1| |
|
|----------------------|:--------------------------:| |
|
| Nvidia H100 80GB GPU | 32 | |
|
| Training Duration | 7 days | |
|
|
|
|
|
### Configuration |
|
|
|
| HyperParameter | Gemma2 9B CPT Sahabat AI v1| |
|
|-------------------|:--------------------------:| |
|
| Precision | bfloat16 | |
|
| Optimizer | decoupled_adamw | |
|
| Scheduler | weight_stable_decay | |
|
| Learning Rate | 1.0e-5 | |
|
| Global Batch Size | 256 | |
|
| Micro Batch Size | 1 | |
|
|
|
|
|
## The Team (by ascending alphabetical order) |
|
|
|
### AI Singapore |
|
Chan Adwin<br> |
|
Chau Shiau Ching<br> |
|
Cheng Nicholas<br> |
|
Choa Esther<br> |
|
Huang Yuli<br> |
|
Lau Wayne<br> |
|
Lee Chwan Ren<br> |
|
Leong Wai Yi<br> |
|
Leong Wei Qi<br> |
|
Limkonchotiwat Peerat<br> |
|
Liu Bing Jie Darius<br> |
|
Montalan Jann Railey<br> |
|
Ng Boon Cheong Raymond<br> |
|
Ngui Jian Gang<br> |
|
Nguyen Thanh Ngan<br> |
|
Ong Brandon<br> |
|
Ong Tat-Wee David<br> |
|
Ong Zhi Hao<br> |
|
Rengarajan Hamsawardhini<br> |
|
Siow Bryan<br> |
|
Susanto Yosephine<br> |
|
Tai Ngee Chia<br> |
|
Tan Choon Meng<br> |
|
Teng Walter<br> |
|
Teo Eng Sipp Leslie<br> |
|
Teo Wei Yi<br> |
|
Tjhi William<br> |
|
Yeo Yeow Tong<br> |
|
Yong Xianbin<br> |
|
|
|
### PT GoTo Gojek Tokopedia Tbk |
|
Anissa Dininta<br> |
|
Choiri Hendra Hadhil<br> |
|
Goel Priyank<br> |
|
Saini Ajay Kumar<br> |
|
Shalev Ofir<br> |
|
Tan Daryl<br> |
|
Tep Kilian Rithi<br> |
|
Tiwari Anupam<br> |
|
Widjojo Daniel<br> |
|
|
|
<!-- |
|
## 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 [Sahabat Inquiry Form.](https://docs.google.com/forms/d/1_us969eQtEooYOn4XkvGkdP5VHOyCbO6L_sd9kTMnaA/edit) |
|
|
|
|
|
## Disclaimer |
|
|
|
This is the repository for the base model. |
|
The model has _not_ been aligned for safety. |
|
Developers and users should perform their own safety fine-tuning and related security measures. |
|
In no event shall the authors be held liable for any claim, damages, or other liability arising from the use of the released weights and codes. |
|
|
|
|
|
## References |
|
### IndoMMLU Reference |
|
|
|
```bibtex |
|
@inproceedings{koto-etal-2023-indommlu, |
|
title = "Large Language Models Only Pass Primary School Exams in {I}ndonesia: A Comprehensive Test on {I}ndo{MMLU}", |
|
author = "Fajri Koto and Nurul Aisyah and Haonan Li and Timothy Baldwin", |
|
booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP)", |
|
month = December, |
|
year = "2023", |
|
address = "Singapore", |
|
publisher = "Association for Computational Linguistics", |
|
} |
|
} |
|
``` |
