--- license: gemma --- *Gemma-SEA-LION-v4-27B (Base Model) Last updated: 2025-08-20* --- # Model Card for Gemma-SEA-LION-v4-27B Last updated: 2025-08-20 Gemma-SEA-LION-v4-27B is based on Gemma 3 (which supports over 100 languages) and is a multilingual model which has undergone continued pre-training on approximately **500B** tokens across 11 SEA languages: Bahasa Indonesia, Burmese, Chinese, English, Khmer, Lao, Malay, Tagalog, Tamil, Thai and Vietnamese. **SEA-LION** is a collection of Large Language Models (LLMs) which have been pretrained and instruct-tuned for the Southeast Asia (SEA) region. ## Model Details ### Model Description SEA-LION stands for *Southeast Asian Languages In One Network*. We performed continued pre-training in English and SEA languages on Gemma 3 27B IT, a decoder model using the Gemma 3 architecture, to create *Gemma-SEA-LION-v4-27B*. For tokenization, the model employs the default tokenizer used in Gemma 3 27B IT. - **Developed by:** Products Pillar, AI Singapore - **Funded by:** Singapore NRF - **Shared by:** Products Pillar, AI Singapore - **Model type:** Decoder - **Context length:** 128k - **Language(s) (NLP):** Bahasa Indonesia, Burmese, Chinese, English, Khmer, Lao, Malay, Tagalog, Tamil, Thai and Vietnamese - **License:** [Gemma Terms of Use](https://ai.google.dev/gemma/terms) - **Finetuned from model:** [Gemma-3-27B-IT](https://huggingface.co/google/gemma-3-27b-it) ### Model Sources - **Repository:** https://github.com/aisingapore/sealion.git ## Uses ### Out-of-Scope Use 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 claims, damages, or other liabilities arising from the use of the released weights and codes. ## Bias, Risks, and Limitations *The model was not tested for robustness against adversarial prompting.* It is important for users to be aware that our model exhibits certain limitations that warrant consideration. Like many LLMs, the model can hallucinate and occasionally generates irrelevant content, introducing fictional elements that are not grounded in the provided context. Users should also exercise caution in interpreting and validating the model's responses due to the potential inconsistencies. **Limitations** In terms of vision capability, Gemma-SEA-LION-v4-27B has been trained and fine-tuned exclusively on the text back-end. As a result, its vision capabilities are expected to be comparable to those of Gemma 3 IT 27B, and may not exhibit significant improvements or differences in this area. [🤗 google/gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it ) ## How to Get Started with the Model Use the code below to get started with the model. Use the code below to get started with the model using the 🤗 Transformers library. ```python from transformers import pipeline import torch pipe = pipeline( "text-generation", model="aisingapore/Gemma-SEA-LION-v4-27B", device="cuda", torch_dtype=torch.bfloat16 ) messages = [ { "role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}] }, { "role": "user", "content": [ {"type": "text", "text": "Write a poem on southeast asian countries in Indonesian."} ] } ] output = pipe(text=messages, max_new_tokens=200) print(output[0]["generated_text"][-1]["content"]) ``` ## Training Details ### Training Data The dataset comprises Bahasa Indonesia, Burmese, Chinese, English, Khmer, Lao, Malay, Tagalog, Tamil, Thai and Vietnamese languages, collected from a mixture of sources including web data, code, open-source datasets, and synthetically generated datasets, amounting to a total of 500 billion tokens. The 500 billion tokens are sampled from a much larger pool of 1 trillion tokens from open-sourced datasets with the optimal datamix shown below determined by our experiments. | Language | Dataset Name | Total Tokens (B) | Percentage (%) | Total percentage (%) | |-----------------------------------|-------------------------|------------------|----------------|---------------------| | Code | StarCoder (OLMo 2 Version) | 50B | 10 | 10 | | EN | Fineweb-Edu | 80B | 16 | 40 | | | DCLM-OLMo2-HQ | 80B | 16 | | | | Non-CC-EN | 40B | 8 | | | ZH | SEA-LION Pile v1 | 13.5B | 2.7 | 9 | | | Fineweb2 | 13.5B | 2.7 | | | | Fineweb2-HQ | 4.5B | 0.9 | | | VI | SEA-LION Pile v1 | 4.25B | 0.85 | 8.5 | | | SEA-LION Pile v2 | 12.75B | 2.55 | | | | Fineweb2 | 8.5B | 1.7 | | | | Non-CC-VI | 17B | 3.4 | | | ID | SEA-LION Pile v1 | 5.66B | 1.13 | 8.5 | | | SEA-LION Pile v2 | 17B | 3.4 | | | | Fineweb2 | 11.33B | 2.27 | | | | Non-CC-ID | 8.5B | 1.7 | | | TH | SEA-LION Pile v1 | 3.035B | 0.61 | 8.5 | | | SEA-LION Pile v2 | 9.107B | 1.82 | | | | Fineweb2 | 3.035B | 0.61 | | | | WangChanBERTa | 3.035B | 0.61 | | | | Dolmav1 | 3.035B | 0.61 | | | | Non-CC-TH | 21.25B | 4.25 | | | TL, TA, MS, KM, LO and MY | ALL_LANG | 77.5B | 15.5 | 15.5 | Note: - All token counts are counted using Gemma 3 tokenizer. - Pre-training was conducted with batches of 8k token lengths. - SEA-Pile v1 is processed from Common Crawl WET, which is published [here](https://huggingface.co/datasets/aisingapore/sea-lion-pile). The main proportion is from mC4 dataset (corpus [link](https://huggingface.co/datasets/bertin-project/mc4-sampling)). The cutoff date of this version is September 2020. - SEA-Pile v2 is processed from Common Crawl WARC from October 2020 to April 2024. - Tamil news is sourced with permission from [Seithi](https://seithi.mediacorp.sg/) - We utilized 0.5% of synthetically generated datasets for the low-resource language, Khmer. ### Training Procedure #### Training Hyperparameters - **Training regime:** | Hyperparameter | Gemma-SEA-LION-v4-27B | |-------------------|-----------------------| | Precision | bfloat16 | | Optimizer | decoupled_adamw | | Scheduler | CosineAnnealing | | Learning Rate | 4.00E-08 | | Global Batch Size | 1024 | ## Evaluation ### Testing Data, Factors & Metrics #### Testing Data We evaluated Gemma-SEA-LION-v4-27B on general language capabilities. **Testing Data** For the evaluation of general language capabilities, we employed the [SEA-HELM evaluation benchmark](https://arxiv.org/abs/2502.14301) across a variety of tasks. These tasks include Question Answering (QA), Sentiment Analysis (Sentiment), Toxicity Detection (Toxicity), Metaphor Understanding, Translation in both directions (Eng>Lang & Lang>Eng), Abstractive Summarisation (Abssum), Causal Reasoning (Causal), Natural Language Inference (NLI), Linguistic Diagnostics (LINDSEA) and Global-MMLU-Lite. #### Factors All evaluations were run with the model specific generation parameters defined in the model config. Each evaluation comprised of 8 runs with different seeds and the final results were averaged across these runs. For all tasks, the model was expected to provide an answer tag from which the answer was automatically extracted. For tasks where options were provided, the answer should comprise one of the pre-defined options. The evaluation was done **five-shot** with native prompts on a sample of 100-1000 instances for each dataset. #### Metrics The following metrics were used: | Task | Metric | |---------------------------------|-------------------------------| | Sentiment Analysis | Accuracy | | Extractive QA (ID, VI, TH, TA) | ChrF++ | | MCQ-QA (TL, MY, MS) | Accuracy | | Metaphor | Accuracy | | Abstractive Summarisation | Rouge-L | | Translations | MetricX-24 score (with reference) | | Causal Reasoning | Accuracy | | Natural Language Inference | Accuracy | | LINDSEA | Accuracy | | Global MMLU Lite | Accuracy | | Toxicity Detection | Accuracy | ### Results Coming soon. #### Summary TBC ## Environmental Impact 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:** Nvidia H200 140GB GPUs - **Hours used:** 214 hrs - **Cloud Provider:** SMC H200 - **Compute Region:** Singapore - **Carbon Emitted:** appx. 98 kg CO2 e ## More Information This is the repository for the commercial instruction-tuned 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 claims, damages, or other liabilities arising from the use of the released weights and codes. For more info, please contact us at sealion@aisingapore.org ## Team Adithya Venkatadri Hulagadri, Adwin Chan Hok Teng, Anocha Sutaveephamochanon, Brandon Ong Jin Jie, Bryan Siow Wei Kang, David Ong Tat-Wee, Esther Choa Hsueh Mei, Evelyn Tan Chor Phin, Hamsawardhini Rengarajan, Huang Yuli, Jann Railey Estrada Montalan, Jessica Tan Siao Wei, Jonathan Heng, Karthik Nagarajan, Lee Chwan Ren, Leong Wai Yi, Leong Wei Qi, Leslie Teo, Mark Pereira, Muhammad Ridzuan Bin Mokhtar, Ngee Chia Tai, Ngui Jian Gang, Nguyen Thanh Ngan, Nicholas Cheng Zi Yi, Ong Zhi Hao, Peerat Limkonchotiwat, Raymond Ng Boon Cheong, Sajeban Antonyrex, Susanto Yosephine, Tan Choon Meng, Walter Teng Kok Wai, Wayne Lau, William Tjhi Chandra, Yeo Yeow Tong, Yong Xianbin, Liew Rachel, Liu Bing Jie Darius, Teo Wei Yi, Lin Zhou, Roshan Gopalakrishnan, Cuahtemoc Anda, Sri Devi Wijaya and Partha Nandi ## Contact sealion@aisingapore.org