Model Card for Gemma-SEA-LION-v4-27B-IT-NVFP4

Last updated: 2025-08-25

SEA-LION is a collection of Large Language Models (LLMs) which have been pretrained and instruct-tuned for the Southeast Asia (SEA) region.

As of 25 Aug 2025, Gemma-SEA-LION-v4-27B-IT excels at Southeast Asian (SEA) tasks when compared to other open models with fewer than 200 billion parameters and demonstrates performance comparable to that of larger and top closed models. Gemma-SEA-LION-v4-27B-IT was quantized to create Gemma-SEA-LION-v4-27B-IT-NVFP4. Gemma-SEA-LION-v4-27B-IT-NVFP4 has little degradation (<0.5%) in performance compared to Gemma-SEA-LION-v4-27B-IT (for detailed rankings, please refer to the leaderboard), and this version can run on a laptop with vLLM and CUDA-compatible GPU.

Gemma-SEA-LION-v4-27B-IT-NVFP4 inherits Gemma 3’s:

• Large 128K context length

• Image and text understanding capabilities, including document comprehension, visual Q&A, and image-grounded reasoning

• Advanced function calling and structured outputs to allow for seamless integration into larger systems

Model Description

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

Quantization was performed on Gemma-SEA-LION-v4-27B-IT to produce optimized variants that reduce memory requirements while maintaining model quality. These quantized models support inference on a range of consumer-grade GPUs and are compatible with various inference engines.

For tokenization, the model employs the default tokenizer used in Gemma 3 27B Instruct.

• Developed by: AI Products Pillar, AI Singapore
• Funded by: Singapore NRF
• Model type: Decoder
• Context length: 128k tokens
• Language(s) (NLP): Burmese, English, Indonesian, Khmer, Lao, Malay, Mandarin, Tagalog, Tamil, Thai and Vietnamese
• License: Gemma Terms of Use
• Quantized from model: Gemma-SEA-LION-v4-27B-IT

Model Weights included in this repository:

• Gemma-SEA-LION-v4-27B-IT-NVFP4

How to Get Started with the Model

Use the code from OpenAI-Compatible Server - vLLM for your personal hardware setting.

Training Details

Training Data

The dataset comprises Burmese, English, Indonesian, Khmer, Lao, Malay, Mandarin, 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 sampled from our bucket of 1 trillion tokens.

Evaluation

Testing Data, Factors & Metrics

Testing Data

We evaluated Gemma-SEA-LION-v4-27B-IT-NVFP4 on general language, multi-turn chat and instruction-following capabilities.

Testing Data

General language capabilities

For the evaluation of general language capabilities, we employed the SEA-HELM evaluation benchmark 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 Summarisation (Abssum), Causal Reasoning (Causal), Natural Language Inference (NLI), Linguistic Diagnostics (LINDSEA), Cultural Knowledge (Kalahi) and Global MMLU Lite.

Instruction-following and Multi-turn Chat

We evaluated the models on instruction-following and multi-turn chat capabilities with SEA-IFEval (based on IFEval) and SEA-MTBench (based on MT-Bench) respectively. The two datasets were originally in English, the linguists and native speakers in the team worked together to filter, localise and translate the datasets into the respective target languages to ensure that the examples remained reasonable, meaningful and natural.

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 zero-shot with native prompts on a sample of 100-1000 instances for each dataset.

SEA-IFEval

SEA-IFEval evaluates a model's ability to adhere to constraints provided in the prompt, for example beginning a response with a specific word/phrase or answering with a certain number of sections. Additionally, accuracy is normalised by the proportion of responses in the correct language (if the model performs the task correctly but responds in the wrong language, it is judged to have failed the task).

SEA-MTBench

SEA-MTBench evaluates a model's ability to engage in multi-turn (2 turns) conversations and respond in ways that align with human needs. We use gpt-4.1-2025-04-14 as the judge model and compare against gpt-4.1-2025-04-14 as the baseline model. The metric used is the weighted win rate against the baseline model (i.e. average win rate across each category: Math, Reasoning, STEM, Humanities, Roleplay, Writing, Extraction).

Metrics

The following metrics were used:

Results

For details on Gemma-SEA-LION-v4-27B-IT-NVFP4 performance, please refer to the SEA-HELM leaderboard, Leaderboard results on SEA-HELM.

Additional remarks:

• SEA-HELM evaluation suite run on one single H200

• Default model parameters as listed in the model card used for evaluation

• Models served through vLLM

Model Performance Across Different GPU Architectures

Additional Remarks:

• TTFT and Tokens per Second: measured with vLLM on localhost and concurrency = 1.

• Reported results are the median (p50) values, calculated across 10 requests.

• Model size taken from vLLM upon loading

• Tests conducted on a system with a single GPU Chip.

• Context Length: 8,096-8,192 tokens maximum, output ~ 1K

• Prompt: Complex multilingual technical explanation (768 tokens)

• FP8 and NVFP4 is created using https://github.com/vllm-project/llm-compressor.git

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-IT 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

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.

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 the National Research Foundation or the National University of Singapore.

For more info, please contact us at sealion@aisingapore.org

Team

Antonyrex Sajeban, Chan Hok Teng Adwin, Cheng Zi Yi Nicholas, Choa Hsueh Mei Esther, Heng Jonathan, Huang Yuli, Hulagadri Adithya Venkatadri, Jann Railey Estrada Montalan, Kang Siow Wei Bryan, Lau Wayne, Lee Chwan Ren, Leong Wai Yi, Leong Wei Qi, Limkonchotiwat Peerat, Muhammad Ridzuan Bin Mokhtar, Nagarajan Karthik, Ng Boon Cheong Raymond, Ngee Chia Tai, Ngui Jian Gang, Nguyen Thanh Ngan, Ong Jin Jie Brandon, Ong Tat-Wee David, Ong Zhi Hao, Pereira Mark, Rengarajan Hamsawardhini, Susanto Yosephine, Sutaveephamochanon Anocha, Tan Choon Meng, Tan Chor Phin Evelyn, Tan Siao Wei Jessica, Teng Kok Wai Walter, Teo Eng Sipp Leslie, Tjhi William, Yeo Yeow Tong, Yong Xianbin, Liew Rachel, Liu Bing Jie Darius, Teo Wei Yi, Zhou Lin (NCS), Gopalakrishnan Roshan (NCS), Anda Cuahtemoc (NCS), Sri Devi Wijaya (NCS), Nandi Partha (NCS), Elliott Chris (Google), Mohseni Mohammadreza (Google), Sharan Mayank (Google), Wei Fanny (Google), Tang Jiuqiang (Google), Xu Xiang (Google), Yu Ting (Google), Loh Michelle (Google), Mangal Saurabh (Google), Mukherjee Pratyusha (Google), Sim Stephanie (Google)

Acknowledgement

This project is supported by the National Research Foundation Singapore and Infocomm Media Development Authority (IMDA), Singapore under its National Large Language Model Funding Initiative.

Contact

sealion@aisingapore.org