Qwen-SEA-LION-v4-32B-IT-4BIT (GPTQ model)
Last update: 2025-10-17
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 17 Oct 2025, Qwen-SEA-LION-v4-32B-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. Qwen-SEA-LION-v4-32B-IT was quantized to create Qwen-SEA-LION-v4-32B-IT-4BIT. The quantized version has little degradation (<0.3% on average) in performance compared to Qwen-SEA-LION-v4-32B-IT (for detailed rankings, please refer to the leaderboard), and Qwen-SEA-LION-v4-32B-IT-4BIT version can run on a laptop.
Qwen-SEA-LION-v4-32B-IT inherits the following features from Qwen3-32B:
- 32,768 of context length natively
Model Details
Model Description
SEA-LION stands for Southeast Asian Languages In One Network.
Quantization was performed on Qwen-SEA-LION-v4-32B-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 Qwen3-32B.
- Developed by: AI Products Pillar, AI Singapore
- Funded by: Singapore NRF
- Shared by: AI Products Pillar, AI Singapore
- Model type: Decoder
- Context Length: 32k tokens
- Language(s) (NLP): Burmese, English, Indonesian, Khmer, Lao, Malay, Mandarin, Tagalog, Tamil, Thai, and Vietnamese
- License: MIT
- Continue pretrained from model: Qwen-3-32B
Uses
Available Quantized Versions
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
Caveats || Risks
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, Qwen-SEA-LION-v4-32B-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 Qwen3 32B and may not exhibit significant improvements or differences in this area. (https://huggingface.co/Qwen/Qwen3-32B)
How to Get Started with the Model
Use the code below to get started with the model using the 🤗 Transformers library.
The model defaults to non-thinking mode. To enable thinking mode, please use
enable_thinking=True.
from transformers import AutoModelForCausalLM, AutoTokenizer
model_name = "aisingapore/Qwen-SEA-LION-v4-32B-IT-4BIT"
# load the tokenizer and the model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype="auto",
device_map="auto"
)
# prepare the model input
prompt = "Give me a short introduction to large language model."
messages = [
{"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=True # Switches between thinking and non-thinking modes. Default is True.
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
# conduct text completion
generated_ids = model.generate(
**model_inputs,
max_new_tokens=32768
)
output_ids = generated_ids[0][len(model_inputs.input_ids[0]):].tolist()
# parsing thinking content
try:
# rindex finding 151668 ()
index = len(output_ids) - output_ids[::-1].index(151668)
except ValueError:
index = 0
thinking_content = tokenizer.decode(output_ids[:index], skip_special_tokens=True).strip("\n")
content = tokenizer.decode(output_ids[index:], skip_special_tokens=True).strip("\n")
print("thinking content:", thinking_content)
print("content:", content)
Training Details
Training Data
The dataset comprises Burmese, Indonesian, 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 100 billion tokens sampled.
Evaluation
Model Performance
For details on Qwen-SEA-LION-v4-32B-IT performance, please refer to the SEA-HELM leaderboard, https://leaderboard.sea-lion.ai/ .
Resource Metrics
| Quantized Variant | Num. Parameters | Storage Size (GB) | VRAM Required (GB) | Time to First Token (s) | Tokens per Second |
|---|---|---|---|---|---|
| BF16 | 32B | 65.57 | 61.03 | 2.20 | 52.98 |
| 8-bit (GPTQ) | 32B | 34.34 | 32.04 | 0.35 | 68.85 |
| 4-bit (GPTQ) | 32B | 19.93 | 19.43 | 0.34 | 78.20 |
Additional Remarks:
- TTFT and Tokens per Sec: 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
- Input size 4K, output 1K
- Tests conducted on a system with an NVIDIA H200 GPU
- For more details on the quantized model, please refer to config.json
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 using this sealion@aisingapore.org
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.
Team
Ahn Jeongmi, 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, 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, Siow Wei Kang Bryan, Susanto Yosephine, Sutaveephamochanon Anocha, Tan Choon Meng, Tan Chor Phin Evelyn, Tan Siao Wei Jessica, Tan Yixian, Tee Jun Yun, Teng Kok Wai Walter, Teo Eng Sipp Leslie, Tjhi William, Yeo Yeow Tong, Yong Xianbin, Zhang Zhou, Liew Rachel, Liu Bing Jie Darius, Tep Kilian Rithi (GoTo)