Update README.md

README.md

@@ -31,7 +31,8 @@ Last updated: 2026-02-04
 
 SEA-LION stands for *Southeast Asian Languages In One Network*. 
 
-This model is a fine-tuned version of [Gemma 3 12B IT](https://huggingface.co/google/gemma-3-12b-it) on 1M instruction-following pairs. For more details on training data, please refer to the paper [SEA-Guard]().
+This model is a fine-tuned version of [Gemma 3 12B IT](https://huggingface.co/google/gemma-3-12b-it) on 1M instruction-following pairs. 
+For more details on training data, please refer to the paper [SEA-Guard](https://arxiv.org/abs/2602.01618).
 
 For tokenization, the model employs the default tokenizer used in Gemma 3.
 - **Developed by:** AI Products Pillar, AI Singapore
@@ -128,7 +129,7 @@ print(prompt,decoded)
 
 ## Training and evaluation data
 
-For more details on training data, please refer to the paper [SEA-Guard]().
+For more details on training data, please refer to the paper [SEA-Guard](https://arxiv.org/abs/2602.01618).
 
 ## Training procedure