Instructions to use anocha-mcomp/lora_model with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use anocha-mcomp/lora_model with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("anocha-mcomp/lora_model", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Unsloth Desktop
Update model card from Confluence: Model Cards: Qwen-SEA-LION-v4-4B-VL
Browse files
model_cards_qwen-sea-lion-v4-4b-vl.md
CHANGED
|
@@ -1,10 +1,10 @@
|
|
| 1 |
# Qwen-SEA-LION-v4-4B-VL
|
| 2 |
|
| 3 |
-
Last update: 2025-
|
| 4 |
|
| 5 |
SEA-LION is a collection of Large Language Models (LLMs) which have been pretrained and instruct-tuned for the Southeast Asia (SEA) region.
|
| 6 |
|
| 7 |
-
**Qwen-SEA-LION-v4-4B-VL** is a 4-billion parameter Vision-Language Model (VLM) built upon the Qwen3-VL-4B-Instruct architecture. To ensure **domain adaptation** for the region, the model underwent rigorous supervised fine-tuning (SFT) on a curated dataset of approximately **
|
| 8 |
|
| 9 |
Qwen-SEA-LION-v4-4B-VL inherits the following features from Qwen3-VL:
|
| 10 |
|
|
@@ -19,7 +19,7 @@ Qwen-SEA-LION-v4-4B-VL inherits the following features from Qwen3-VL:
|
|
| 19 |
|
| 20 |
SEA-LION stands for *Southeast Asian Languages In One Network*.
|
| 21 |
|
| 22 |
-
We performed post-training in English and SEA languages on Qwen3-VL-4B-Instruct, a decoder model using the Qwen3 architecture,
|
| 23 |
|
| 24 |
For tokenization, the model employs the default tokenizer used in Qwen3-VL.
|
| 25 |
|
|
@@ -28,14 +28,13 @@ For tokenization, the model employs the default tokenizer used in Qwen3-VL.
|
|
| 28 |
- **Shared by:** AI Products Pillar, AI Singapore
|
| 29 |
- **Model type:** Decoder
|
| 30 |
- **Context length:** 256k
|
| 31 |
-
- **Language(s):** fine-tuned on Burmese, Indonesian,
|
| 32 |
- **License:** [Apache-2.0](https://choosealicense.com/licenses/apache-2.0/)
|
| 33 |
- **Finetuned from model:** [Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct)
|
| 34 |
|
| 35 |
### Model Sources
|
| 36 |
|
| 37 |
-
- **Repository:**
|
| 38 |
-
- **Demo:** [playground.sea-lion.ai](https://playground.sea-lion.ai/chat)
|
| 39 |
|
| 40 |
## Uses
|
| 41 |
|
|
@@ -168,25 +167,23 @@ The following metrics were used for text capabilities:
|
|
| 168 |
| Natural Language Inference | Accuracy |
|
| 169 |
| LINDSEA | Accuracy |
|
| 170 |
| Global MMLU Lite | Accuracy |
|
| 171 |
-
|
|
| 172 |
| Kalahi | Accuracy |
|
| 173 |
| SEA-IFEval | Accuracy |
|
| 174 |
| SEA-MTBench | Win rate against a reference |
|
| 175 |
|
| 176 |
### Results
|
| 177 |
|
| 178 |
-
<
|
| 179 |
-
|
| 180 |
-
For details on Qwen-SEA-LION-v4-4B-VL performance, please refer to the SEA-HELM leaderboard, <https://leaderboard.sea-lion.ai/>.
|
| 181 |
|
| 182 |
### Retaining VL Capabilities
|
| 183 |
|
| 184 |
-
We also evaluated
|
| 185 |
|
| 186 |
-
- Visual Question Answering (VQA): We
|
| 187 |
- Image Captioning: We employed the XM3600 dataset, evaluating strictly on examples relevant to the SEA region.
|
| 188 |
|
| 189 |
-
Key Insight: Despite our fine-tuning process focusing primarily on text data (approximately 8 million regional Q&A and instruction pairs), our
|
| 190 |
|
| 191 |
#### Factors
|
| 192 |
|
|
@@ -196,12 +193,12 @@ The evaluation was done **zero-shot** with native prompts.
|
|
| 196 |
|
| 197 |
The following metrics were used to measure performance:
|
| 198 |
|
| 199 |
-
- **
|
| 200 |
- **RefCLIP Score** was used for the XM3600 image captioning task.
|
| 201 |
|
| 202 |
### Results
|
| 203 |
|
| 204 |
-
[Attachment: image-20251126-101525.png]The VL capabilities were retained despite training using text only
|
| 205 |
|
| 206 |
## More Information
|
| 207 |
|
|
@@ -211,7 +208,7 @@ For more info, please contact us at [sealion@aisingapore.org](mailto:sealion@ais
|
|
| 211 |
|
| 212 |
## Team
|
| 213 |
|
| 214 |
-
Ahn Jeongmi, Antonyrex Sajeban, Chan Hok Teng Adwin, Cheng Zi Yi Nicholas, Choa Hsueh Mei Esther, Heng Jonathan, Huang Yuli,
|
| 215 |
|
| 216 |
## Contact
|
| 217 |
|
|
|
|
| 1 |
# Qwen-SEA-LION-v4-4B-VL
|
| 2 |
|
| 3 |
+
Last update: 2025-12-4
|
| 4 |
|
| 5 |
SEA-LION is a collection of Large Language Models (LLMs) which have been pretrained and instruct-tuned for the Southeast Asia (SEA) region.
|
| 6 |
|
| 7 |
+
**Qwen-SEA-LION-v4-4B-VL** is a 4-billion parameter Vision-Language Model (VLM) built upon the Qwen3-VL-4B-Instruct architecture. To ensure **domain adaptation** for the region, the model underwent rigorous supervised fine-tuning (SFT) on a curated dataset of approximately **9 million** instruction-text pairs. This extensive post-training instills **multilingual** and **multicultural** fluency, covering English and 7 key SEA languages: Burmese, Indonesian, Filipino, Malay, Tamil, Thai, and Vietnamese.
|
| 8 |
|
| 9 |
Qwen-SEA-LION-v4-4B-VL inherits the following features from Qwen3-VL:
|
| 10 |
|
|
|
|
| 19 |
|
| 20 |
SEA-LION stands for *Southeast Asian Languages In One Network*.
|
| 21 |
|
| 22 |
+
We performed post-training in English and SEA languages on Qwen3-VL-4B-Instruct, a decoder model using the Qwen3 architecture, to create Qwen-SEA-LION-v4-4B-VL.
|
| 23 |
|
| 24 |
For tokenization, the model employs the default tokenizer used in Qwen3-VL.
|
| 25 |
|
|
|
|
| 28 |
- **Shared by:** AI Products Pillar, AI Singapore
|
| 29 |
- **Model type:** Decoder
|
| 30 |
- **Context length:** 256k
|
| 31 |
+
- **Language(s):** fine-tuned on Burmese, Indonesian, Filipino, Malay, Tamil, Thai, and Vietnamese
|
| 32 |
- **License:** [Apache-2.0](https://choosealicense.com/licenses/apache-2.0/)
|
| 33 |
- **Finetuned from model:** [Qwen3-VL-4B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-4B-Instruct)
|
| 34 |
|
| 35 |
### Model Sources
|
| 36 |
|
| 37 |
+
- **Repository:** [SEA-LION v4 - an aisingapore Collection](https://huggingface.co/collections/aisingapore/sea-lion-v4)
|
|
|
|
| 38 |
|
| 39 |
## Uses
|
| 40 |
|
|
|
|
| 167 |
| Natural Language Inference | Accuracy |
|
| 168 |
| LINDSEA | Accuracy |
|
| 169 |
| Global MMLU Lite | Accuracy |
|
| 170 |
+
| ThaiExam | Accuracy |
|
| 171 |
| Kalahi | Accuracy |
|
| 172 |
| SEA-IFEval | Accuracy |
|
| 173 |
| SEA-MTBench | Win rate against a reference |
|
| 174 |
|
| 175 |
### Results
|
| 176 |
|
| 177 |
+
[Attachment: 8B & 4B screenshot.png]For details on Qwen-SEA-LION-v4-4B-VL performance, please refer to the <https://leaderboard.sea-lion.ai/>.
|
|
|
|
|
|
|
| 178 |
|
| 179 |
### Retaining VL Capabilities
|
| 180 |
|
| 181 |
+
We also evaluated our models on two types of tasks using datasets specifically focused on Southeast Asian examples to benchmark and compared our models' performances against the original base models (Qwen3-VL-4B/8B).
|
| 182 |
|
| 183 |
+
- Visual Question Answering (VQA): We utilised Multiple Choice Question (MCQ) style tasks, including MARVL, CVQA, and WorldCuisines.
|
| 184 |
- Image Captioning: We employed the XM3600 dataset, evaluating strictly on examples relevant to the SEA region.
|
| 185 |
|
| 186 |
+
Key Insight: Despite our fine-tuning process focusing primarily on text data (approximately 8 million regional Q&A and instruction pairs), our evaluations confirm that Qwen-SEA-LION-v4 (4B/8B) successfully retains the high-performance vision-language capabilities of the original base models.
|
| 187 |
|
| 188 |
#### Factors
|
| 189 |
|
|
|
|
| 193 |
|
| 194 |
The following metrics were used to measure performance:
|
| 195 |
|
| 196 |
+
- **Normalised accuracy** was the primary metric for the VQA tasks (CVQA, MARVL, and WorldCuisines).
|
| 197 |
- **RefCLIP Score** was used for the XM3600 image captioning task.
|
| 198 |
|
| 199 |
### Results
|
| 200 |
|
| 201 |
+
[Attachment: image-20251126-101525.png]The VL capabilities were retained despite training using text only
|
| 202 |
|
| 203 |
## More Information
|
| 204 |
|
|
|
|
| 208 |
|
| 209 |
## Team
|
| 210 |
|
| 211 |
+
Ahn Jeongmi, Antonyrex Sajeban, Chan Hok Teng Adwin, Cheng Zi Yi Nicholas, Choa Hsueh Mei Esther, Heng Jonathan, Huang Yuli, Jann Railey Estrada Montalan, Kang Siow Wei Bryan, Lau Wayne, Lee Chwan Ren, Leong Wai Yi, Leong Wei Qi, Liew Rachel, Limkonchotiwat Peerat, Muhammad Ridzuan Bin Mokhtar, Nagarajan Karthik, Ng Boon Cheong Raymond, Ngee Chia Tai, Ngui Jian Gang, Nguyen Thanh Ngan, Ong Tat-Wee David, Ong Zhi Hao, Pereira Mark, Poon Joseph, Rengarajan Hamsawardhini, 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
|
| 212 |
|
| 213 |
## Contact
|
| 214 |
|