anocha-mcomp commited on
Commit
20f9a0a
·
verified ·
1 Parent(s): 42396c6

Update model card from Confluence: Model Card: Gemma-SEA-LION-v4-4B-VL

Browse files
Files changed (1) hide show
  1. model_card_gemma-sea-lion-v4-4b-vl.md +26 -12
model_card_gemma-sea-lion-v4-4b-vl.md CHANGED
@@ -4,9 +4,7 @@
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
- **Gemma-SEA-LION-v4-4B-VL** is a 4-billion parameter Vision-Language Model (VLM) built upon the gemma-3-4b-it architecture. To ensure **domain adaptation** for the region, the model underwent rigorous post-training on a curated dataset of approximately **8.54 million** instruction-text pairs.
8
-
9
- This extensive post-training instills **multilingual** and **multicultural** fluency, covering English and key SEA languages with a heavy emphasis on Burmese, Malay, Tagalog and Tamil.
10
 
11
  Gemma-SEA-LION-v4-4B-VL inherits the following features from gemma-3-4b-it:
12
 
@@ -20,7 +18,7 @@ Gemma-SEA-LION-v4-4B-VL inherits the following features from gemma-3-4b-it:
20
 
21
  SEA-LION stands for *Southeast Asian Languages In One Network*.
22
 
23
- We performed post-training in English and SEA languages on gemma-3-4b-it, a decoder model using the gemma-3 architecture, and post-training to create Gemma-SEA-LION-v4-4B-VL.
24
 
25
  For tokenization, the model employs the default tokenizer used in gemma-3-4b-it.
26
 
@@ -52,8 +50,7 @@ The model has not been aligned for safety. Developers and users should perform t
52
  Use the code below to get started with the model with 🤗 Transformers libraries.
53
 
54
  ```
55
- pip install transformers>=4.50.0
56
-
57
  ```
58
 
59
  ```
@@ -75,7 +72,7 @@ messages = [
75
  "role": "user",
76
  "content": [
77
  {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
78
- {"type": "text", "text": "Response in Indonesian. What animal is on the candy? Describe in Indonesian and English."}
79
  ]
80
  }
81
  ]
@@ -103,6 +100,15 @@ The instruction fine-tuning text dataset comprises of a collection of OSS & synt
103
  - <https://huggingface.co/datasets/allenai/Dolci-Instruct-SFT-Tool-Use>
104
  - <https://huggingface.co/datasets/Agent-Ark/Toucan-1.5M>
105
 
 
 
 
 
 
 
 
 
 
106
  **Datasets for Reinforcement Learning:**
107
 
108
  - Instruction following: <https://huggingface.co/datasets/nvidia/Nemotron-Cascade-RL-Instruction-Following>
@@ -118,8 +124,6 @@ The instruction fine-tuning text dataset comprises of a collection of OSS & synt
118
 
119
  ### Testing Data, Factors & Metrics
120
 
121
- #### Testing Data
122
-
123
  We evaluated Gemma-SEA-LION-v4-4B-VL on general language capabilities and LLM-specific capabilities using SEA-HELM.
124
 
125
  **Testing Data**
@@ -171,7 +175,17 @@ The following metrics were used for text capabilities:
171
 
172
  ### Results
173
 
174
- For details on Gemma-SEA-LION-v4-4B-VL performance, please refer to the SEA-HELM leaderboard, <https://leaderboard.sea-lion.ai/> .
 
 
 
 
 
 
 
 
 
 
175
 
176
  ### Retaining VL Capabilities
177
 
@@ -204,9 +218,9 @@ For more info, please contact us at [sealion@aisingapore.org](mailto:sealion@ais
204
 
205
  ## Team
206
 
207
- Ahmed Dabeer, 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, 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
208
 
209
- Gemma team??? 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)
210
 
211
  ## Acknowledgement
212
 
 
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
+ **Gemma-SEA-LION-v4-4B-VL** is a 4-billion parameter Vision-Language Model (VLM) built upon the gemma-3-4b-it architecture. To ensure **domain adaptation** for the region, the model underwent rigorous post-training on a curated dataset of approximately **8.54 million** instruction-text pairs. This curated dataset included various SEA Languages such as Burmese, Malay, Tagalog and Tamil. This extensive post-training instills **multilingual** and **multicultural** fluency for both SEA languages and English.
 
 
8
 
9
  Gemma-SEA-LION-v4-4B-VL inherits the following features from gemma-3-4b-it:
10
 
 
18
 
19
  SEA-LION stands for *Southeast Asian Languages In One Network*.
20
 
21
+ We performed post-training in SEA languages and English on gemma-3-4b-it, a decoder model using the gemma-3 architecture, and post-training to create Gemma-SEA-LION-v4-4B-VL.
22
 
23
  For tokenization, the model employs the default tokenizer used in gemma-3-4b-it.
24
 
 
50
  Use the code below to get started with the model with 🤗 Transformers libraries.
51
 
52
  ```
53
+ pip install transformers>=4.50.0
 
54
  ```
55
 
56
  ```
 
72
  "role": "user",
73
  "content": [
74
  {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
75
+ {"type": "text", "text": "Respond and provide a brief description in Indonesian. What animal is on the candy?"}
76
  ]
77
  }
78
  ]
 
100
  - <https://huggingface.co/datasets/allenai/Dolci-Instruct-SFT-Tool-Use>
101
  - <https://huggingface.co/datasets/Agent-Ark/Toucan-1.5M>
102
 
103
+ **Datasets for OCR:**
104
+
105
+ - <https://huggingface.co/datasets/Goekdeniz-Guelmez/unsloth-LaTeX_OCR-for-mlx-vlm> (68,700)
106
+ - <https://huggingface.co/datasets/sushanthr/fineweb_ocr/viewer/default/train?row=0> (150,000)
107
+ - <https://huggingface.co/datasets/sushanthr/wiki_ocr> (50,000)
108
+ - <https://huggingface.co/datasets/Chanrith123333/khmer_english_ocr_image_line> (11.3 million)
109
+ - <https://huggingface.co/datasets/chuuhtetnaing/burmese_ocr_data_hf> (9070)
110
+ - <https://huggingface.co/datasets/SoyVitou/KhmerST-Dataset-OCR-Cropped> (3560)
111
+
112
  **Datasets for Reinforcement Learning:**
113
 
114
  - Instruction following: <https://huggingface.co/datasets/nvidia/Nemotron-Cascade-RL-Instruction-Following>
 
124
 
125
  ### Testing Data, Factors & Metrics
126
 
 
 
127
  We evaluated Gemma-SEA-LION-v4-4B-VL on general language capabilities and LLM-specific capabilities using SEA-HELM.
128
 
129
  **Testing Data**
 
175
 
176
  ### Results
177
 
178
+ For details on Gemma-SEA-LION-v4-4B-VL performance, please refer to the SEA-HELM leaderboard, <https://leaderboard.sea-lion.ai/>.
179
+
180
+ ### Tool calling
181
+
182
+ We evaluated the tool calling capabilities of our model
183
+
184
+ #### Results
185
+
186
+ ### OCR
187
+
188
+ #### Results
189
 
190
  ### Retaining VL Capabilities
191
 
 
218
 
219
  ## Team
220
 
221
+ Ahmed Dabeer, 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, 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, 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)
222
 
223
+ ~~Gemma team??? 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)~~
224
 
225
  ## Acknowledgement
226