Instructions to use SEACrowd/SEA-LION-VL-100226 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SEACrowd/SEA-LION-VL-100226 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="SEACrowd/SEA-LION-VL-100226")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForImageTextToText

processor = AutoProcessor.from_pretrained("SEACrowd/SEA-LION-VL-100226")
model = AutoModelForImageTextToText.from_pretrained("SEACrowd/SEA-LION-VL-100226")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use SEACrowd/SEA-LION-VL-100226 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SEACrowd/SEA-LION-VL-100226"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SEACrowd/SEA-LION-VL-100226",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/SEACrowd/SEA-LION-VL-100226

SGLang

How to use SEACrowd/SEA-LION-VL-100226 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "SEACrowd/SEA-LION-VL-100226" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SEACrowd/SEA-LION-VL-100226",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "SEACrowd/SEA-LION-VL-100226" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SEACrowd/SEA-LION-VL-100226",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use SEACrowd/SEA-LION-VL-100226 with Docker Model Runner:
```
docker model run hf.co/SEACrowd/SEA-LION-VL-100226
```

See axolotl config

axolotl version: 0.12.1

base_model: aisingapore/Gemma-SEA-LION-v4-27B-IT
# Automatically upload checkpoint and final model to HF
# hub_model_id: username/custom_model_name

skip_prepare_dataset: false
remove_unused_columns: false
sample_packing: false

ddp_find_unused_parameters: true
deepspeed: deepspeed_configs/zero3.json

chat_template: gemma3
eot_tokens:
  - <end_of_turn>
datasets:
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_textcaps_EN_SEA
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xm3600_EN_SEA
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_1
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_2
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_3
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_4
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_5
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_6
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_7
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_8
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_9
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_xflickr_EN_SEA_chunk_10
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_1
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_2
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_3
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_4
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_5
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_6
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_7
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_8
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_9
    type: chat_template
    split: train
    field_messages: messages
  - path: /mnt/weka/all/holy/translation_parallel/pretrain_dataset_sea_vl_EN_SEA_chunk_10
    type: chat_template
    split: train
    field_messages: messages
dataset_prepared_path: peerat_test_path
val_set_size: 0.01
output_dir: ./outputs-pt/sealion-v4-gemma-3-27b-hero_pre_train_v2

sequence_len: 2048
pad_to_sequence_len: false

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 4
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 2e-5

bf16: true
fp16:
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
resume_from_checkpoint:
logging_steps: 1
flash_attention: false
# attn_implementation: sdpa

warmup_ratio: 0.1
evals_per_epoch: 1
save_steps: 1000
save_total_limit: 3
weight_decay: 0.0

# save_first_step: true  # uncomment this to validate checkpoint saving works with your config

outputs-pt/sealion-v4-gemma-3-27b-hero_pre_train_v2

This model is a fine-tuned version of aisingapore/Gemma-SEA-LION-v4-27B-IT on the None dataset.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 80
gradient_accumulation_steps: 2
total_train_batch_size: 640
total_eval_batch_size: 320
optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 1156
training_steps: 11568