Instructions to use FriendliAI/deepseek-vl2-tiny with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use FriendliAI/deepseek-vl2-tiny with Transformers:

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

pipe = pipeline("image-text-to-text", model="FriendliAI/deepseek-vl2-tiny")
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 AutoModel
model = AutoModel.from_pretrained("FriendliAI/deepseek-vl2-tiny", dtype="auto")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use FriendliAI/deepseek-vl2-tiny with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "FriendliAI/deepseek-vl2-tiny"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "FriendliAI/deepseek-vl2-tiny",
		"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/FriendliAI/deepseek-vl2-tiny

SGLang

How to use FriendliAI/deepseek-vl2-tiny 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 "FriendliAI/deepseek-vl2-tiny" \
    --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": "FriendliAI/deepseek-vl2-tiny",
		"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 "FriendliAI/deepseek-vl2-tiny" \
        --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": "FriendliAI/deepseek-vl2-tiny",
		"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 FriendliAI/deepseek-vl2-tiny with Docker Model Runner:
```
docker model run hf.co/FriendliAI/deepseek-vl2-tiny
```

deepseek-vl2-tiny

File size: 1,226 Bytes

e2d480f

{
  "add_special_token": false,
  "candidate_resolutions": [
    [
      384,
      384
    ],
    [
      384,
      768
    ],
    [
      768,
      384
    ],
    [
      384,
      1152
    ],
    [
      1152,
      384
    ],
    [
      384,
      1536
    ],
    [
      1536,
      384
    ],
    [
      768,
      768
    ],
    [
      384,
      1920
    ],
    [
      1920,
      384
    ],
    [
      384,
      2304
    ],
    [
      2304,
      384
    ],
    [
      768,
      1152
    ],
    [
      1152,
      768
    ],
    [
      384,
      2688
    ],
    [
      2688,
      384
    ],
    [
      384,
      3072
    ],
    [
      3072,
      384
    ],
    [
      768,
      1536
    ],
    [
      1536,
      768
    ],
    [
      384,
      3456
    ],
    [
      3456,
      384
    ],
    [
      1152,
      1152
    ]
  ],
  "downsample_ratio": 2,
  "ignore_id": -100,
  "image_mean": [
    0.5,
    0.5,
    0.5
  ],
  "image_std": [
    0.5,
    0.5,
    0.5
  ],
  "image_token": "<image>",
  "mask_prompt": false,
  "normalize": true,
  "pad_token": "<\uff5c\u2581pad\u2581\uff5c>",
  "patch_size": 14,
  "processor_class": "DeepseekVLV2Processor",
  "sft_format": "deepseek"
}