deepseek-vl2#

  • Context Length: 4096

  • Model Name: deepseek-vl2

  • Languages: en, zh

  • Abilities: chat, vision

  • Description: DeepSeek-VL2, an advanced series of large Mixture-of-Experts (MoE) Vision-Language Models that significantly improves upon its predecessor, DeepSeek-VL. DeepSeek-VL2 demonstrates superior capabilities across various tasks, including but not limited to visual question answering, optical character recognition, document/table/chart understanding, and visual grounding.

Specifications#

Model Spec 1 (pytorch, 27 Billion)#

  • Model Format: pytorch

  • Model Size (in billions): 27

  • Quantizations: none

  • Engines: Transformers

  • Model ID: deepseek-ai/deepseek-vl2

  • Model Hubs: Hugging Face, ModelScope

Execute the following command to launch the model, remember to replace ${quantization} with your chosen quantization method from the options listed above:

xinference launch --model-engine ${engine} --model-name deepseek-vl2 --size-in-billions 27 --model-format pytorch --quantization ${quantization}

Model Spec 2 (pytorch, 16 Billion)#

  • Model Format: pytorch

  • Model Size (in billions): 16

  • Quantizations: none

  • Engines: Transformers

  • Model ID: deepseek-ai/deepseek-vl2-small

  • Model Hubs: Hugging Face, ModelScope

Execute the following command to launch the model, remember to replace ${quantization} with your chosen quantization method from the options listed above:

xinference launch --model-engine ${engine} --model-name deepseek-vl2 --size-in-billions 16 --model-format pytorch --quantization ${quantization}

Model Spec 3 (pytorch, 3 Billion)#

  • Model Format: pytorch

  • Model Size (in billions): 3

  • Quantizations: none

  • Engines: Transformers

  • Model ID: deepseek-ai/deepseek-vl2-tiny

  • Model Hubs: Hugging Face, ModelScope

Execute the following command to launch the model, remember to replace ${quantization} with your chosen quantization method from the options listed above:

xinference launch --model-engine ${engine} --model-name deepseek-vl2 --size-in-billions 3 --model-format pytorch --quantization ${quantization}