MiniCPM-V-4.6-Thinking#

  • Context Length: 262144

  • Model Name: MiniCPM-V-4.6-Thinking

  • Languages: en, zh

  • Abilities: chat, vision, reasoning

  • Description: MiniCPM-V 4.6 Thinking is a reasoning-tuned version of MiniCPM-V 4.6. It supports deep reasoning for more complex problem solving, while inheriting the single-image, multi-image, and video understanding capabilities of the MiniCPM-V family.

Specifications#

Model Spec 1 (pytorch, 1 Billion)#

  • Model Format: pytorch

  • Model Size (in billions): 1

  • Quantizations: none

  • Engines: Transformers, SGLang

  • Model ID: openbmb/MiniCPM-V-4.6-Thinking

  • 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 MiniCPM-V-4.6-Thinking --size-in-billions 1 --model-format pytorch --quantization ${quantization}

Model Spec 2 (bnb, 1 Billion)#

  • Model Format: bnb

  • Model Size (in billions): 1

  • Quantizations: 4-bit

  • Engines: Transformers, SGLang

  • Model ID: openbmb/MiniCPM-V-4.6-Thinking-BNB

  • 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 MiniCPM-V-4.6-Thinking --size-in-billions 1 --model-format bnb --quantization ${quantization}

Model Spec 3 (awq, 1 Billion)#

  • Model Format: awq

  • Model Size (in billions): 1

  • Quantizations: Int4

  • Engines: Transformers, SGLang

  • Model ID: openbmb/MiniCPM-V-4.6-Thinking-AWQ

  • 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 MiniCPM-V-4.6-Thinking --size-in-billions 1 --model-format awq --quantization ${quantization}

Model Spec 4 (gptq, 1 Billion)#

  • Model Format: gptq

  • Model Size (in billions): 1

  • Quantizations: Int4

  • Engines: Transformers, SGLang

  • Model ID: openbmb/MiniCPM-V-4.6-Thinking-GPTQ

  • 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 MiniCPM-V-4.6-Thinking --size-in-billions 1 --model-format gptq --quantization ${quantization}

Model Spec 5 (ggufv2, 1 Billion)#

  • Model Format: ggufv2

  • Model Size (in billions): 1

  • Quantizations: Q4_K_M, Q4_K_S, Q5_K_M, Q5_K_S, Q6_K, Q8_0, F16

  • Engines: llama.cpp

  • Model ID: openbmb/MiniCPM-V-4.6-Thinking-gguf

  • 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 MiniCPM-V-4.6-Thinking --size-in-billions 1 --model-format ggufv2 --quantization ${quantization}