KAT-V1#

  • Context Length: 131072

  • Model Name: KAT-V1

  • Languages: en, zh

  • Abilities: chat

  • Description: Kwaipilot-AutoThink ranks first among all open-source models on LiveCodeBench Pro, a challenging benchmark explicitly designed to prevent data leakage, and even surpasses strong proprietary systems such as Seed and o3-mini.

Specifications#

Model Spec 1 (pytorch, 40 Billion)#

  • Model Format: pytorch

  • Model Size (in billions): 40

  • Quantizations: none

  • Engines: vLLM, Transformers, SGLang

  • Model ID: Kwaipilot/KAT-V1-40B

  • 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 KAT-V1 --size-in-billions 40 --model-format pytorch --quantization ${quantization}

Model Spec 2 (gptq, 40 Billion)#

  • Model Format: gptq

  • Model Size (in billions): 40

  • Quantizations: Int4-Int8Mix

  • Engines: vLLM, Transformers, SGLang

  • Model ID: QuantTrio/KAT-V1-40B-GPTQ-Int4-Int8Mix

  • 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 KAT-V1 --size-in-billions 40 --model-format gptq --quantization ${quantization}

Model Spec 3 (awq, 40 Billion)#

  • Model Format: awq

  • Model Size (in billions): 40

  • Quantizations: Int4

  • Engines: vLLM, Transformers, SGLang

  • Model ID: QuantTrio/KAT-V1-40B-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 KAT-V1 --size-in-billions 40 --model-format awq --quantization ${quantization}