Kimi-K3#

  • Context Length: 1048576

  • Model Name: Kimi-K3

  • Languages: en, zh

  • Abilities: chat, vision, reasoning, tools

  • Description: Kimi K3 is an open-weight, native multimodal agentic model and our most capable model to date.

Specifications#

Model Spec 1 (pytorch, 2779_93 Billion)#

  • Model Format: pytorch

  • Model Size (in billions): 2779_93

  • Quantizations: none

  • Engines: vLLM

  • Model ID: moonshotai/Kimi-K3

  • 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 Kimi-K3 --size-in-billions 2779_93 --model-format pytorch --quantization ${quantization}

Model Spec 2 (ggufv2, 2779_93 Billion)#

  • Model Format: ggufv2

  • Model Size (in billions): 2779_93

  • Quantizations: UD-IQ1_M, UD-IQ1_S, UD-IQ2_XXS, UD-Q1_0, UD-Q2_K_XL, UD-Q4_K_XL, UD-Q8_K_XL, UD-TQ1_0, UD-TQ2_0

  • Engines: llama.cpp

  • Model ID: unsloth/Kimi-K3-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 Kimi-K3 --size-in-billions 2779_93 --model-format ggufv2 --quantization ${quantization}