LLM EXPLORER 58,133 MODELS INDEXED

Kimi K3 Cubic 2.5Bit by QuantTrio

By QuantTrio · 39 downloads
HF Repo ↗ Copy URL

Kimi K3 Cubic 2.5Bit is an open-source language model by QuantTrio. Features: 874.6b LLM, VRAM: 85.2GB, License: other, LLM Explorer Score: 0.29.

  Arxiv:2608.06763   8-bit   Base model:moonshotai/kimi-k3 Base model:quantized:moonshota...   Cubic-quantization   Custom code   Feature-extraction   Image-text-to-text   Kimi k3   Multimodal   Region:us   Safetensors   Sharded   Tensorflow   Vllm   W2a16   W2a8   W3a16   W3a8   W4a16   W4a8
Model Card on HF 🤗: https://huggingface.co/QuantTrio/Kimi-K3-Cubic-2.5Bit 

Kimi K3 Cubic 2.5Bit Parameters and Internals

LLM NameKimi K3 Cubic 2.5Bit
Repository 🤗https://huggingface.co/QuantTrio/Kimi-K3-Cubic-2.5Bit 
Base Model(s)  Kimi K3   moonshotai/Kimi-K3
Model Size874.6b
Required VRAM85.2 GB
Updated2026-08-11
MaintainerQuantTrio
Model Typekimi_k3
Model Files  2.3 GB: 1-of-341   3.1 GB: 2-of-341   3.1 GB: 3-of-341   3.1 GB: 4-of-341   3.1 GB: 5-of-341   0.7 GB: 6-of-341   3.1 GB: 7-of-341   3.1 GB: 8-of-341   3.1 GB: 9-of-341   3.1 GB: 10-of-341   0.7 GB: 11-of-341   3.1 GB: 12-of-341   3.1 GB: 13-of-341   3.1 GB: 14-of-341   3.1 GB: 15-of-341   0.3 GB: 16-of-341   3.1 GB: 17-of-341   3.1 GB: 18-of-341   3.1 GB: 19-of-341   3.1 GB: 20-of-341   0.2 GB: 21-of-341   3.1 GB: 22-of-341   3.1 GB: 23-of-341   3.1 GB: 24-of-341   3.1 GB: 25-of-341   0.2 GB: 26-of-341   3.1 GB: 27-of-341   3.1 GB: 28-of-341   3.1 GB: 29-of-341   3.1 GB: 30-of-341   0.2 GB: 31-of-341   3.1 GB: 32-of-341   3.1 GB: 33-of-341
Model ArchitectureKimiK3ForConditionalGeneration
Licenseother