moonshotai/Kimi-Linear-48B-A3B-Instruct · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Machine Learning Model Repository
Summary
The Hugging Face repo provides the Kimi‑K2‑Instruct model in GGUF format for local inference, detailing hardware needs, quantization guidance, and performance highlights.
Highlights
- Mixture‑of‑Experts LLM with 1 trillion total parameters and 32 B active parameters
- Quantized GGUF files run efficiently on consumer GPUs using a llama.cpp fork
- Recommended minimum 128 GB unified RAM for small quant runs, 16 GB VRAM for acceptable speed
- Guidance includes temperature setting, token limits, and expert selection per token
- Offers both base and instruction‑tuned variants for research and chat use
auto-generated
Context
Audience
Developers and researchers interested in running large MoE language models locally
DomainArtificial Intelligence
Formatmodel checkpoint in GGUF format with documentation
Accessopen source download from Hugging Face
Topics
Related
unsloth Dynamic quantizerllama.cpp forkKimi‑K2 technical blogLiveCodeBench benchmark
