CoolFace
Modelpublic

devendradhakad/autodroid-Abiray-Nanbeige4.2-3B-GGUF

sourceHugging Faceapache-2.0updated 10d agoView on Hugging Face
0likes66downloads
Model Card

Nanbeige4.2-3B - GGUF Quants

This repository contains GGUF quantizations for [Nanbeige/Nanbeige4.2-3B](https://huggingface.co/Nanbeige/Nanbeige4.2-3B).

  • —Original Model: Nanbeige/Nanbeige4.2-3B
  • —Base Architecture: Looped Transformer (3B non-embedding parameters)
  • —Quantization Format: GGUF (Q4_K_M, Q4_K_S, Q5_K_M, Q6_K, Q8_0)

Available Files & Quantization Details

File NameSizeQuant MethodDescription
Nanbeige4.2-3B-Q4_K_M.gguf~2.57 GBQ4_K_M4-bit medium. Recommended balance of speed, memory usage, and quality.
Nanbeige4.2-3B-Q4_K_S.gguf~2.50 GBQ4_K_S4-bit small. Slightly lower memory footprint.
Nanbeige4.2-3B-Q5_K_M.gguf~2.99 GBQ5_K_M5-bit medium. Higher precision with slight increase in size.
Nanbeige4.2-3B-Q6_K.gguf~3.42 GBQ6_K6-bit quantization. Very close to FP16 performance.
Nanbeige4.2-3B-Q8_0.gguf~4.43 GBQ8_08-bit quantization. Maximum quality for GGUF.

Usage Guide

1. Running with llama.cpp

For full support, clone the official or nanbeige42 fork of llama.cpp:

bash
# Clone the repository with Nanbeige support
git clone -b nanbeige42 [https://github.com/Nanbeige/llama.cpp.git](https://github.com/Nanbeige/llama.cpp.git)
cd llama.cpp

# Build with CUDA support
cmake -B build -DGGML_CUDA=ON
cmake --build build --config Release -j

# Download a model from this repository
huggingface-cli download Abiray/Nanbeige4.2-3B-GGUF Nanbeige4.2-3B-Q4_K_M.gguf --local-dir .

# Run CLI inference
./build/bin/llama-cli -m Nanbeige4.2-3B-Q4_K_M.gguf -ngl 99 -p "Which number is bigger, 9.11 or 9.8?"