bcacdwk/slidesparse-checkpoints

SlideSparse Checkpoints

Pre-converted sparse model checkpoints using the SlideSparse technique.

Overview

This repository contains model weights converted with various sparsity configurations:

• 2:4 - Standard N:M sparsity (50% sparse)
• 2:6 - Extended sparsity (67% sparse)
• 2:8 - Higher sparsity (75% sparse)
• 2:10 - Maximum sparsity (80% sparse)

Models Included

Base Model	Quantization	Sparsity Variants
Llama-3.2-1B	INT8, FP8	2:4, 2:6, 2:8, 2:10
Llama-3.2-3B	INT8, FP8	2:4, 2:6, 2:8, 2:10
Qwen2.5-7B	INT8, FP8	2:4, 2:6, 2:8, 2:10
Qwen2.5-14B	INT8, FP8	2:4, 2:6, 2:8, 2:10

Source Models

These checkpoints are derived from:

• RedHatAI/Llama-3.2-1B-Instruct-quantized.w8a8
• RedHatAI/Llama-3.2-3B-Instruct-quantized.w8a8
• RedHatAI/Qwen2.5-7B-Instruct-quantized.w8a8
• RedHatAI/Qwen2.5-14B-Instruct-quantized.w8a8

License

• Qwen models: Apache 2.0
• Llama models: Please refer to Meta's Llama license

Usage

# Download all checkpoints
huggingface-cli download bcacdwk/slidesparse-checkpoints --local-dir ./checkpoints_slidesparse

# Download specific model
huggingface-cli download bcacdwk/slidesparse-checkpoints Llama3.2-1B-INT8-SlideSparse-2_4 --local-dir ./checkpoints_slidesparse/Llama3.2-1B-INT8-SlideSparse-2_4

Citation

If you use these checkpoints, please cite the SlideSparse paper (coming soon).