Skip to content

ZunhaiSu/Super-Experts-Profilling

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Installation

conda create -n Super_Experts python==3.12 -y
conda activate Super_Experts
conda install pytorch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 pytorch-cuda=11.8 -c pytorch -c nvidia
pip install -r requirements.txt

Quick start

Profiling Down_proj Outliers of Qwen3-30B-A3B

CUDA_VISIBLE_DEVICES=0 python3 run.py \
--model_name Your Model Path \
--save_path ./output/Qwen3-30B-A3B \
--profile_outliers \
--vis_outliers_heatmap

Profiling Super Experts of Qwen3-30B-A3B

CUDA_VISIBLE_DEVICES=0 python3 run.py \
--model_name Your Model Path \
--save_path ./output/Qwen3-30B-A3B \
--profile_super_experts \
--vis_super_experts_line_plot

PPL test of Original Qwen3-30B-A3B

CUDA_VISIBLE_DEVICES=0 python3 run.py \
--model_name Your Model Path \
--save_path ./output/Qwen3-30B-A3B \
--dataset_name wikitext2 \
--eval_ppl

PPL Test of Qwen3-30B-A3B After Prune Super Experts

CUDA_VISIBLE_DEVICES=0 python3 run.py \
--model_name Your Model Path \
--save_path ./output/Qwen3-30B-A3B \
--eval_ppl \
--dataset wikitext2 \
--prune_super_experts

About

Unveiling Super Experts in Mixture-of-Experts Large Language Models

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published