RAPIDS - Zero Code Change Acceleration

RAPIDS - Zero Code Change Acceleration

1. RAPIDS 라이브러리를 통한 GPU 기반 ML모델 학습
2. Zero Code Change Acceleration를 통한 GPU 및 CPU 상호호환성 확보
3. GPU 학습 & CPU 추론 시스템 구축을 통한 리소스 최적화

설치 프로세스

• Conda 다운로드 및 설치

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

• RAPIDS 라이브러리 설치

conda create -n rapids-25.04 -c rapidsai-nightly -c conda-forge -c nvidia  \
    rapids=25.04 python=3.12 'cuda-version>=12.0,<=12.8'

Training in GPU

• train.py

from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
import pickle

# create dataset
X, y = make_classification(n_samples=500000, n_features=100, random_state=0)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)
 
# define model
rf = RandomForestClassifier(n_estimators=100, random_state=0, n_jobs=-1)

# train
rf.fit(X_train, y_train)

# save model with pickle
pickle.dump(rf, open("model.pkl", "wb"))

• run with cuml.accel

python -m cuml.accel train.py

Inference in CPU

• inference.py

import os
# This is for test
os.environ["CUDA_VISIBLE_DEVICES"] = ""
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
import pickle
 
# create dataset
X, y = make_classification(n_samples=500000, n_features=100, random_state=0)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=0)
 
# load model
rf = pickle.load(open("rf.pkl", "rb"))

# inference
rf.predict(X_test)

Jupyter Notebook용 acceleration 설정

# 첫 셀에서 아래 명령 실행
%load_ext cuml.accel

... existing codes ...