e2eAIOK 0.1.0

Intel® End-to-End AI Optimization Kit

Intel® End-to-End AI Optimization Kit

Intel® End-to-End AI Optimization Kit (code name AIDK) is a set of kits for E2E AI democratization on CPU. It is a pipeline framework that streamlines AI democratization technologies in each stage of E2E AI pipeline, including data processing, feature engineering, training, hyper-parameter tunning, and inference. Intel® End-to-End AI Optimization Kit delivers high performance, lightweight models efficiently on commodity hardware.

Motivation and Solution

Modern end to end machine learning or deep learning system usually includes a lifecycle of data processing, feature engineering, training, inference and serving. In this project, we provided four Major components:

• RecDP (Feature Engineering)

• Democratized Modelzoo(Recsys, Computer Vision, NLP, speech recognition, etc)

• DeNas(train-free Neural Architecture Search)

• TLK(Transfer Learning Kit)

Indices and tables

• Examples

  • Built-in Models

    • DLRM - recsys, facebook
    • DIEN - recsys, alibaba
    • WnD - recsys, google
    • RNNT - speech recognition
    • RESNET - computer vision
    • BERT - Natual Language Processing
    • MiniGO - minimalist engine modeled after AlphaGo Zero

  • DENAS Models

    • ComputerVision - computer vision
    • BERT - Natural Language Processing
    • ASR - speech recognition

  • TLK Models

    • TLK - transfer learning kit

  • DENAS-TLK Joint Models

    • BERT - Natural Language Processing

  • hugging face intergration

    • NLP_DeNas_KD - hugging face + DENAS NLP + TLK KD
    • CV_DeNas_KD - hugging face CV + TLK