microsoft/archai

Accelerate your Neural Architecture Search (NAS) through fast, reproducible and modular research.

What it solves

Archai is designed to accelerate Neural Architecture Search (NAS), making the process of finding the most efficient deep network architectures for specific applications faster, more reproducible, and modular.

How it works

The framework allows users to define a search space (such as for Transformers) and set specific optimization objectives, such as reducing latency, memory usage, or parameter count. It then employs search algorithms, like EvolutionParetoSearch, to iterate through different network configurations and identify a Pareto-optimal frontier of the best-performing architectures.

Who it’s for

It is intended for researchers and developers who need to generate efficient deep learning models, particularly those working with Natural Language Processing (NLP) and computer vision tasks like face segmentation.

Highlights

  • Modular Design: Supports flexible search spaces and objectives.
  • Pareto-Optimization: Finds a balance between multiple competing objectives (e.g., speed vs. accuracy).
  • Broad Application: Includes end-to-end tasks for text generation and face segmentation.
  • PyTorch Integration: Built to work with PyTorch 1.7.0+.

Related

  • Project
    fastai/fastai

    A deep learning library built on PyTorch that provides high-level components for rapid productivity and low-level components for research flexibility.

  • Project
    amzn/pecos

    A modular machine learning framework designed for fast learning and inference on problems with extremely large output spaces, such as extreme multi-label ranking and large-scale retrieval.

  • Project
    benchen4395/alpha_agentic_search

    A stable, low-latency multi-agent search framework featuring a 5-layer hierarchical memory RAG stack that improves over time as it archives successful answers.

  • Project
    pytorch/pytorch

    A Python-based machine learning framework that provides GPU-accelerated tensor computation and a dynamic autograd system for building deep neural networks.

  • Project
    kyegomez/zeta

    A modular PyTorch framework providing reusable, high-performance building blocks for assembling state-of-the-art AI models, including attention mechanisms, quantization, and multi-modal architectures.