CoreAI-Opt
  • API Reference
  • GitHub Repo
/

Introduction

  • Installation
  • How to use coreai-opt
  • Integration with Core AI

Examples

  • Toy Model Examples (Jupyter Notebooks)
    • Applying quantization to an MNIST model
    • Applying palettization to an MNIST model
    • Applying joint palettization and activation quantization to an MNIST model
  • Model Examples
    • ResNet50 model: Weight + Activation Quantization (PTQ)
    • Mixed-precision palettization with ResNet50
    • EDSR model: Weight palettization + activation quantization (PTQ)

Core Algorithms and APIs

  • Quantization
    • Basics
    • API Overview
    • Config API
    • Deeper Dive
  • Palettization
    • Basics
    • API Overview
    • Config API

Other Workflows and Utilities

  • Joint Compression
  • Mixed-Precision Compression
  • Inspecting PyTorch Model Structure
  • 16-bit PyTorch Model Casting
  • Compressing Core AI Models

Reference

  • API Reference
    • coreai_opt.CoreMLExportError
    • coreai_opt.ExportBackend
    • coreai_opt.casting.cast_fp32_to_fp16
    • coreai_opt.casting.cast_int32_to_int16
    • coreai_opt.casting.cast_to_16_bit_precision
    • coreai_opt.config.CompressionConfig
    • coreai_opt.config.CompressionSpec
    • coreai_opt.config.CompressionType
    • coreai_opt.config.ModuleCompressionConfig
    • coreai_opt.config.OpCompressionConfig
    • coreai_opt.config.WeightOnlyModuleValidationMixin
    • coreai_opt.config.WeightOnlyOpValidationMixin
    • coreai_opt.config.spec.CompressionComponentFactoryBase
    • coreai_opt.config.spec.CompressionSimulatorBase
    • coreai_opt.config.spec.CompressionTargetTensor
    • coreai_opt.coreai_utils.CompressionGranularity
    • coreai_opt.coreai_utils.DType
    • coreai_opt.coreai_utils.palettize_weights
    • coreai_opt.coreai_utils.quantize_weights
    • coreai_opt.coreai_utils.sparsify_weights
    • coreai_opt.coreai_utils.common.QScheme
    • coreai_opt.inspection.ModelInspector
    • coreai_opt.inspection.ModelSummary
    • coreai_opt.inspection.ModuleContext
    • coreai_opt.inspection.ModuleInfo
    • coreai_opt.inspection.OpInfo
    • coreai_opt.inspection.SourceFrame
    • coreai_opt.palettization.KMeansPalettizer
    • coreai_opt.palettization.KMeansPalettizerConfig
      • presets
        • w4
        • w6
        • w8
    • coreai_opt.palettization.ModuleKMeansPalettizerConfig
      • presets
        • w4
        • w6
        • w8
    • coreai_opt.palettization.PalettizationSpec
    • coreai_opt.palettization.config.OpKMeansPalettizerConfig
    • coreai_opt.palettization.spec.PalettizationGranularity
    • coreai_opt.palettization.spec.PerGroupedChannelGranularity
    • coreai_opt.palettization.spec.PerTensorGranularity
    • coreai_opt.palettization.spec.default_weight_palettization_spec
    • coreai_opt.pruning.MagnitudePruner
    • coreai_opt.pruning.MagnitudePrunerConfig
    • coreai_opt.pruning.ModuleMagnitudePrunerConfig
    • coreai_opt.pruning.PruningSpec
    • coreai_opt.pruning.config.ConstantSparsitySchedule
    • coreai_opt.pruning.config.OpMagnitudePrunerConfig
    • coreai_opt.pruning.config.PolynomialDecaySchedule
    • coreai_opt.pruning.config.SparsityScheduleBase
    • coreai_opt.pruning.spec.ChannelStructured
    • coreai_opt.pruning.spec.PruneImplBase
    • coreai_opt.pruning.spec.PruningScheme
    • coreai_opt.pruning.spec.Unstructured
    • coreai_opt.pruning.spec.default_weight_pruning_spec
    • coreai_opt.quantization.ExecutionMode
    • coreai_opt.quantization.ModuleQuantizerConfig
      • presets
        • w4
        • w4_per_block
        • w8
    • coreai_opt.quantization.QuantizationSpec
    • coreai_opt.quantization.Quantizer
    • coreai_opt.quantization.QuantizerConfig
      • presets
        • w4
        • w4_per_block
        • w8
    • coreai_opt.quantization.config.OpQuantizerConfig
    • coreai_opt.quantization.config.QATSchedule
    • coreai_opt.quantization.spec.GlobalMinMaxQParamsCalculator
    • coreai_opt.quantization.spec.MinMaxRangeCalculator
    • coreai_opt.quantization.spec.MovingAverageQParamsCalculator
    • coreai_opt.quantization.spec.PerBlockGranularity
    • coreai_opt.quantization.spec.PerChannelGranularity
    • coreai_opt.quantization.spec.PerTensorGranularity
    • coreai_opt.quantization.spec.QParamsCalculatorBase
    • coreai_opt.quantization.spec.QuantizationComponentFactory
    • coreai_opt.quantization.spec.QuantizationFormulation
    • coreai_opt.quantization.spec.QuantizationGranularity
    • coreai_opt.quantization.spec.QuantizationScheme
    • coreai_opt.quantization.spec.RangeCalculatorBase
    • coreai_opt.quantization.spec.RunningRangeMixin
    • coreai_opt.quantization.spec.StaticQParamsCalculator
    • coreai_opt.quantization.spec.default_activation_quantization_spec
    • coreai_opt.quantization.spec.default_weight_quantization_spec
    • coreai_opt.quantization.spec.fake_quantize.FakeQuantizeImplBase

Resources

  • Changelog
  1. CoreAI-Opt /
  2. Toy Model Examples (Jupyter Notebooks)
View as Markdown

Toy Model Examples (Jupyter Notebooks)ΒΆ

These tutorials demonstrate model compression techniques using simple toy models. Each tutorial is a Jupyter notebook that you can download and run locally.

  • Applying quantization to an MNIST model
  • Applying palettization to an MNIST model
  • Applying joint palettization and activation quantization to an MNIST model
Previous
Integration with Core AI
Next
Applying quantization to an MNIST model

2026, Apple, Inc. All rights reserved

Made with Sphinx and Shibuya theme.