SOPHONSDKUserGuide
(HEAD detached at 936e3aa)

Catalogue

  • 1. Declaration
  • 2. Introduction
  • 3. SDK Porting Development Overview and Packages
  • 4. Quick Start
  • 5. MLIR-Model Migration Process
    • 5.1. Overview of MLIR-Migration Tool
    • 5.2. Using TPU-MLIR to Generate FLOAT BModel
    • 5.3. Using TPU-MLIR for Model Quantization
  • 6. Algorithmic transplant
  • 7. Appendix
SOPHONSDKUserGuide
  • 5. MLIR-Model Migration Process
  • View page source

5. MLIR-Model Migration Process

  • 5.1. Overview of MLIR-Migration Tool
  • 5.2. Using TPU-MLIR to Generate FLOAT BModel
    • 5.2.1. Loading tpu-mlir
    • 5.2.2. Preparing the working directory
    • 5.2.3. ONNX to MLIR
  • 5.3. Using TPU-MLIR for Model Quantization
    • 5.3.1. Load tpu-mlir
    • 5.3.2. Prepare the working directory
    • 5.3.3. ONNX to MLIR
    • 5.3.4. MLIR to F16 Model
    • 5.3.5. MLIR to INT8 Model
      • 5.3.5.1. Generating the Calibration Table
      • 5.3.5.2. Compiling to INT8 Symmetric Quantized Model
    • 5.3.6. Effect Comparison
    • 5.3.7. Model performance testing
      • 5.3.7.1. Install the libsophon environment
      • 5.3.7.2. Check the performance of the BModel
Previous Next

© Copyright 2024, SOPHGO.

Built with Sphinx using a theme provided by Read the Docs.