Get in Touch

Course Outline

Overview of Cambricon and MLU Architecture

  • Review of Cambricon’s artificial intelligence processor offerings
  • MLU architecture and instruction processing pipeline
  • Compatible model categories and application scenarios for government use

Development Toolchain Installation

  • Installation of BANGPy and Neuware SDK components
  • Configuration of Python and C++ development environments
  • Assessment of model compatibility and data preprocessing requirements

Model Development Using BANGPy

  • Management of tensor structures and dimensions
  • Construction of computation graphs
  • Implementation of custom operations within the BANGPy framework

Deployment via Neuware Runtime

  • Model conversion procedures and loading mechanisms
  • Control of execution flows and inference processes
  • Best practices for deployment in both edge computing and data center environments

Performance Optimization Strategies

  • Memory mapping techniques and layer-specific tuning
  • Execution tracing and system profiling methods
  • Identification of common performance bottlenecks and resolution strategies

Integration of MLU into Applications

  • Leveraging Neuware APIs for seamless application integration
  • Implementation of streaming capabilities and multi-model support
  • Configuration of hybrid CPU-MLU inference workflows

Comprehensive Project and Use Case Analysis

  • Practical exercise: Deployment of vision or natural language processing models
  • Execution of edge inference utilizing BANGPy integration
  • Evaluation of model accuracy and throughput metrics

Summary and Strategic Next Steps

Requirements

  • Familiarity with the architecture of machine learning models
  • Proficiency in Python and/or C++
  • Knowledge of techniques for model deployment and performance acceleration

Intended Audience

  • Engineers specializing in embedded AI systems
  • Machine learning professionals deploying solutions to edge computing or datacenter environments
  • Software developers engaged with Chinese AI infrastructure initiatives
 21 Hours

Number of participants


Price per participant

Testimonials (1)

Upcoming Courses

Related Categories