Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Course Outline
Overview of CANN Optimization Capabilities
- Strategies for managing inference performance within the CANN environment
- Optimization priorities for edge-based and embedded artificial intelligence systems
- Governance of AI Core utilization and memory allocation protocols
Using Graph Engine for Analysis
- Fundamentals of the Graph Engine and its execution pipeline
- Visualization of operator graphs and runtime operational metrics
- Modifying computational graphs to enhance optimization outcomes
Profiling Tools and Performance Metrics
- Employment of the CANN Profiling Tool for comprehensive workload analysis
- Assessment of kernel execution duration and identification of performance bottlenecks
- Memory access profiling and implementation of tiling strategies
Custom Operator Development with TIK
- Summary of TIK capabilities and the operator programming framework
- Development of custom operators using the TIK DSL
- Testing protocols and performance benchmarking for custom operators
Advanced Operator Optimization with TVM
- Introduction to TVM integration within the CANN architecture
- Auto-tuning methodologies for computational graphs
- Criteria and procedures for selecting between TVM and TIK implementations
Memory Optimization Techniques
- Management of memory layout and buffer placement efficiency
- Methods for reducing on-chip memory consumption
- Best practices for asynchronous execution and resource reuse
Real-World Deployment and Case Studies
- Case study: performance tuning for smart city camera pipeline applications
- Case study: optimization of the autonomous vehicle inference stack
- Guidelines for iterative profiling and continuous process improvement
Summary and Next Steps
Requirements
- Demonstrated proficiency in deep learning model architectures and associated training processes.
- Practical experience deploying models via CANN, TensorFlow, or PyTorch environments.
- Competency in Linux command-line interfaces, shell scripting, and Python development.
Target Audience
- AI performance engineers seeking standardized solutions for government applications.
- Inference optimization specialists responsible for system efficiency.
- Developers specializing in edge AI or real-time computing systems for government use cases.
14 Hours
