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.
Duration 14 hours
Course Outline
Summary of CANN Performance Enhancement Features
- Management of inference workloads within the CANN framework
- Strategic targets for optimizing edge and embedded AI infrastructure
- Comprehension of AI Core resource allocation and memory management
Leveraging the Graph Engine for Analytical Purposes
- Orientation to the Graph Engine and its execution workflows
- Visualization of operator dependencies and runtime performance indicators
- Modification of computational structures for efficiency enhancements
Diagnostic Instruments and Performance Indicators
- Application of the CANN Profiling Tool for workload assessment
- Evaluation of kernel execution duration and identification of constraints
- Analysis of memory access patterns and tiling methodologies
Development of Custom Operators via TIK
- Overview of TIK and the associated operator programming framework
- Creation of tailored operators utilizing the TIK Domain-Specific Language
- Validation and benchmarking of operator efficiency metrics
Advanced Operator Refinement with TVM
- Introduction to the integration of TVM with the CANN ecosystem
- Automated tuning methods for computational networks
- Criteria for transitioning between TVM and TIK implementations
Memory Management Strategies
- Oversight of memory structures and buffer positioning
- Methods to decrease on-chip memory expenditure
- Standard procedures for asynchronous operations and resource reuse
Practical Implementation and Illustrative Scenarios
- Scenario analysis: performance adjustment for intelligent urban surveillance systems
- Scenario analysis: optimization of inference architectures for autonomous vehicles
- Protocols for iterative profiling and sustained enhancement
Conclusions and Future Directives
Requirements
- Comprehensive knowledge of deep learning model architectures and training processes
- Proficiency in model deployment using CANN, TensorFlow, or PyTorch
- Competence in Linux command-line interfaces, shell scripting, and Python programming
Intended Participants
- AI performance specialists
- Inference optimization experts
- Engineers working on edge AI or real-time systems for government
