Someone wanted to learn this too, so Grasp built them a personal learning path.
You can keep exploring this one or create a path like this one shaped around you.
Recommended resources
Module 2
Advanced C++: Templates, Concurrency, and Performance
Module 3
CPU Performance Engineering and Profiling
Module 4
High-Performance CPU Kernels with SIMD and OpenMP
Module 5
CUDA Programming Fundamentals
Module 6
Advanced CUDA Kernels for Deep Learning
Module 7
Implementing Core Deep Learning Operators
Module 8
Inference Engines and Graph Optimization
Module 9
Specialized Techniques for LLM Inference
Module 10
Capstone Project: Optimizing CNN Inference
Module 11