Has Anyone Experimented with NCM + ANE?
I've been curious about whether Next-Concept Modeling (NCM) , a concept-based generative modeling architecture, could benefit from Apple's Neural Engine (ANE) .
With Apple Silicon offering dedicated hardware for machine learning, I'm wondering whether NCM could be optimized for local inference through Core ML.
Some Questions Worth Exploring
ANE Acceleration: Has anyone successfully deployed an NCM architecture on Apple's Neural Engine?
FP16 Precision: Could half-precision inference improve execution speed without significantly affecting prediction accuracy?
Core ML Compatibility: What challenges might arise when converting NCM components from PyTorch to Core ML?
Hardware Limitations: Could an M4 MacBook Air with 16GB unified memory handle this architecture efficiently?
Performance: Would the ANE offer meaningful advantages over CPU or GPU inference for this type of model?
Why This Interests Me
Much of the machine learning discussion focuses on GPUs and CUDA,…