[qeternity]

Yes, absolutely in deep learning. Custom fused CUDA kernels everywhere.

[Scene_Cast2]

Yep. MoE, FlashAttention, or sparse retrieval architectures for example.