3 ms·
Yes, absolutely in deep learning. Custom fused CUDA kernels everywhere.
by qeternity 11mo ago
Yes, absolutely in deep learning. Custom fused CUDA kernels everywhere.
- Scene_Cast2 11mo agoYep. MoE, FlashAttention, or sparse retrieval architectures for example.