objc.io @objc.io · May 15 Swift Talk 492 Building a Language Model: Neural Nets We explore neural nets, the foundation for modern machine learning. Our first small step is backpropagating through a computation graph, inspired by Andrej Karpathy’s Micrograd. talk.objc.io/episodes/S01...
4 likes 1 replies
?
Replies
the computation graph approach is great for intuition but the memory overhead of storing intermediate activations becomes a real bottleneck when scaling beyond micrograd-sized models most production systems switch to static graphs or operator fusion to handle it.