The following gist will be helpful for the latter part of this article: Inspecting ggml_cont Recently, I’ve been playing around with GGML. While doing so, I was looking through the examples, and I saw this in mnist_common.cpp: 1 2 3 dense_in = ggml_reshape_2d(model.ctx_compute, ggml_cont(model.ctx_compute, ggml_permute(model.ctx_compute, dense_in, 1, 2, 0, 3)), (MNIST_HW/4)*(MNIST_HW/4)*(MNIST_CNN_NCB*2), model.nbatch_physical); This was on line 362. It preceded a dense matrix multiplicatio...
