For context :

When running the GPT inference example, from the README, at the step make -j4 gpt-2 gpt-j, we get the following error :

make: *** No rule to make target `gpt-2'.  Stop.

Any idea how to fix this ?

The gpt-2 example has been modified since that document was written, and currently that are multiple versions of it, meant to show how to use different features of ggml. The names of the targets are gpt-2-ctx, gpt-2-alloc, gpt-2-backend and gpt-2-backend2. You can use any of these instead.

Despite what the README says, the gpt-2-backend version supports full GPU offloading, and the gpt-2-backend2 version supports partial offloading, ie. running a part of the model in the GPU and the rest on the CPU, similar to llama.cpp.

This is still an issue, we need to update the documentation, and possibly make one of these the default gpt-2 target.