GGML_ASSERT: D:\a\llama.cpp\llama.cpp\llama.cpp:12198: hparams.n_embd_head_k % ggml_blck_size(type_k) == 0

[techvd]

I'm having trouble using recent versions of llama.cpp with recent versions of llama-cpp-python. This is on Windows 11.

When I try to load a GGUF model, I get this assert:
GGML_ASSERT: D:\a\llama.cpp\llama.cpp\llama.cpp:12198: hparams.n_embd_head_k % ggml_blck_size(type_k) == 0
And the program crashes.
This happens with the most recent version (b2360) but a few of the previous versions I tried have the same problem as well.

Any idea what could be causing this issue?

In the past updating both llama.cpp and llama-cpp-python to the latest versions always used to work.

An older combination of llama.cpp b2234 and llama-cpp-python 0.2.47 works fine. However, I can't use b2234 with the latest version of llama-cpp-python because it gives this error:
AttributeError: function 'llama_rope_type' not found. Did you mean: 'llama_vocab_type'?

I hope that makes sense.

[github-actions]

This issue was closed because it has been inactive for 14 days since being marked as stale.