Closed
Description
While playing around, I noticed the embeddings are only 512 floats rather than the 4096 you get when using the standalone application.
So I went digging and I found the culprit which was a copy-paste residue in the function llama_n_embd
llama-cpp-python/llama_cpp/llama_cpp.py
Lines 220 to 221 in 6d1bda4
It's calling
llama_n_ctx rather than llama_n_embd.
I don't think this warrants a pull request as it is a very easy issue to fix, so I made a simple issue instead.
Keep up the good work :)