Jan Kaniewski

Try https://github.com/getnamo/Llama-Unreal/releases/tag/v0.4.0 with the CPU only build to see if it works out of the box. NB this has the refactor so if you used blueprints off of the old...

Just to confirm v0.4 is failing with cuda = false in build.cs for you guys?

Apparently building llama.cpp yourself locally can resolve this. Hints at static lib config maybe?

See https://github.com/getnamo/Llama-Unreal/pull/10 for an alternative path. Thanks to @ellllie-42 pr you can now specify a LLAMA_PATH and use your standard CUDA_PATH if you want a dev friendly custom build env....

Finally addressed with https://github.com/getnamo/Llama-Unreal/releases/tag/v0.9.0 release (windows only at this time). Re-open if still broken.

#### Resources: - memory db c lib: https://github.com/facebookresearch/faiss/blob/main/c_api/INSTALL.md - using embedding model to generate vectors: https://github.com/ggerganov/llama.cpp/blob/ceb/bert/examples/embedding/embedding.cpp already built in support for Bert style embedders. - NB best results are achieved...

Another interesting approach: https://arxiv.org/html/2412.06078v1

being tackled in: https://github.com/getnamo/Llama-Unreal/tree/feat-hnswlib Picking hnswlib for vector search and in-built llama.cpp functionality for embedding models. More work needed for full loop tests. Added tasks to description for tracking.

Embedding looks good, but vectordb in testing on real vectors has some sort of vector pointer de-ref issue TBC

UE 5.5 Windows is now supported with https://github.com/getnamo/Llama-Unreal/releases/tag/v0.9.0. Mac build still currently pending more dev time.