Why does llama.cpp implement GGML again in the repo instead of referencing the GGML repo?

[Ucag]

I noticed that llama.cpp reimplement ggml in the repo. Why not use the existing ggml repo ? Is there any difference between the two implementations? Can I use ggml repo to replace the implementation of llama.cpp?

[phymbert]

It is a playground for GGML as described here:

• https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#description
• Sync to llama.cpp: sync : ggml #6351
• Sync from llama.cpp: sync : llama.cpp ggml#773