cturan/llama.cpp

огледало от https://github.com/cturan/llama.cpp

Автор	SHA1 Съобщение	Дата
Acly	638d330246 ggml : fix graph reallocation with multiple chunks (#16396)	преди 3 месеца
Acly	f2a789e334 ggml : split graph allocations according to backend max buffer size (#15815)	преди 3 месеца