Add build instructions for ARM64

Author: XbGH-MS

__readme__.txt

@@ -1,20 +1,34 @@
 Offspring for hoivb612
 https://github.com/hoivb612/llama.cpp
 
-#include <sys/stat.h>
-#include <sys/types.h>
-#include <string>
-#include <hash>
+===========================================
 
-static std::hash<std::string> hasher;
-static const char* dir = "./llama_cache";
+For ARM64: 
+cmake --preset arm64-windows-llvm-release -D GGML_LLAMAFILE=OFF -D GGML_OPENMP=OFF -B build.arm
+cd build.arm
+cmake --build . --config RelWithDebInfo --target llama-bench xbapp
 
-// create the cache dir if it does not exist yet
-struct stat info;
-if (stat(dir, &info) != 0) {
-    mkdir(dir, 0777);
-}
+bin\llama-bench.exe -m c:\llama.cpp\models\Llama-3.2-3B-Instruct-Q4_0_4_8.gguf -t 8 -p 128 -n 64
+| model                          |       size |     params | backend    | threads |          test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | ------: | ------------: | -------------------: |
+| llama 3B Q4_0_4_8              |   2.08 GiB |     3.61 B | CPU        |       8 |         pp128 |        306.69 ± 9.23 |
+| llama 3B Q4_0_4_8              |   2.08 GiB |     3.61 B | CPU        |       8 |          tg64 |         45.39 ± 0.74 |
+
+bin\llama-bench.exe -m c:\llama.cpp\models\Llama-3.2-3B-Instruct-Q2_K-Second.gguf -t 8 -p 128 -n 64
+| model                          |       size |     params | backend    | threads |          test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | ------: | ------------: | -------------------: |
+| llama 3B Q2_K - Medium         |   1.56 GiB |     3.61 B | CPU        |       8 |         pp128 |         71.69 ± 0.43 |
+| llama 3B Q2_K - Medium         |   1.56 GiB |     3.61 B | CPU        |       8 |          tg64 |         46.66 ± 0.47 |
+
+bin\llama-bench.exe -m c:\llama.cpp\models\Phi-3.5-mini-instruct-Q4_0_4_8.gguf -t 8 -p 128 -n 64
+| model                          |       size |     params | backend    | threads |          test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | ------: | ------------: | -------------------: |
+| phi3 3B Q4_0_4_8               |   2.03 GiB |     3.82 B | CPU        |       8 |         pp128 |        233.87 ± 6.45 |
+| phi3 3B Q4_0_4_8               |   2.03 GiB |     3.82 B | CPU        |       8 |          tg64 |         40.70 ± 0.47 |
+
+bin\llama-bench.exe -m c:\llama.cpp\models\Phi-3.5-mini-instruct-Q2_K.gguf -t 8 -p 128 -n 64
+| model                          |       size |     params | backend    | threads |          test |                  t/s |
+| ------------------------------ | ---------: | ---------: | ---------- | ------: | ------------: | -------------------: |
+| phi3 3B Q2_K - Medium          |   1.32 GiB |     3.82 B | CPU        |       8 |         pp128 |         50.47 ± 5.81 |
+| phi3 3B Q2_K - Medium          |   1.32 GiB |     3.82 B | CPU        |       8 |          tg64 |         34.63 ± 0.20 |
 
-// default generated file name
-std::string pfx_path(dir);
-std::string full_file_path = pfx_path + "/" + std::to_string(hasher(pfx));