addpatch: ollama, ver=0.5.4-1

[wszqkzqk]

• Set ARCH to loong64 to avoid automatically using loongarch64 as GOARCH
• Update math32 to build on loong64
• Disable cuda and rocm

[wszqkzqk]

Errors:

ggml-cpu-quants.c:2263:20: error: incompatible types when initializing type ‘__m128’ using type ‘__vector(2) long long int’
ggml-cpu-quants.c:2264:20: error: incompatible types when initializing type ‘__m128’ using type ‘__vector(2) long long int’
ggml-cpu-quants.c:2265:20: error: incompatible types when initializing type ‘__m128’ using type ‘__vector(2) long long int’
ggml-cpu-quants.c:2266:20: error: incompatible types when initializing type ‘__m128’ using type ‘__vector(2) long long int’
ggml-cpu-quants.c:2271:30: error: incompatible types when initializing type ‘__m128’ using type ‘__m128i’
ggml-cpu-quants.c:2278:31: error: implicit declaration of function ‘mul_sum_i8_pairs’ [-Wimplicit-function-declaration]
ggml-cpu-quants.c:2278:31: error: incompatible types when initializing type ‘__m128i’ using type ‘int’
ggml-cpu-quants.c:2283:31: error: incompatible types when initializing type ‘__m128i’ using type ‘int’
ggml-cpu-quants.c:2289:30: error: incompatible types when initializing type ‘__m128’ using type ‘__m128i’
ggml-cpu-quants.c:2296:31: error: incompatible types when initializing type ‘__m128i’ using type ‘int’
ggml-cpu-quants.c:2301:31: error: incompatible types when initializing type ‘__m128i’ using type ‘int’
ggml-cpu-quants.c:2322:12: error: implicit declaration of function ‘hsum_float_4x4’ [-Wimplicit-function-declaration]

These errors need to be fixed upstream and it seems that these errors are fixed:

• ggerganov/llama.cpp@d5c0582
• ggerganov/llama.cpp@0c27e6f

But it's not so easy to back port them now.