Distribute and run LLMs with a single file.
C++ cross-platformggufllama-cpplocal-ailocal-inferencelocal-llmopen-source-aisingle-file-executablespeech-to-text
Add missing CPUID check
Justine Tunneye75caa154be8
diff --git a/llamafile/sgemm.cpp b/llamafile/sgemm.cppindex 553283dd..d7324c1f 100644--- a/llamafile/sgemm.cpp+++ b/llamafile/sgemm.cpp@@ -58,13 +58,15 @@ static const struct GemmFuncs {// AMD Excavator (2015-2022)sgemm = llamafile_sgemm_amd_avx2;mixmul = llamafile_mixmul_amd_avx2;- iqk_mixmul = iqk_mul_mat_moe;+ if (X86_HAVE(F16C))+ iqk_mixmul = iqk_mul_mat_moe;}} else {// AMD Piledriver (2011-2014)sgemm = llamafile_sgemm_amd_fma;mixmul = llamafile_mixmul_amd_fma;- iqk_mixmul = iqk_mul_mat_moe;+ if (X86_HAVE(F16C))+ iqk_mixmul = iqk_mul_mat_moe;}} else {// Intel Sandybridge/Ivybridge (2010-2012)