Merge branch 'main' into export-D80261684

Author: shoumikhin

.github/workflows/apple-perf.yml

@@ -230,7 +230,7 @@ jobs:
                 model.use_sdpa_with_kv_cache=true \
                 backend.xnnpack.enabled=true \
                 backend.xnnpack.extended_ops=true \
-                base.preq_mode="8da4w_output_8da8w" \
+                base.preq_mode="preq_8da4w_out_8da8w" \
                 base.preq_group_size=32 \
                 export.max_seq_length=2048 \
                 export.max_context_length=2048 \
@@ -256,7 +256,7 @@ jobs:
                 base.params="${DOWNLOADED_PATH}/params.json" \
                 quantization.use_qat=true \
                 base.use_lora=16 \
-                base.preq_mode="8da4w_output_8da8w" \
+                base.preq_mode="preq_8da4w_out_8da8w" \
                 base.preq_group_size=32 \
                 base.preq_embedding_quantize=\'8,0\' \
                 model.use_sdpa_with_kv_cache=true \

examples/models/llama/CMakeLists.txt

@@ -219,6 +219,8 @@ endif()
 target_include_directories(llama_main PUBLIC ${_common_include_directories})
 target_link_libraries(llama_main PUBLIC llama_runner ${link_libraries})
 target_compile_options(llama_main PUBLIC ${_common_compile_options})
-if(NOT APPLE)
+if(APPLE)
+  target_link_options(llama_main PRIVATE -Wl,-rpath,@executable_path)
+elseif(UNIX)
   set_target_properties(llama_main PROPERTIES LINK_FLAGS "-Wl,-rpath='$ORIGIN'")
 endif()