From aa4c1bb8b0071246d140d0027caee00a1128a4dc Mon Sep 17 00:00:00 2001 From: Xiuchuan Zhai Date: Mon, 25 Nov 2024 04:52:30 -0500 Subject: [PATCH] fix a arm error --- .../src/transformations/cpu_opset/x64/op/llm_mlp.hpp | 3 ++- .../src/transformations/transformation_pipeline.cpp | 5 +++-- 2 files changed, 5 insertions(+), 3 deletions(-) diff --git a/src/plugins/intel_cpu/src/transformations/cpu_opset/x64/op/llm_mlp.hpp b/src/plugins/intel_cpu/src/transformations/cpu_opset/x64/op/llm_mlp.hpp index b69a5253e09275..40e1e44b93a91b 100644 --- a/src/plugins/intel_cpu/src/transformations/cpu_opset/x64/op/llm_mlp.hpp +++ b/src/plugins/intel_cpu/src/transformations/cpu_opset/x64/op/llm_mlp.hpp @@ -33,7 +33,8 @@ class LLMMLPNode : public ov::op::Op { // 2: up_proj // 3: down_proj LLMMLPNode(const OutputVector& args, const Config& cfg, const ov::element::Type output_type = ov::element::undefined) - : Op(args), m_args(args), m_config(cfg), m_output_type(output_type) { + : Op(args), m_config(cfg), m_output_type(output_type) { + m_args = args; validate_and_infer_types(); } diff --git a/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp b/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp index 856c25fab127f4..6b79e2ab60ad19 100644 --- a/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp +++ b/src/plugins/intel_cpu/src/transformations/transformation_pipeline.cpp @@ -405,19 +405,20 @@ void Transformations::PreLpt(const std::vector& defaultPrecis precisions_map fp_convert_precision_map = {{ov::element::f32, ov::element::f16}}; #if defined(OPENVINO_ARCH_ARM) || defined(OPENVINO_ARCH_ARM64) type_to_fuse_map fuse_map = {{ov::opset1::FakeQuantize::get_type_info_static(), fuse_type_to_fq}}; + const bool store_original_precision_as_rt_attribute = false; #else type_to_fuse_map fuse_map = {{ov::op::PagedAttentionExtension::get_type_info_static(), fuse_type_to_pa}}; + const bool store_original_precision_as_rt_attribute = true; #endif const bool keep_precision_sensitive_in_fp32 = true; const bool need_convert_input_output_precision = false; - const bool save_original_precision_attribute = true; CPU_REGISTER_PASS_COMMON(manager, ov::pass::ConvertPrecision, fp_convert_precision_map, fuse_map, keep_precision_sensitive_in_fp32, need_convert_input_output_precision, - save_original_precision_attribute); + store_original_precision_as_rt_attribute); } CPU_REGISTER_PASS_COMMON(manager, ov::pass::KeepConstAndDecompression); CPU_SET_CALLBACK_COMMON(manager,