enforce inference precision to f16 for GNR acc verify

liubo-intel · Dec 24, 2024 · de77c8d · de77c8d
1 parent 92edc91
commit de77c8d
Showing 1 changed file with 10 additions and 0 deletions.
diff --git a/src/plugins/intel_cpu/src/config.cpp b/src/plugins/intel_cpu/src/config.cpp
@@ -296,6 +296,11 @@ void Config::readProperties(const ov::AnyMap& prop, const ModelType modelType) {
                 } else {
                     OPENVINO_THROW("invalid value");
                 }
+                // enforce inference precision to f16 for GNR acc verify
+                if (mayiuse(avx512_core_amx_fp16)) {
+                    // inferencePrecision = ov::element::f16;
+                    inferencePrecision = ov::element::bf16;
+                }
             } catch (ov::Exception&) {
                 OPENVINO_THROW("Wrong value ",
                                val.as<std::string>(),
@@ -401,6 +406,11 @@ void Config::readProperties(const ov::AnyMap& prop, const ModelType modelType) {
         } else {
             inferencePrecision = ov::element::undefined;
         }
+        // enforce inference precision to f16 for GNR acc verify
+        if (mayiuse(avx512_core_amx_fp16)) {
+            // inferencePrecision = ov::element::f16;
+            inferencePrecision = ov::element::bf16;
+        }
     }
     // enable ACL fast math in PERFORMANCE mode
 #if defined(OV_CPU_WITH_ACL)