[NPU] Add NF4 precision support in NPU plugin

Signed-off-by: Stefania Hergane <stefania-persida.hergane@intel.com>
openvinotoolkit · Dec 4, 2024 · 83d7867 · 83d7867
1 parent 571e98d
commit 83d7867
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 2 deletions.
diff --git a/src/plugins/intel_npu/src/backend/src/zero_infer_request.cpp b/src/plugins/intel_npu/src/backend/src/zero_infer_request.cpp
@@ -647,6 +647,8 @@ void ZeroInferRequest::check_network_precision(const ov::element::Type_t precisi
         break;
     case ov::element::Type_t::bf16:
         break;
+    case ov::element::Type_t::nf4:
+        break;
     case ov::element::Type_t::u4:
         break;
     case ov::element::Type_t::i4:
@@ -670,8 +672,9 @@ void ZeroInferRequest::check_network_precision(const ov::element::Type_t precisi
     case ov::element::Type_t::f64:
         break;
     default:
-        OPENVINO_THROW("Unsupported tensor precision: " + ov::element::Type(precision).get_type_name() +
-                       "! Supported precisions: FP32, FP16, BF16, U4, I4, U8, I8, U16, I16, U32, I32, U64, I64, FP64");
+        OPENVINO_THROW(
+            "Unsupported tensor precision: " + ov::element::Type(precision).get_type_name() +
+            "! Supported precisions: FP32, FP16, BF16, NF4, U4, I4, U8, I8, U16, I16, U32, I32, U64, I64, FP64");
     }
 }
 

diff --git a/src/plugins/intel_npu/src/compiler_adapter/src/driver_compiler_adapter.cpp b/src/plugins/intel_npu/src/compiler_adapter/src/driver_compiler_adapter.cpp
@@ -81,6 +81,8 @@ std::string ovPrecisionToLegacyPrecisionString(const ov::element::Type& precisio
         return "FP64";
     case ov::element::Type_t::bf16:
         return "BF16";
+    case ov::element::Type_t::nf4:
+        return "NF4";
     case ov::element::Type_t::i4:
         return "I4";
     case ov::element::Type_t::i8: