[onert] Support block quantization in CircleLoader (Samsung#13884)

This commit updates CircleLoader and BaseLoader to support block quantization. ONE-DCO-1.0-Signed-off-by: Hyeongseok Oh <[email protected]>
mhs4670go · Sep 3, 2024 · b104d9c · b104d9c
1 parent 8593953
commit b104d9c
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 1 deletion.
diff --git a/runtime/onert/core/src/loader/BaseLoader.h b/runtime/onert/core/src/loader/BaseLoader.h
@@ -94,7 +94,7 @@ template <typename LoaderDomain> class BaseLoader
 
   // Helper functions
   ir::Activation convertActivation(ActivationFunctionType type);
-  ir::DataType tensorTypeToDataType(TensorType type);
+  virtual ir::DataType tensorTypeToDataType(const TensorType type);
   ir::OperandIndex tensorIdxToOperandIdx(int32_t tensorIdx);
   flexbuffers::Map getCustomOpAttrMap(const Operator *op);
 

diff --git a/runtime/onert/core/src/loader/CircleLoader.cc b/runtime/onert/core/src/loader/CircleLoader.cc
@@ -87,6 +87,17 @@ class CircleLoader final : public loader::BaseLoader<LoaderDomain>
     }
   }
 
+protected:
+  ir::DataType tensorTypeToDataType(const TensorType type) override
+  {
+    if (type == TensorType::TensorType_Q4_0)
+      return ir::DataType::QUANT_GGML_Q4_0;
+    if (type == TensorType::TensorType_Q8_0)
+      return ir::DataType::QUANT_GGML_Q8_0;
+
+    return BaseLoader::tensorTypeToDataType(type);
+  }
+
 private:
   std::unique_ptr<ir::Graph> loadSubgraph(const circle::SubGraph *circle_subg) override
   {