Able to run quantized model but outputs are wrong

Author: BujSet

samples/modules/executorch/arm-fvp-tutorials/models/mv2/prj.conf

@@ -27,7 +27,7 @@ CONFIG_CBPRINTF_FP_SUPPORT=y
 CONFIG_CODE_DATA_RELOCATION=y
 
 # Add model specific configs, mv2 needs at least ~11 MB, quantized needs 12MB?
-CONFIG_EXECUTORCH_METHOD_ALLOCATOR_POOL_SIZE=11534336
-CONFIG_EXECUTORCH_INCLUDE_ARM_QUANTIZE_LIBS=n
-#CONFIG_EXECUTORCH_METHOD_ALLOCATOR_POOL_SIZE=12582912
+#CONFIG_EXECUTORCH_METHOD_ALLOCATOR_POOL_SIZE=11534336
+CONFIG_EXECUTORCH_INCLUDE_ARM_QUANTIZE_LIBS=y
+CONFIG_EXECUTORCH_METHOD_ALLOCATOR_POOL_SIZE=12582912
 

samples/modules/executorch/arm-fvp-tutorials/models/mv2/src/mv2_output_verifier.cpp

@@ -120,6 +120,48 @@ MV2OutputVerifier::_get_class(size_t idx, char *name_buf, size_t buf_len) {
  }
  }
 }
+void
+MV2OutputVerifier::_print(std::vector<executorch::runtime::EValue>& outputs) {
+ // Print the outputs.
+ ET_LOG(Info, "Printing outputs.");
+ for (int i = 0; i < outputs.size(); ++i) {
+ if (outputs[i].isTensor()) {
+ executorch::aten::Tensor tensor = outputs[i].toTensor();
+ // The output might be collected and parsed so printf() is used instead
+ // of ET_LOG() here
+ for (int j = 0; j < tensor.numel(); ++j) {
+ if (tensor.scalar_type() == executorch::aten::ScalarType::Int) {
+ printf(
+ "Output[%d][%d]: (int) %d\n",
+ i,
+ j,
+ tensor.const_data_ptr<int>()[j]);
+ } else if (tensor.scalar_type() == executorch::aten::ScalarType::Float) {
+ printf(
+ "Output[%d][%d]: (float) %f\n",
+ i,
+ j,
+ tensor.const_data_ptr<float>()[j]);
+ } else if (tensor.scalar_type() == executorch::aten::ScalarType::Char) {
+ printf(
+ "Output[%d][%d]: (char) %d\n",
+ i,
+ j,
+ tensor.const_data_ptr<int8_t>()[j]);
+ } else if (tensor.scalar_type() == executorch::aten::ScalarType::Bool) {
+ printf(
+ "Output[%d][%d]: (bool) %s (0x%x)\n",
+ i,
+ j,
+ tensor.const_data_ptr<int8_t>()[j] ? "true " : "false",
+ tensor.const_data_ptr<int8_t>()[j]);
+ }
+ }
+ } else {
+ printf("Output[%d]: Not Tensor\n", i);
+ }
+ }
+}
 
 int
 MV2OutputVerifier::verify(std::vector<executorch::runtime::EValue>& outputs) {

samples/modules/executorch/arm-fvp-tutorials/models/mv2/src/mv2_output_verifier.hpp

@@ -17,8 +17,10 @@ class MV2OutputVerifier : public OutputVerifier {
  int verify(std::vector<executorch::runtime::EValue>& outputs);
  private:
 float _max(std::vector<executorch::runtime::EValue>& outputs);
+void _print(std::vector<executorch::runtime::EValue>& outputs);
 void _softmax_in_place(std::vector<executorch::runtime::EValue>& outputs, float max_val);
 void _top_k(std::vector<executorch::runtime::EValue>& in_probs, size_t k, std::vector<float>& out_probs, std::vector<size_t>& out_idxs);
 void _get_class(size_t idx, char *name_buf, size_t buf_len);
+
 };
 #endif //__MV2_OUTPUT_VERIFICATION_H__