SDK 9.2 QDQ Model Compilation Accuracy Issue

[99999-LQL]

We encountered abnormal accuracy while compiling a QDQ model on SDK 9.2. Below is the configuration used:
Parameter Configuration:
tidl_tools_path: /SDK92/edgeai-tidl-tools/tidl_tools
artifacts_folder: /SDK92/edgeai-tidl-tools/model-artifacts/mobilenet_v2_lite_wt-v2_qat-v2-wc8-at8_20231120_model/
platform: J7
version: 7.2
tensor_bits: 8
debug_level: 2
max_num_subgraphs: 16
deny_list: ''
deny_list:layer_type: ''
deny_list:layer_name: ''
model_type: ''
accuracy_level: 0
advanced_options:calibration_frames: 1
advanced_options:calibration_iterations: 1
advanced_options:output_feature_16bit_names_list: ''
advanced_options:params_16bit_names_list: ''
advanced_options:mixed_precision_factor: -1
advanced_options:quantization_scale_type: 4
advanced_options:high_resolution_optimization: 0
advanced_options:pre_batchnorm_fold: 1
ti_internal_nc_flag: 1601
advanced_options:activation_clipping: 1
advanced_options:weight_clipping: 1
advanced_options:bias_calibration: 1
advanced_options:add_data_convert_ops: 3
advanced_options:channel_wise_quantization: 0
advanced_options:inference_mode: 0
advanced_options:num_cores: 1
advanced_options:prequantized_model: 1
Problems Encountered:

1. Compilation Errors: When compiling mobilenet_v2_lite_wt-v2_qat-v2-wc8-at8_20231120_model.onnx, the following error was reported:
   Input tensor name - x
   Output tensor name - 1631
   Graph Domain TO version : 18In TIDL_onnxRtImportInit subgraph_name=1631
   Layer 0, subgraph id 1631, name=1631
   Layer 1, subgraph id 1631, name=x
   In TIDL_runtimesOptimizeNet: LayerIndex = 315, dataIndex = 314
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Unable to merge Dequantize upwards - DQ without initializer?
   Error: Layer 14, /0_4/Conv:/0_4/Conv_output_0 is missing inputs in the network and cannot be topologically sorted
   Input 0: /0_4/DequantizeLinear_output_0, dataId=229
2. After simplifying onnx, the compilation is OK, but the precision is not aligned。
   Inference Image:：airshow.jpg
   Inference QDQ ONNX: 0 17.885122 warplane, military plan
   Inference bin：0 18.246786 warplane, military plane
   The precision between the QDQ ONNX model and the binary model is not aligned.

Request: Please investigate the cause of the dequantization issues during compilation and help resolve the accuracy misalignment after simplification.