BERT is not available in ZCU102？

[zhendexuebuhui]

I am trying to run the quantized bert-tiny ONNX model based on the VOE of ZCU102, through the C++ API. There is an error of "indices element out of data bounds". I don't think there is any problem with the input tensor. Is there any incompatibility problem?

Environment

• ZCU102 DPUCZDX8G
• Vitis-AI 3.0（xilinx-zcu102-dpu-v2022.2-v3.0.0.img）
• pt_bert-tiny_SQuADv1.1_384_453M_3.0

Error

Gather Node indices element out of data bounds

2025-01-06 17:12:16.610315550 [E:onnxruntime:, sequential_executor.cc:369 Execute] Non-zero status code returned while running Gather node. Name:'Gather_31' Status Message: indices element out of data bounds, idx=281472786415824 must be within the inclusive range [-2,1]
terminate called after throwing an instance of 'Ort::Exception'
what(): Non-zero status code returned while running Gather node. Name:'Gather_31' Status Message: indices element out of data bounds, idx=281472786415824 must be within the inclusive range [-2,1]
Aborted

Model in Netron

Tried

1. Print the value of debugging indices(that is, input_2). Check without error

2. Re-train bert-tiny, quantify, transform ONNX, and run again. Still reporting the error

3. Check .onnx model by onnx.checker.check_model(). Check without error

[zhendexuebuhui]

The same error occurred in pt_bert-base_SQuADv1.1_384_70.66G_3.0.