导出int8模型报错

[MissMIA0]

大佬你好，我在onnx加入了plugin之后运行convert_to_trt.py希望得到int8的trt模型，出现错误（导出fp16正常）。如果不加入plugin直接将onnx转trt int8不会报错。是我的做法有问题吗 报错如下： [08/31/2024-01:14:44] [TRT] [V] LayerNorm-0 (PluginV2[0x80000019]) profiling completed in 0.0047566 seconds. Fastest Tactic: 0x0000000000000000 Time: 0.00550033 [08/31/2024-01:14:44] [TRT] [V] >>>>>>>>>>>>>>> Chose Runner Type: PluginV2 Tactic: 0x0000000000000000 [08/31/2024-01:14:44] [TRT] [V] *************** Autotuning format combination: Int8(36864,36864:4,144,1), Int8(1), Int8(1) -> Int8(36864,36864:4,144,1) *************** [08/31/2024-01:14:44] [TRT] [V] --------------- Timing Runner: LayerNorm-0 (PluginV2[0x80000019]) [08/31/2024-01:14:44] [TRT] [F] Assertion failed: [computeLayerNormQDQ] Unsupport hidden dimension 144 C:_src\plugin\layerNormPlugin\layerNormKernel.cu:295 Aborting...