LayerNormalization failure of TensorRT 8.2.4.2

[lin-lcx]

Description

Does 8.2.4.2 support LayerNormalization?
With the same onnx model and conversion instructions, I successfully converted on 8.6.1.6, but not on 8.2.4.2. But I have to adapt to the 8.2.4.2 environment

[01/31/2024-10:15:34] [I] [TRT] No importer registered for op: LayerNormalization. Attempting to import as plugin.
[01/31/2024-10:15:34] [I] [TRT] Searching for plugin: LayerNormalization, plugin_version: 1, plugin_namespace: 
[01/31/2024-10:15:34] [E] [TRT] ModelImporter.cpp:773: While parsing node number 16 [LayerNormalization -> "/embeddings/norm/LayerNormalization_output_0"]:
[01/31/2024-10:15:34] [E] [TRT] ModelImporter.cpp:774: --- Begin node ---
[01/31/2024-10:15:34] [E] [TRT] ModelImporter.cpp:775: input: "/embeddings/patch_embeddings/Transpose_output_0"
input: "embeddings.norm.weight"
input: "embeddings.norm.bias"
output: "/embeddings/norm/LayerNormalization_output_0"
name: "/embeddings/norm/LayerNormalization"
op_type: "LayerNormalization"
attribute {
  name: "axis"
  i: -1
  type: INT
}
attribute {
  name: "epsilon"
  f: 1e-05
  type: FLOAT
}

Environment

TensorRT Version:8.2.4.2

NVIDIA GPU:2080ti

NVIDIA Driver Version:

CUDA Version:11.1

CUDNN Version:

[zerollzeng]

Could you please try latest 9.2/9.3 release? Thanks!

[lin-lcx]

Due to engineering environment restrictions, I must use the 8.2.4.2 environment. I am using trtexec

[zerollzeng]

IIRC most of the transformer support are added since TRT 8.6. And we won't back port feature back to old release.

[ttyio]

closing since no activity for more than 3 weeks, thanks all!

[ZeppelinSCB]

I'm facing the same situation where I have to use TRT 8.2, may I know how you solved the problem?

[sushilkhadkaanon]

Hi @lin-lcx @ZeppelinSCB , How did you guys solve the issue? I have the similar situation, I have to adapt to the TensorRT version 8.4.1,
Did you guys write a custom LayerNormalization plugin, if so Could you help me on that?