CPUExecutionProvider
    -4 Add
    -3 LayerNormalization axis=-1 epsilon=9.999999747378752e-06 stash_type=1
    +3 SkipLayerNormalization epsilon=9.999999747378752e-06
    -1 LayerNormalization axis=-1 epsilon=9.999999974752427e-07 stash_type=1
    +1 SkipLayerNormalization epsilon=9.999999974752427e-07
    178 nodes, weights 1c71e1598778
    FuseBatchNormIntoConv x3 cafd8b4b
    FuseSkipLayerNorm, LayerNormBiasFusion, LayerNormFusion x4 dfce4419
    LayerNormBiasFusion, LayerNormFusion x1 2a80a1b2
    _FlattenToSqueeze x1 f6285ec7
    _FoldBiasAdd x54 a455ec75
    _FuseErfGelu x13 837b6556
    _UnflattenToUnsqueeze x1 1f22e914
    unstamped x101 48bb1c11

CUDAExecutionProvider
    -4 Add
    -3 LayerNormalization axis=-1 epsilon=9.999999747378752e-06 stash_type=1
    +3 SkipLayerNormalization epsilon=9.999999747378752e-06
    -1 ArgMax axis=2 keepdims=0
    -1 LayerNormalization axis=-1 epsilon=9.999999974752427e-07 stash_type=1
    -1 ReduceMax keepdims=1
    +1 SkipLayerNormalization epsilon=9.999999974752427e-07
    +1 Squeeze
    +1 TopK axis=2 largest=1 sorted=1
    178 nodes, weights 1c71e1598778
    FuseBatchNormIntoConv x3 cafd8b4b
    FuseGreedyCtcTopK x2 60800ef1
    FuseSkipLayerNorm, LayerNormBiasFusion, LayerNormFusion x4 dfce4419
    LayerNormBiasFusion, LayerNormFusion x1 2a80a1b2
    _FlattenToSqueeze x1 f6285ec7
    _FoldBiasAdd x54 a455ec75
    _FuseErfGelu x13 837b6556
    _UnflattenToUnsqueeze x1 1f22e914
    unstamped x99 543cce05

CoreMLExecutionProvider
    -1 Transpose perm=(0, 3, 1, 2)
    181 nodes, weights 1c71e1598778
    FuseBatchNormIntoConv x3 cafd8b4b
    LayerNormBiasFusion, LayerNormFusion x5 2fcd7a68
    _FlattenToSqueeze x1 f6285ec7
    _FoldBiasAdd x54 a455ec75
    _FuseErfGelu x13 837b6556
    _UnflattenToUnsqueeze x1 ea488eee
    unstamped x104 77530022

MIGraphXExecutionProvider
    -2 Conv auto_pad=SAME_UPPER dilations=(1, 1) group=1 kernel_shape=(2, 2) strides=(1, 1)
    +2 Conv dilations=(1, 1) group=1 kernel_shape=(3, 3) pads=(1, 1, 1, 1) strides=(1, 1)
    182 nodes, weights 6b83d5a75df4
    FuseBatchNormIntoConv x3 cafd8b4b
    LayerNormBiasFusion, LayerNormFusion x5 2fcd7a68
    SymmetrizeConvPads, _FoldBiasAdd x2 c440deb2
    _FlattenToSqueeze x1 f6285ec7
    _FoldBiasAdd x52 f35bd2c4
    _FuseErfGelu x13 837b6556
    _UnflattenToUnsqueeze x1 ea488eee
    unstamped x105 f6fa16d5

NvTensorRTRTXExecutionProvider
    -1 ArgMax axis=2 keepdims=0
    -1 ReduceMax keepdims=1
    +1 Squeeze
    +1 TopK axis=2 largest=1 sorted=1
    182 nodes, weights 1c71e1598778
    FuseBatchNormIntoConv x3 cafd8b4b
    FuseGreedyCtcTopK x2 60800ef1
    LayerNormBiasFusion, LayerNormFusion x5 2fcd7a68
    _FlattenToSqueeze x1 f6285ec7
    _FoldBiasAdd x54 a455ec75
    _FuseErfGelu x13 837b6556
    _UnflattenToUnsqueeze x1 ea488eee
    unstamped x103 e65db882

OpenVINOExecutionProvider
    -1 ArgMax axis=2 keepdims=0
    -1 Cast to=6
    -1 Exp
    -1 Reciprocal
    -1 ReduceMax keepdims=1
    -1 ReduceSum keepdims=0
    +1 Softmax axis=2
    -1 Sub
    176 nodes, weights 1c71e1598778
    FuseBatchNormIntoConv x3 cafd8b4b
    LayerNormBiasFusion, LayerNormFusion x5 2fcd7a68
    _FlattenToSqueeze x1 f6285ec7
    _FoldBiasAdd x54 a455ec75
    _FuseErfGelu x13 837b6556
    _UnflattenToUnsqueeze x1 ea488eee
    unstamped x99 d09d18d3

RKNPU width=2048
    +26 Mul
    +14 Add
    +13 Div
    +13 Erf
    -13 Gelu approximate=none
    +10 Transpose perm=(0, 3, 2, 1)
    -2 Sub
    -1 ArgMax axis=2 keepdims=0
    -1 Cast to=1
    -1 Cast to=6
    -1 Concat axis=1
    +1 Concat axis=3
    +1 Conv auto_pad=NOTSET dilations=(1, 1) group=1 kernel_shape=(3, 3) pads=(1, 1, 1, 1) strides=(2, 2)
    -1 Exp
    -1 Reciprocal
    -1 ReduceMax keepdims=1
    -1 ReduceSum keepdims=0
    +1 Split axis=1
    +1 Transpose perm=(0, 2, 1)
    -1 Transpose perm=(0, 3, 1, 2)
    +1 Unsqueeze
    239 nodes, weights d10aaa3f70ba
    config {'mean_values': [[127.5, 127.5, 127.5]], 'std_values': [[1.0, 1.0, 1.0]]}
    contract {"dims":[{"width":2048},{"width":224},{"width":320},{"width":448},{"width":640},{"width":1280}]}
    opset 19
    DecomposeGelu, _FuseErfGelu x65 a950e748
    FoldConcatIntoConv, _FoldBiasAdd x3 40db084c
    FuseBatchNormIntoConv x3 cafd8b4b
    LayerNormBiasFusion, LayerNormFusion x5 2fcd7a68
    _FlattenToSqueeze x1 f6285ec7
    _FoldBiasAdd x53 176f5c08
    _UnflattenToUnsqueeze x1 ea488eee
    unstamped x108 aec73a75

TensorrtExecutionProvider
    -1 ArgMax axis=2 keepdims=0
    -1 ReduceMax keepdims=1
    +1 Squeeze
    +1 TopK axis=2 largest=1 sorted=1
    182 nodes, weights 1c71e1598778
    FuseBatchNormIntoConv x3 cafd8b4b
    FuseGreedyCtcTopK x2 60800ef1
    LayerNormBiasFusion, LayerNormFusion x5 2fcd7a68
    _FlattenToSqueeze x1 f6285ec7
    _FoldBiasAdd x54 a455ec75
    _FuseErfGelu x13 837b6556
    _UnflattenToUnsqueeze x1 ea488eee
    unstamped x103 e65db882
