CPUExecutionProvider
    -47 Add
    -47 LayerNormalization axis=-1 epsilon=9.999999974752427e-07 stash_type=1
    +47 SkipLayerNormalization epsilon=9.999999974752427e-07
    324 nodes, weights ad5df5f65509
    FuseSkipLayerNorm x47 71e1d8de
    _Fp16TokenEmbedding x2 0d702441
    _ScalarGatherToSlice, _SelectBeforeLayerNorm x1 c46fabc2
    _SelectBeforeLayerNorm x1 eba1fbae
    unstamped x273 fab046c2

CUDAExecutionProvider
    -47 Add
    -47 LayerNormalization axis=-1 epsilon=9.999999974752427e-07 stash_type=1
    +47 SkipLayerNormalization epsilon=9.999999974752427e-07
    324 nodes, weights ad5df5f65509
    FuseSkipLayerNorm x47 71e1d8de
    _Fp16TokenEmbedding x2 0d702441
    _ScalarGatherToSlice, _SelectBeforeLayerNorm x1 c46fabc2
    _SelectBeforeLayerNorm x1 eba1fbae
    unstamped x273 fab046c2

CoreMLExecutionProvider
    +120 MatMul
    +96 Reshape
    +96 Slice
    +96 Transpose perm=(0, 2, 1, 3)
    +72 Add
    +25 Mul
    -24 Attention is_causal=0 kv_num_heads=16 q_num_heads=16 qk_matmul_output_mode=0 softcap=0.0
    +24 Softmax axis=-1
    +24 Transpose perm=(0, 1, 3, 2)
    -1 ReduceL2 keepdims=1 noop_with_empty_axes=0
    +1 ReduceSum keepdims=1
    +1 Sqrt
    901 nodes, weights b76788abc27a
    DecomposeAttention x312 f1a1e3bb
    DecomposeReduceL2 x3 2f55de2f
    SplitLargeReduction x264 043e0ad8
    _Fp16TokenEmbedding x2 0d702441
    _ScalarGatherToSlice, _SelectBeforeLayerNorm x1 c46fabc2
    _SelectBeforeLayerNorm x1 eba1fbae
    unstamped x318 921faf75

MIGraphXExecutionProvider
    +96 Reshape
    +96 Transpose perm=(0, 2, 1, 3)
    +48 MatMul
    -24 Attention is_causal=0 kv_num_heads=16 q_num_heads=16 qk_matmul_output_mode=0 softcap=0.0
    +24 Mul
    +24 Softmax axis=-1
    +24 Transpose perm=(0, 1, 3, 2)
    659 nodes, weights b616a5f26154
    DecomposeAttention x312 f1a1e3bb
    _Fp16TokenEmbedding x2 0d702441
    _ScalarGatherToSlice, _SelectBeforeLayerNorm x1 c46fabc2
    _SelectBeforeLayerNorm x1 eba1fbae
    unstamped x343 d978fcc3

NvTensorRTRTXExecutionProvider
    +96 Reshape
    +96 Transpose perm=(0, 2, 1, 3)
    +48 MatMul
    -24 Attention is_causal=0 kv_num_heads=16 q_num_heads=16 qk_matmul_output_mode=0 softcap=0.0
    +24 Mul
    +24 Softmax axis=-1
    +24 Transpose perm=(0, 1, 3, 2)
    659 nodes, weights b616a5f26154
    DecomposeAttention x312 f1a1e3bb
    _Fp16TokenEmbedding x2 0d702441
    _ScalarGatherToSlice, _SelectBeforeLayerNorm x1 c46fabc2
    _SelectBeforeLayerNorm x1 eba1fbae
    unstamped x343 d978fcc3

OpenVINOExecutionProvider
    +96 Reshape
    +96 Transpose perm=(0, 2, 1, 3)
    +48 MatMul
    -24 Attention is_causal=0 kv_num_heads=16 q_num_heads=16 qk_matmul_output_mode=0 softcap=0.0
    +24 Mul
    +24 Softmax axis=-1
    +24 Transpose perm=(0, 1, 3, 2)
    659 nodes, weights b616a5f26154
    DecomposeAttention x312 f1a1e3bb
    _Fp16TokenEmbedding x2 0d702441
    _ScalarGatherToSlice, _SelectBeforeLayerNorm x1 c46fabc2
    _SelectBeforeLayerNorm x1 eba1fbae
    unstamped x343 d978fcc3

RKNPU static
    +336 Slice
    +288 MatMul
    +264 Add
    +96 Reshape
    +96 Transpose perm=(0, 2, 1, 3)
    +72 Mul
    -24 Attention is_causal=0 kv_num_heads=16 q_num_heads=16 qk_matmul_output_mode=0 softcap=0.0
    +24 Div
    +24 Erf
    -24 Gelu approximate=tanh
    +24 Softmax axis=-1
    +24 Transpose perm=(0, 1, 3, 2)
    -1 Gather axis=0
    1570 nodes, weights 8164b384695f
    contract {"dims":[{}],"embedding":"text.token_embedding.weight_fp16"}
    opset 19
    DecomposeAttention x312 f1a1e3bb
    DecomposeGelu x120 388372f5
    SplitLargeReduction x912 e32053cc
    _Fp16TokenEmbedding x1 ee8d996d
    _ScalarGatherToSlice, _SelectBeforeLayerNorm x1 c46fabc2
    _SelectBeforeLayerNorm x1 eba1fbae
    unstamped x223 0e00ace1

TensorrtExecutionProvider
    +96 Reshape
    +96 Transpose perm=(0, 2, 1, 3)
    +48 MatMul
    -24 Attention is_causal=0 kv_num_heads=16 q_num_heads=16 qk_matmul_output_mode=0 softcap=0.0
    +24 Mul
    +24 Softmax axis=-1
    +24 Transpose perm=(0, 1, 3, 2)
    659 nodes, weights b616a5f26154
    DecomposeAttention x312 f1a1e3bb
    _Fp16TokenEmbedding x2 0d702441
    _ScalarGatherToSlice, _SelectBeforeLayerNorm x1 c46fabc2
    _SelectBeforeLayerNorm x1 eba1fbae
    unstamped x343 d978fcc3
