program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.1"}})]
{
    func n12<ios18>(tensor<fp16, [10, 1, 768, 2]> conv, tensor<fp16, [7, 1, 4, 31, 96]> k, tensor<fp16, [1, 1, 1, 52]> m1, tensor<fp16, [1, 1, 1, 26]> m2, tensor<fp16, [1, 1, 12, 43]> mask, tensor<fp16, [1, 128, 104]> mel, tensor<fp16, [7, 1, 4, 31, 96]> v) {
            tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 128, 104]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = mel)[name = string("input_1_cast_fp16")];
            string x_1_pad_type_0 = const()[name = string("x_1_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> x_1_pad_0 = const()[name = string("x_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> x_1_strides_0 = const()[name = string("x_1_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> x_1_dilations_0 = const()[name = string("x_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_1_groups_0 = const()[name = string("x_1_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_in_weight_to_fp16 = const()[name = string("downsample_conv_in_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [256]> downsample_conv_in_bias_to_fp16 = const()[name = string("downsample_conv_in_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4736)))];
            tensor<fp16, [1, 256, 64, 52]> x_1_cast_fp16 = conv(bias = downsample_conv_in_bias_to_fp16, dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = downsample_conv_in_weight_to_fp16, x = input_1_cast_fp16)[name = string("x_1_cast_fp16")];
            fp16 var_31_to_fp16 = const()[name = string("op_31_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 64, 52]> var_32_cast_fp16 = mul(x = x_1_cast_fp16, y = var_31_to_fp16)[name = string("op_32_cast_fp16")];
            tensor<fp16, [1, 256, 64, 52]> var_35_cast_fp16 = tanh(x = var_32_cast_fp16)[name = string("op_35_cast_fp16")];
            fp16 var_36_to_fp16 = const()[name = string("op_36_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 64, 52]> var_37_cast_fp16 = add(x = var_35_cast_fp16, y = var_36_to_fp16)[name = string("op_37_cast_fp16")];
            tensor<fp16, [1, 256, 64, 52]> h_1_cast_fp16 = mul(x = var_32_cast_fp16, y = var_37_cast_fp16)[name = string("h_1_cast_fp16")];
            tensor<fp16, [1, 256, 64, 52]> input_3_cast_fp16 = mul(x = h_1_cast_fp16, y = m1)[name = string("input_3_cast_fp16")];
            string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(256)];
            tensor<int32, [2]> input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_mid_weight_to_fp16 = const()[name = string("downsample_conv_mid_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5312)))];
            tensor<fp16, [256]> downsample_conv_mid_bias_to_fp16 = const()[name = string("downsample_conv_mid_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))];
            tensor<fp16, [1, 256, 32, 26]> input_5_cast_fp16 = conv(bias = downsample_conv_mid_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = downsample_conv_mid_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string x_3_pad_type_0 = const()[name = string("x_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_3_strides_0 = const()[name = string("x_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_3_pad_0 = const()[name = string("x_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_3_dilations_0 = const()[name = string("x_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_3_groups_0 = const()[name = string("x_3_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> downsample_lin_mid_weight_to_fp16 = const()[name = string("downsample_lin_mid_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10560)))];
            tensor<fp16, [256]> downsample_lin_mid_bias_to_fp16 = const()[name = string("downsample_lin_mid_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141696)))];
            tensor<fp16, [1, 256, 32, 26]> x_3_cast_fp16 = conv(bias = downsample_lin_mid_bias_to_fp16, dilations = x_3_dilations_0, groups = x_3_groups_0, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = x_3_strides_0, weight = downsample_lin_mid_weight_to_fp16, x = input_5_cast_fp16)[name = string("x_3_cast_fp16")];
            fp16 var_54_to_fp16 = const()[name = string("op_54_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 32, 26]> var_55_cast_fp16 = mul(x = x_3_cast_fp16, y = var_54_to_fp16)[name = string("op_55_cast_fp16")];
            tensor<fp16, [1, 256, 32, 26]> var_58_cast_fp16 = tanh(x = var_55_cast_fp16)[name = string("op_58_cast_fp16")];
            fp16 var_59_to_fp16 = const()[name = string("op_59_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 32, 26]> var_60_cast_fp16 = add(x = var_58_cast_fp16, y = var_59_to_fp16)[name = string("op_60_cast_fp16")];
            tensor<fp16, [1, 256, 32, 26]> h_3_cast_fp16 = mul(x = var_55_cast_fp16, y = var_60_cast_fp16)[name = string("h_3_cast_fp16")];
            tensor<fp16, [1, 256, 32, 26]> input_7_cast_fp16 = mul(x = h_3_cast_fp16, y = m2)[name = string("input_7_cast_fp16")];
            string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(256)];
            tensor<int32, [2]> input_9_dilations_0 = const()[name = string("input_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_end_weight_to_fp16 = const()[name = string("downsample_conv_end_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142272)))];
            tensor<fp16, [256]> downsample_conv_end_bias_to_fp16 = const()[name = string("downsample_conv_end_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146944)))];
            tensor<fp16, [1, 256, 16, 13]> input_9_cast_fp16 = conv(bias = downsample_conv_end_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = downsample_conv_end_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
            string x_5_pad_type_0 = const()[name = string("x_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_5_strides_0 = const()[name = string("x_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_5_pad_0 = const()[name = string("x_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_5_dilations_0 = const()[name = string("x_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_5_groups_0 = const()[name = string("x_5_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> downsample_lin_end_weight_to_fp16 = const()[name = string("downsample_lin_end_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147520)))];
            tensor<fp16, [256]> downsample_lin_end_bias_to_fp16 = const()[name = string("downsample_lin_end_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278656)))];
            tensor<fp16, [1, 256, 16, 13]> x_5_cast_fp16 = conv(bias = downsample_lin_end_bias_to_fp16, dilations = x_5_dilations_0, groups = x_5_groups_0, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = x_5_strides_0, weight = downsample_lin_end_weight_to_fp16, x = input_9_cast_fp16)[name = string("x_5_cast_fp16")];
            fp16 var_77_to_fp16 = const()[name = string("op_77_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 16, 13]> var_78_cast_fp16 = mul(x = x_5_cast_fp16, y = var_77_to_fp16)[name = string("op_78_cast_fp16")];
            tensor<fp16, [1, 256, 16, 13]> var_81_cast_fp16 = tanh(x = var_78_cast_fp16)[name = string("op_81_cast_fp16")];
            fp16 var_82_to_fp16 = const()[name = string("op_82_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 16, 13]> var_83_cast_fp16 = add(x = var_81_cast_fp16, y = var_82_to_fp16)[name = string("op_83_cast_fp16")];
            tensor<fp16, [1, 256, 16, 13]> h_cast_fp16 = mul(x = var_78_cast_fp16, y = var_83_cast_fp16)[name = string("h_cast_fp16")];
            tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 4096, 13])];
            tensor<fp16, [1, 4096, 13]> input_11_cast_fp16 = reshape(shape = concat_0, x = h_cast_fp16)[name = string("input_11_cast_fp16")];
            string d_pad_type_0 = const()[name = string("d_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> d_strides_0 = const()[name = string("d_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> d_pad_0 = const()[name = string("d_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> d_dilations_0 = const()[name = string("d_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 d_groups_0 = const()[name = string("d_groups_0"), val = int32(1)];
            tensor<fp16, [768, 4096, 1]> downsample_lin_out_weight_to_fp16 = const()[name = string("downsample_lin_out_weight_to_fp16"), val = tensor<fp16, [768, 4096, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279232)))];
            tensor<fp16, [768]> downsample_lin_out_bias_to_fp16 = const()[name = string("downsample_lin_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6570752)))];
            tensor<fp16, [1, 768, 13]> d_cast_fp16 = conv(bias = downsample_lin_out_bias_to_fp16, dilations = d_dilations_0, groups = d_groups_0, pad = d_pad_0, pad_type = d_pad_type_0, strides = d_strides_0, weight = downsample_lin_out_weight_to_fp16, x = input_11_cast_fp16)[name = string("d_cast_fp16")];
            tensor<int32, [3]> var_107_begin_0 = const()[name = string("op_107_begin_0"), val = tensor<int32, [3]>([0, 0, 1])];
            tensor<int32, [3]> var_107_end_0 = const()[name = string("op_107_end_0"), val = tensor<int32, [3]>([1, 768, 13])];
            tensor<bool, [3]> var_107_end_mask_0 = const()[name = string("op_107_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> var_107_cast_fp16 = slice_by_index(begin = var_107_begin_0, end = var_107_end_0, end_mask = var_107_end_mask_0, x = d_cast_fp16)[name = string("op_107_cast_fp16")];
            tensor<int32, [3]> x_7_perm_0 = const()[name = string("x_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_127 = const()[name = string("op_127"), val = int32(-1)];
            bool var_128 = const()[name = string("op_128"), val = bool(true)];
            tensor<fp16, [1, 12, 768]> x_7_cast_fp16 = transpose(perm = x_7_perm_0, x = var_107_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [1, 12, 768]> var_720_cast_fp16 = abs(x = x_7_cast_fp16)[name = string("op_720_cast_fp16")];
            tensor<int32, [1]> var_722_axes_0 = const()[name = string("op_722_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_722_cast_fp16 = reduce_max(axes = var_722_axes_0, keep_dims = var_128, x = var_720_cast_fp16)[name = string("op_722_cast_fp16")];
            fp16 var_129_to_fp16 = const()[name = string("op_129_to_fp16"), val = fp16(0x1.a38p-14)];
            tensor<fp16, [1, 12, 1]> s_1_cast_fp16 = maximum(x = var_722_cast_fp16, y = var_129_to_fp16)[name = string("s_1_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_1_cast_fp16 = real_div(x = x_7_cast_fp16, y = s_1_cast_fp16)[name = string("y_1_cast_fp16")];
            fp16 var_130_promoted_to_fp16 = const()[name = string("op_130_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_725_cast_fp16 = pow(x = y_1_cast_fp16, y = var_130_promoted_to_fp16)[name = string("op_725_cast_fp16")];
            tensor<int32, [1]> var_727_axes_0 = const()[name = string("op_727_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_727_keep_dims_0 = const()[name = string("op_727_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_727_cast_fp16 = reduce_mean(axes = var_727_axes_0, keep_dims = var_727_keep_dims_0, x = var_725_cast_fp16)[name = string("op_727_cast_fp16")];
            fp32 var_728_epsilon_0 = const()[name = string("op_728_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_728_cast_fp16 = inverse(epsilon = var_728_epsilon_0, x = s_1_cast_fp16)[name = string("op_728_cast_fp16")];
            fp16 var_729_to_fp16 = const()[name = string("op_729_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_730_cast_fp16 = mul(x = var_728_cast_fp16, y = var_729_to_fp16)[name = string("op_730_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_731_cast_fp16 = real_div(x = var_730_cast_fp16, y = s_1_cast_fp16)[name = string("op_731_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_732_cast_fp16 = add(x = var_727_cast_fp16, y = var_731_cast_fp16)[name = string("op_732_cast_fp16")];
            fp32 var_733_epsilon_0 = const()[name = string("op_733_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_733_cast_fp16 = rsqrt(epsilon = var_733_epsilon_0, x = var_732_cast_fp16)[name = string("op_733_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_734_cast_fp16 = mul(x = y_1_cast_fp16, y = var_733_cast_fp16)[name = string("op_734_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_0_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6572352)))];
            tensor<fp16, [1, 12, 768]> input_13_cast_fp16 = mul(x = core_backbone_layers_0_operator_norm_weight_to_fp16, y = var_734_cast_fp16)[name = string("input_13_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_0_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6573952)))];
            tensor<fp16, [2304]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10112960)))];
            tensor<fp16, [1, 12, 2304]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_0_conv_in_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<int32, [3]> bcx_1_perm_0 = const()[name = string("bcx_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_1_begin_0 = const()[name = string("b_1_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_1_end_0 = const()[name = string("b_1_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_1_end_mask_0 = const()[name = string("b_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_1_cast_fp16 = transpose(perm = bcx_1_perm_0, x = linear_0_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [1, 768, 12]> b_1_cast_fp16 = slice_by_index(begin = b_1_begin_0, end = b_1_end_0, end_mask = b_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("b_1_cast_fp16")];
            tensor<int32, [3]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("c_1_cast_fp16")];
            tensor<int32, [3]> xx_1_begin_0 = const()[name = string("xx_1_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_1_end_0 = const()[name = string("xx_1_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_1_end_mask_0 = const()[name = string("xx_1_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_1_cast_fp16 = slice_by_index(begin = xx_1_begin_0, end = xx_1_end_0, end_mask = xx_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("xx_1_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_1_cast_fp16 = mul(x = b_1_cast_fp16, y = xx_1_cast_fp16)[name = string("new_1_cast_fp16")];
            tensor<int32, [4]> input_17_begin_0 = const()[name = string("input_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_17_end_0 = const()[name = string("input_17_end_0"), val = tensor<int32, [4]>([1, 1, 768, 2])];
            tensor<bool, [4]> input_17_end_mask_0 = const()[name = string("input_17_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_17_squeeze_mask_0 = const()[name = string("input_17_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_17_cast_fp16 = slice_by_index(begin = input_17_begin_0, end = input_17_end_0, end_mask = input_17_end_mask_0, squeeze_mask = input_17_squeeze_mask_0, x = conv)[name = string("input_17_cast_fp16")];
            tensor<int32, [6]> var_749_pad_0 = const()[name = string("op_749_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_749_mode_0 = const()[name = string("op_749_mode_0"), val = string("constant")];
            fp16 const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_749_cast_fp16 = pad(constant_val = const_4_to_fp16, mode = var_749_mode_0, pad = var_749_pad_0, x = new_1_cast_fp16)[name = string("op_749_cast_fp16")];
            tensor<int32, [6]> var_751_pad_0 = const()[name = string("op_751_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_751_mode_0 = const()[name = string("op_751_mode_0"), val = string("constant")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_751_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = var_751_mode_0, pad = var_751_pad_0, x = input_17_cast_fp16)[name = string("op_751_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_1_cast_fp16 = add(x = var_749_cast_fp16, y = var_751_cast_fp16)[name = string("win_1_cast_fp16")];
            tensor<int32, [3]> var_753_begin_0 = const()[name = string("op_753_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_753_end_0 = const()[name = string("op_753_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_753_end_mask_0 = const()[name = string("op_753_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_753_cast_fp16 = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = win_1_cast_fp16)[name = string("op_753_cast_fp16")];
            string var_759_pad_type_0 = const()[name = string("op_759_pad_type_0"), val = string("valid")];
            int32 var_759_groups_0 = const()[name = string("op_759_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_759_strides_0 = const()[name = string("op_759_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_759_pad_0 = const()[name = string("op_759_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_759_dilations_0 = const()[name = string("op_759_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_0_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10117632)))];
            tensor<fp16, [1, 768, 12]> var_759_cast_fp16 = conv(dilations = var_759_dilations_0, groups = var_759_groups_0, pad = var_759_pad_0, pad_type = var_759_pad_type_0, strides = var_759_strides_0, weight = core_backbone_layers_0_conv_conv_weight_to_fp16, x = win_1_cast_fp16)[name = string("op_759_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_761_cast_fp16 = mul(x = c_1_cast_fp16, y = var_759_cast_fp16)[name = string("op_761_cast_fp16")];
            tensor<int32, [3]> input_19_perm_0 = const()[name = string("input_19_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_0_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10122304)))];
            tensor<fp16, [768]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11302016)))];
            tensor<fp16, [1, 12, 768]> input_19_cast_fp16 = transpose(perm = input_19_perm_0, x = var_761_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [1, 12, 768]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_conv_out_proj_weight_to_fp16, x = input_19_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_9_cast_fp16 = add(x = x_7_cast_fp16, y = linear_1_cast_fp16)[name = string("x_9_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_767_cast_fp16 = abs(x = x_9_cast_fp16)[name = string("op_767_cast_fp16")];
            tensor<int32, [1]> var_769_axes_0 = const()[name = string("op_769_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_769_cast_fp16 = reduce_max(axes = var_769_axes_0, keep_dims = var_128, x = var_767_cast_fp16)[name = string("op_769_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_3_cast_fp16 = maximum(x = var_769_cast_fp16, y = var_129_to_fp16)[name = string("s_3_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_5_cast_fp16 = real_div(x = x_9_cast_fp16, y = s_3_cast_fp16)[name = string("y_5_cast_fp16")];
            fp16 var_130_promoted_1_to_fp16 = const()[name = string("op_130_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_772_cast_fp16 = pow(x = y_5_cast_fp16, y = var_130_promoted_1_to_fp16)[name = string("op_772_cast_fp16")];
            tensor<int32, [1]> var_774_axes_0 = const()[name = string("op_774_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_774_keep_dims_0 = const()[name = string("op_774_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_774_cast_fp16 = reduce_mean(axes = var_774_axes_0, keep_dims = var_774_keep_dims_0, x = var_772_cast_fp16)[name = string("op_774_cast_fp16")];
            fp32 var_775_epsilon_0 = const()[name = string("op_775_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_775_cast_fp16 = inverse(epsilon = var_775_epsilon_0, x = s_3_cast_fp16)[name = string("op_775_cast_fp16")];
            fp16 var_776_to_fp16 = const()[name = string("op_776_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_777_cast_fp16 = mul(x = var_775_cast_fp16, y = var_776_to_fp16)[name = string("op_777_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_778_cast_fp16 = real_div(x = var_777_cast_fp16, y = s_3_cast_fp16)[name = string("op_778_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_779_cast_fp16 = add(x = var_774_cast_fp16, y = var_778_cast_fp16)[name = string("op_779_cast_fp16")];
            fp32 var_780_epsilon_0 = const()[name = string("op_780_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_780_cast_fp16 = rsqrt(epsilon = var_780_epsilon_0, x = var_779_cast_fp16)[name = string("op_780_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_781_cast_fp16 = mul(x = y_5_cast_fp16, y = var_780_cast_fp16)[name = string("op_781_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_0_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11303616)))];
            tensor<fp16, [1, 12, 768]> input_21_cast_fp16 = mul(x = core_backbone_layers_0_ffn_norm_weight_to_fp16, y = var_781_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_0_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11305216)))];
            tensor<fp16, [1792]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14057792)))];
            tensor<fp16, [1, 12, 1792]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w1_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_0_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14061440)))];
            tensor<fp16, [1, 12, 1792]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w3_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_3_cast_fp16")];
            fp16 var_790_to_fp16 = const()[name = string("op_790_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_791_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_790_to_fp16)[name = string("op_791_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_794_cast_fp16 = tanh(x = var_791_cast_fp16)[name = string("op_794_cast_fp16")];
            fp16 var_795_to_fp16 = const()[name = string("op_795_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_796_cast_fp16 = add(x = var_794_cast_fp16, y = var_795_to_fp16)[name = string("op_796_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_797_cast_fp16 = mul(x = var_791_cast_fp16, y = var_796_cast_fp16)[name = string("op_797_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_23_cast_fp16 = mul(x = var_797_cast_fp16, y = linear_3_cast_fp16)[name = string("input_23_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_0_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16814016)))];
            tensor<fp16, [1, 12, 768]> linear_4_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w2_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_13_cast_fp16 = add(x = x_9_cast_fp16, y = linear_4_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_803_cast_fp16 = abs(x = x_13_cast_fp16)[name = string("op_803_cast_fp16")];
            tensor<int32, [1]> var_805_axes_0 = const()[name = string("op_805_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_805_cast_fp16 = reduce_max(axes = var_805_axes_0, keep_dims = var_128, x = var_803_cast_fp16)[name = string("op_805_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_5_cast_fp16 = maximum(x = var_805_cast_fp16, y = var_129_to_fp16)[name = string("s_5_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_7_cast_fp16 = real_div(x = x_13_cast_fp16, y = s_5_cast_fp16)[name = string("y_7_cast_fp16")];
            fp16 var_130_promoted_2_to_fp16 = const()[name = string("op_130_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_808_cast_fp16 = pow(x = y_7_cast_fp16, y = var_130_promoted_2_to_fp16)[name = string("op_808_cast_fp16")];
            tensor<int32, [1]> var_810_axes_0 = const()[name = string("op_810_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_810_keep_dims_0 = const()[name = string("op_810_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_810_cast_fp16 = reduce_mean(axes = var_810_axes_0, keep_dims = var_810_keep_dims_0, x = var_808_cast_fp16)[name = string("op_810_cast_fp16")];
            fp32 var_811_epsilon_0 = const()[name = string("op_811_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_811_cast_fp16 = inverse(epsilon = var_811_epsilon_0, x = s_5_cast_fp16)[name = string("op_811_cast_fp16")];
            fp16 var_812_to_fp16 = const()[name = string("op_812_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_813_cast_fp16 = mul(x = var_811_cast_fp16, y = var_812_to_fp16)[name = string("op_813_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_814_cast_fp16 = real_div(x = var_813_cast_fp16, y = s_5_cast_fp16)[name = string("op_814_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_815_cast_fp16 = add(x = var_810_cast_fp16, y = var_814_cast_fp16)[name = string("op_815_cast_fp16")];
            fp32 var_816_epsilon_0 = const()[name = string("op_816_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_816_cast_fp16 = rsqrt(epsilon = var_816_epsilon_0, x = var_815_cast_fp16)[name = string("op_816_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_817_cast_fp16 = mul(x = y_7_cast_fp16, y = var_816_cast_fp16)[name = string("op_817_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_1_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19566592)))];
            tensor<fp16, [1, 12, 768]> input_25_cast_fp16 = mul(x = core_backbone_layers_1_operator_norm_weight_to_fp16, y = var_817_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_1_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19568192)))];
            tensor<fp16, [1, 12, 2304]> linear_5_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_1_conv_in_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [3]> bcx_3_perm_0 = const()[name = string("bcx_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_5_begin_0 = const()[name = string("b_5_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_5_end_0 = const()[name = string("b_5_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_5_end_mask_0 = const()[name = string("b_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_3_cast_fp16 = transpose(perm = bcx_3_perm_0, x = linear_5_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [1, 768, 12]> b_5_cast_fp16 = slice_by_index(begin = b_5_begin_0, end = b_5_end_0, end_mask = b_5_end_mask_0, x = bcx_3_cast_fp16)[name = string("b_5_cast_fp16")];
            tensor<int32, [3]> c_3_begin_0 = const()[name = string("c_3_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_3_end_0 = const()[name = string("c_3_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_3_end_mask_0 = const()[name = string("c_3_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_3_cast_fp16 = slice_by_index(begin = c_3_begin_0, end = c_3_end_0, end_mask = c_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("c_3_cast_fp16")];
            tensor<int32, [3]> xx_3_begin_0 = const()[name = string("xx_3_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_3_end_0 = const()[name = string("xx_3_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_3_end_mask_0 = const()[name = string("xx_3_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_3_cast_fp16 = slice_by_index(begin = xx_3_begin_0, end = xx_3_end_0, end_mask = xx_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("xx_3_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_3_cast_fp16 = mul(x = b_5_cast_fp16, y = xx_3_cast_fp16)[name = string("new_3_cast_fp16")];
            tensor<int32, [4]> input_29_begin_0 = const()[name = string("input_29_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])];
            tensor<int32, [4]> input_29_end_0 = const()[name = string("input_29_end_0"), val = tensor<int32, [4]>([2, 1, 768, 2])];
            tensor<bool, [4]> input_29_end_mask_0 = const()[name = string("input_29_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_29_squeeze_mask_0 = const()[name = string("input_29_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_29_cast_fp16 = slice_by_index(begin = input_29_begin_0, end = input_29_end_0, end_mask = input_29_end_mask_0, squeeze_mask = input_29_squeeze_mask_0, x = conv)[name = string("input_29_cast_fp16")];
            tensor<int32, [6]> var_832_pad_0 = const()[name = string("op_832_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_832_mode_0 = const()[name = string("op_832_mode_0"), val = string("constant")];
            fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_832_cast_fp16 = pad(constant_val = const_6_to_fp16, mode = var_832_mode_0, pad = var_832_pad_0, x = new_3_cast_fp16)[name = string("op_832_cast_fp16")];
            tensor<int32, [6]> var_834_pad_0 = const()[name = string("op_834_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_834_mode_0 = const()[name = string("op_834_mode_0"), val = string("constant")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_834_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = var_834_mode_0, pad = var_834_pad_0, x = input_29_cast_fp16)[name = string("op_834_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_5_cast_fp16 = add(x = var_832_cast_fp16, y = var_834_cast_fp16)[name = string("win_5_cast_fp16")];
            tensor<int32, [3]> var_836_begin_0 = const()[name = string("op_836_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_836_end_0 = const()[name = string("op_836_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_836_end_mask_0 = const()[name = string("op_836_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_836_cast_fp16 = slice_by_index(begin = var_836_begin_0, end = var_836_end_0, end_mask = var_836_end_mask_0, x = win_5_cast_fp16)[name = string("op_836_cast_fp16")];
            string var_842_pad_type_0 = const()[name = string("op_842_pad_type_0"), val = string("valid")];
            int32 var_842_groups_0 = const()[name = string("op_842_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_842_strides_0 = const()[name = string("op_842_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_842_pad_0 = const()[name = string("op_842_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_842_dilations_0 = const()[name = string("op_842_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_1_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23107200)))];
            tensor<fp16, [1, 768, 12]> var_842_cast_fp16 = conv(dilations = var_842_dilations_0, groups = var_842_groups_0, pad = var_842_pad_0, pad_type = var_842_pad_type_0, strides = var_842_strides_0, weight = core_backbone_layers_1_conv_conv_weight_to_fp16, x = win_5_cast_fp16)[name = string("op_842_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_844_cast_fp16 = mul(x = c_3_cast_fp16, y = var_842_cast_fp16)[name = string("op_844_cast_fp16")];
            tensor<int32, [3]> input_31_perm_0 = const()[name = string("input_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_1_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23111872)))];
            tensor<fp16, [1, 12, 768]> input_31_cast_fp16 = transpose(perm = input_31_perm_0, x = var_844_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [1, 12, 768]> linear_6_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_conv_out_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_15_cast_fp16 = add(x = x_13_cast_fp16, y = linear_6_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_850_cast_fp16 = abs(x = x_15_cast_fp16)[name = string("op_850_cast_fp16")];
            tensor<int32, [1]> var_852_axes_0 = const()[name = string("op_852_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_852_cast_fp16 = reduce_max(axes = var_852_axes_0, keep_dims = var_128, x = var_850_cast_fp16)[name = string("op_852_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_7_cast_fp16 = maximum(x = var_852_cast_fp16, y = var_129_to_fp16)[name = string("s_7_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_11_cast_fp16 = real_div(x = x_15_cast_fp16, y = s_7_cast_fp16)[name = string("y_11_cast_fp16")];
            fp16 var_130_promoted_3_to_fp16 = const()[name = string("op_130_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_855_cast_fp16 = pow(x = y_11_cast_fp16, y = var_130_promoted_3_to_fp16)[name = string("op_855_cast_fp16")];
            tensor<int32, [1]> var_857_axes_0 = const()[name = string("op_857_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_857_keep_dims_0 = const()[name = string("op_857_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_857_cast_fp16 = reduce_mean(axes = var_857_axes_0, keep_dims = var_857_keep_dims_0, x = var_855_cast_fp16)[name = string("op_857_cast_fp16")];
            fp32 var_858_epsilon_0 = const()[name = string("op_858_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_858_cast_fp16 = inverse(epsilon = var_858_epsilon_0, x = s_7_cast_fp16)[name = string("op_858_cast_fp16")];
            fp16 var_859_to_fp16 = const()[name = string("op_859_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_860_cast_fp16 = mul(x = var_858_cast_fp16, y = var_859_to_fp16)[name = string("op_860_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_861_cast_fp16 = real_div(x = var_860_cast_fp16, y = s_7_cast_fp16)[name = string("op_861_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_862_cast_fp16 = add(x = var_857_cast_fp16, y = var_861_cast_fp16)[name = string("op_862_cast_fp16")];
            fp32 var_863_epsilon_0 = const()[name = string("op_863_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_863_cast_fp16 = rsqrt(epsilon = var_863_epsilon_0, x = var_862_cast_fp16)[name = string("op_863_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_864_cast_fp16 = mul(x = y_11_cast_fp16, y = var_863_cast_fp16)[name = string("op_864_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_1_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24291584)))];
            tensor<fp16, [1, 12, 768]> input_33_cast_fp16 = mul(x = core_backbone_layers_1_ffn_norm_weight_to_fp16, y = var_864_cast_fp16)[name = string("input_33_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_1_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24293184)))];
            tensor<fp16, [1, 12, 1792]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w1_weight_to_fp16, x = input_33_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_1_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27045760)))];
            tensor<fp16, [1, 12, 1792]> linear_8_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w3_weight_to_fp16, x = input_33_cast_fp16)[name = string("linear_8_cast_fp16")];
            fp16 var_873_to_fp16 = const()[name = string("op_873_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_874_cast_fp16 = mul(x = linear_7_cast_fp16, y = var_873_to_fp16)[name = string("op_874_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_877_cast_fp16 = tanh(x = var_874_cast_fp16)[name = string("op_877_cast_fp16")];
            fp16 var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_879_cast_fp16 = add(x = var_877_cast_fp16, y = var_878_to_fp16)[name = string("op_879_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_880_cast_fp16 = mul(x = var_874_cast_fp16, y = var_879_cast_fp16)[name = string("op_880_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_35_cast_fp16 = mul(x = var_880_cast_fp16, y = linear_8_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_1_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29798336)))];
            tensor<fp16, [1, 12, 768]> linear_9_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w2_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_9_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_19_cast_fp16 = add(x = x_15_cast_fp16, y = linear_9_cast_fp16)[name = string("x_19_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_886_cast_fp16 = abs(x = x_19_cast_fp16)[name = string("op_886_cast_fp16")];
            tensor<int32, [1]> var_888_axes_0 = const()[name = string("op_888_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_888_cast_fp16 = reduce_max(axes = var_888_axes_0, keep_dims = var_128, x = var_886_cast_fp16)[name = string("op_888_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_9_cast_fp16 = maximum(x = var_888_cast_fp16, y = var_129_to_fp16)[name = string("s_9_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_13_cast_fp16 = real_div(x = x_19_cast_fp16, y = s_9_cast_fp16)[name = string("y_13_cast_fp16")];
            fp16 var_130_promoted_4_to_fp16 = const()[name = string("op_130_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_891_cast_fp16 = pow(x = y_13_cast_fp16, y = var_130_promoted_4_to_fp16)[name = string("op_891_cast_fp16")];
            tensor<int32, [1]> var_893_axes_0 = const()[name = string("op_893_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_893_keep_dims_0 = const()[name = string("op_893_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_893_cast_fp16 = reduce_mean(axes = var_893_axes_0, keep_dims = var_893_keep_dims_0, x = var_891_cast_fp16)[name = string("op_893_cast_fp16")];
            fp32 var_894_epsilon_0 = const()[name = string("op_894_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_894_cast_fp16 = inverse(epsilon = var_894_epsilon_0, x = s_9_cast_fp16)[name = string("op_894_cast_fp16")];
            fp16 var_895_to_fp16 = const()[name = string("op_895_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_896_cast_fp16 = mul(x = var_894_cast_fp16, y = var_895_to_fp16)[name = string("op_896_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_897_cast_fp16 = real_div(x = var_896_cast_fp16, y = s_9_cast_fp16)[name = string("op_897_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_898_cast_fp16 = add(x = var_893_cast_fp16, y = var_897_cast_fp16)[name = string("op_898_cast_fp16")];
            fp32 var_899_epsilon_0 = const()[name = string("op_899_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_899_cast_fp16 = rsqrt(epsilon = var_899_epsilon_0, x = var_898_cast_fp16)[name = string("op_899_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_900_cast_fp16 = mul(x = y_13_cast_fp16, y = var_899_cast_fp16)[name = string("op_900_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_2_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32550912)))];
            tensor<fp16, [1, 12, 768]> input_37_cast_fp16 = mul(x = core_backbone_layers_2_operator_norm_weight_to_fp16, y = var_900_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32552512)))];
            tensor<fp16, [1, 12, 768]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_q_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<int32, [4]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [4]>([1, 12, 8, 96])];
            tensor<fp16, [1, 12, 8, 96]> x_21_cast_fp16 = reshape(shape = concat_1, x = linear_10_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_907_cast_fp16 = abs(x = x_21_cast_fp16)[name = string("op_907_cast_fp16")];
            tensor<int32, [1]> var_909_axes_0 = const()[name = string("op_909_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 8, 1]> var_909_cast_fp16 = reduce_max(axes = var_909_axes_0, keep_dims = var_128, x = var_907_cast_fp16)[name = string("op_909_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> s_11_cast_fp16 = maximum(x = var_909_cast_fp16, y = var_129_to_fp16)[name = string("s_11_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> y_15_cast_fp16 = real_div(x = x_21_cast_fp16, y = s_11_cast_fp16)[name = string("y_15_cast_fp16")];
            fp16 var_130_promoted_5_to_fp16 = const()[name = string("op_130_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 8, 96]> var_912_cast_fp16 = pow(x = y_15_cast_fp16, y = var_130_promoted_5_to_fp16)[name = string("op_912_cast_fp16")];
            tensor<int32, [1]> var_914_axes_0 = const()[name = string("op_914_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_914_keep_dims_0 = const()[name = string("op_914_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 8, 1]> var_914_cast_fp16 = reduce_mean(axes = var_914_axes_0, keep_dims = var_914_keep_dims_0, x = var_912_cast_fp16)[name = string("op_914_cast_fp16")];
            fp32 var_915_epsilon_0 = const()[name = string("op_915_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 8, 1]> var_915_cast_fp16 = inverse(epsilon = var_915_epsilon_0, x = s_11_cast_fp16)[name = string("op_915_cast_fp16")];
            fp16 var_916_to_fp16 = const()[name = string("op_916_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 8, 1]> var_917_cast_fp16 = mul(x = var_915_cast_fp16, y = var_916_to_fp16)[name = string("op_917_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_918_cast_fp16 = real_div(x = var_917_cast_fp16, y = s_11_cast_fp16)[name = string("op_918_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_919_cast_fp16 = add(x = var_914_cast_fp16, y = var_918_cast_fp16)[name = string("op_919_cast_fp16")];
            fp32 var_920_epsilon_0 = const()[name = string("op_920_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 8, 1]> var_920_cast_fp16 = rsqrt(epsilon = var_920_epsilon_0, x = var_919_cast_fp16)[name = string("op_920_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_921_cast_fp16 = mul(x = y_15_cast_fp16, y = var_920_cast_fp16)[name = string("op_921_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_2_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33732224)))];
            tensor<fp16, [1, 12, 8, 96]> var_922_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_q_norm_weight_to_fp16, y = var_921_cast_fp16)[name = string("op_922_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33732480)))];
            tensor<fp16, [384]> linear_11_bias_0_to_fp16 = const()[name = string("linear_11_bias_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34322368)))];
            tensor<fp16, [1, 12, 384]> linear_11_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_k_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<int32, [4]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> x_23_cast_fp16 = reshape(shape = concat_2, x = linear_11_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_929_cast_fp16 = abs(x = x_23_cast_fp16)[name = string("op_929_cast_fp16")];
            tensor<int32, [1]> var_931_axes_0 = const()[name = string("op_931_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 4, 1]> var_931_cast_fp16 = reduce_max(axes = var_931_axes_0, keep_dims = var_128, x = var_929_cast_fp16)[name = string("op_931_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> s_13_cast_fp16 = maximum(x = var_931_cast_fp16, y = var_129_to_fp16)[name = string("s_13_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> y_17_cast_fp16 = real_div(x = x_23_cast_fp16, y = s_13_cast_fp16)[name = string("y_17_cast_fp16")];
            fp16 var_130_promoted_6_to_fp16 = const()[name = string("op_130_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 4, 96]> var_934_cast_fp16 = pow(x = y_17_cast_fp16, y = var_130_promoted_6_to_fp16)[name = string("op_934_cast_fp16")];
            tensor<int32, [1]> var_936_axes_0 = const()[name = string("op_936_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_936_keep_dims_0 = const()[name = string("op_936_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 4, 1]> var_936_cast_fp16 = reduce_mean(axes = var_936_axes_0, keep_dims = var_936_keep_dims_0, x = var_934_cast_fp16)[name = string("op_936_cast_fp16")];
            fp32 var_937_epsilon_0 = const()[name = string("op_937_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 4, 1]> var_937_cast_fp16 = inverse(epsilon = var_937_epsilon_0, x = s_13_cast_fp16)[name = string("op_937_cast_fp16")];
            fp16 var_938_to_fp16 = const()[name = string("op_938_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 4, 1]> var_939_cast_fp16 = mul(x = var_937_cast_fp16, y = var_938_to_fp16)[name = string("op_939_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_940_cast_fp16 = real_div(x = var_939_cast_fp16, y = s_13_cast_fp16)[name = string("op_940_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_941_cast_fp16 = add(x = var_936_cast_fp16, y = var_940_cast_fp16)[name = string("op_941_cast_fp16")];
            fp32 var_942_epsilon_0 = const()[name = string("op_942_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 4, 1]> var_942_cast_fp16 = rsqrt(epsilon = var_942_epsilon_0, x = var_941_cast_fp16)[name = string("op_942_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_943_cast_fp16 = mul(x = y_17_cast_fp16, y = var_942_cast_fp16)[name = string("op_943_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_2_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34323200)))];
            tensor<fp16, [1, 12, 4, 96]> var_944_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_k_norm_weight_to_fp16, y = var_943_cast_fp16)[name = string("op_944_cast_fp16")];
            tensor<int32, [4]> new_5_perm_0 = const()[name = string("new_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34323456)))];
            tensor<fp16, [1, 12, 384]> linear_12_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_v_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> var_949_cast_fp16 = reshape(shape = concat_3, x = linear_12_cast_fp16)[name = string("op_949_cast_fp16")];
            tensor<int32, [4]> new_7_perm_0 = const()[name = string("new_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_41_begin_0 = const()[name = string("input_41_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_41_end_0 = const()[name = string("input_41_end_0"), val = tensor<int32, [5]>([1, 1, 4, 31, 96])];
            tensor<bool, [5]> input_41_end_mask_0 = const()[name = string("input_41_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_41_squeeze_mask_0 = const()[name = string("input_41_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_41_cast_fp16 = slice_by_index(begin = input_41_begin_0, end = input_41_end_0, end_mask = input_41_end_mask_0, squeeze_mask = input_41_squeeze_mask_0, x = k)[name = string("input_41_cast_fp16")];
            tensor<int32, [8]> var_954_pad_0 = const()[name = string("op_954_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_954_mode_0 = const()[name = string("op_954_mode_0"), val = string("constant")];
            fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_5_cast_fp16 = transpose(perm = new_5_perm_0, x = var_944_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [1, 4, 43, 96]> var_954_cast_fp16 = pad(constant_val = const_8_to_fp16, mode = var_954_mode_0, pad = var_954_pad_0, x = new_5_cast_fp16)[name = string("op_954_cast_fp16")];
            tensor<int32, [8]> var_956_pad_0 = const()[name = string("op_956_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_956_mode_0 = const()[name = string("op_956_mode_0"), val = string("constant")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_956_cast_fp16 = pad(constant_val = const_9_to_fp16, mode = var_956_mode_0, pad = var_956_pad_0, x = input_41_cast_fp16)[name = string("op_956_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> x_27_cast_fp16 = add(x = var_954_cast_fp16, y = var_956_cast_fp16)[name = string("x_27_cast_fp16")];
            tensor<int32, [5]> input_45_begin_0 = const()[name = string("input_45_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_45_end_0 = const()[name = string("input_45_end_0"), val = tensor<int32, [5]>([1, 1, 4, 31, 96])];
            tensor<bool, [5]> input_45_end_mask_0 = const()[name = string("input_45_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_45_squeeze_mask_0 = const()[name = string("input_45_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_45_cast_fp16 = slice_by_index(begin = input_45_begin_0, end = input_45_end_0, end_mask = input_45_end_mask_0, squeeze_mask = input_45_squeeze_mask_0, x = v)[name = string("input_45_cast_fp16")];
            tensor<int32, [8]> var_961_pad_0 = const()[name = string("op_961_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_961_mode_0 = const()[name = string("op_961_mode_0"), val = string("constant")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_7_cast_fp16 = transpose(perm = new_7_perm_0, x = var_949_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [1, 4, 43, 96]> var_961_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = var_961_mode_0, pad = var_961_pad_0, x = new_7_cast_fp16)[name = string("op_961_cast_fp16")];
            tensor<int32, [8]> var_963_pad_0 = const()[name = string("op_963_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_963_mode_0 = const()[name = string("op_963_mode_0"), val = string("constant")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_963_cast_fp16 = pad(constant_val = const_11_to_fp16, mode = var_963_mode_0, pad = var_963_pad_0, x = input_45_cast_fp16)[name = string("op_963_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> v_3_cast_fp16 = add(x = var_961_cast_fp16, y = var_963_cast_fp16)[name = string("v_3_cast_fp16")];
            tensor<fp16, [1, 12, 1, 96]> const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = tensor<fp16, [1, 12, 1, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34913344)))];
            tensor<fp16, [1, 12, 8, 96]> var_965_cast_fp16 = mul(x = var_922_cast_fp16, y = const_66_to_fp16)[name = string("op_965_cast_fp16")];
            tensor<int32, [2]> var_966_split_sizes_0 = const()[name = string("op_966_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_67 = const()[name = string("const_67"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 48]> var_966_cast_fp16_0, tensor<fp16, [1, 12, 8, 48]> var_966_cast_fp16_1 = split(axis = const_67, split_sizes = var_966_split_sizes_0, x = var_922_cast_fp16)[name = string("op_966_cast_fp16")];
            fp16 const_12_promoted_to_fp16 = const()[name = string("const_12_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 12, 8, 48]> var_968_cast_fp16 = mul(x = var_966_cast_fp16_1, y = const_12_promoted_to_fp16)[name = string("op_968_cast_fp16")];
            bool var_970_interleave_0 = const()[name = string("op_970_interleave_0"), val = bool(false)];
            int32 const_68 = const()[name = string("const_68"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 96]> var_970_cast_fp16 = concat(axis = const_68, interleave = var_970_interleave_0, values = (var_968_cast_fp16, var_966_cast_fp16_0))[name = string("op_970_cast_fp16")];
            tensor<fp16, [1, 12, 1, 96]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [1, 12, 1, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34915712)))];
            tensor<fp16, [1, 12, 8, 96]> var_971_cast_fp16 = mul(x = var_970_cast_fp16, y = const_69_to_fp16)[name = string("op_971_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> q_1_cast_fp16 = add(x = var_965_cast_fp16, y = var_971_cast_fp16)[name = string("q_1_cast_fp16")];
            tensor<fp16, [1, 1, 43, 96]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [1, 1, 43, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34918080)))];
            tensor<fp16, [1, 4, 43, 96]> var_973_cast_fp16 = mul(x = x_27_cast_fp16, y = const_0_to_fp16)[name = string("op_973_cast_fp16")];
            tensor<int32, [2]> var_974_split_sizes_0 = const()[name = string("op_974_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_974_axis_0 = const()[name = string("op_974_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 43, 48]> var_974_cast_fp16_0, tensor<fp16, [1, 4, 43, 48]> var_974_cast_fp16_1 = split(axis = var_974_axis_0, split_sizes = var_974_split_sizes_0, x = x_27_cast_fp16)[name = string("op_974_cast_fp16")];
            fp16 const_13_promoted_to_fp16 = const()[name = string("const_13_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 43, 48]> var_976_cast_fp16 = mul(x = var_974_cast_fp16_1, y = const_13_promoted_to_fp16)[name = string("op_976_cast_fp16")];
            bool var_978_interleave_0 = const()[name = string("op_978_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 43, 96]> var_978_cast_fp16 = concat(axis = var_127, interleave = var_978_interleave_0, values = (var_976_cast_fp16, var_974_cast_fp16_0))[name = string("op_978_cast_fp16")];
            tensor<fp16, [1, 1, 43, 96]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [1, 1, 43, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34926400)))];
            tensor<fp16, [1, 4, 43, 96]> var_979_cast_fp16 = mul(x = var_978_cast_fp16, y = const_1_to_fp16)[name = string("op_979_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> k_3_cast_fp16 = add(x = var_973_cast_fp16, y = var_979_cast_fp16)[name = string("k_3_cast_fp16")];
            tensor<int32, [1]> var_981_axes_0 = const()[name = string("op_981_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_981_cast_fp16 = expand_dims(axes = var_981_axes_0, x = k_3_cast_fp16)[name = string("op_981_cast_fp16")];
            tensor<int32, [5]> var_983_reps_0 = const()[name = string("op_983_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_983_cast_fp16 = tile(reps = var_983_reps_0, x = var_981_cast_fp16)[name = string("op_983_cast_fp16")];
            tensor<int32, [4]> concat_5 = const()[name = string("concat_5"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> k_5_cast_fp16 = reshape(shape = concat_5, x = var_983_cast_fp16)[name = string("k_5_cast_fp16")];
            tensor<int32, [1]> var_985_axes_0 = const()[name = string("op_985_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_985_cast_fp16 = expand_dims(axes = var_985_axes_0, x = v_3_cast_fp16)[name = string("op_985_cast_fp16")];
            tensor<int32, [5]> var_987_reps_0 = const()[name = string("op_987_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_987_cast_fp16 = tile(reps = var_987_reps_0, x = var_985_cast_fp16)[name = string("op_987_cast_fp16")];
            tensor<int32, [4]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> v_5_cast_fp16 = reshape(shape = concat_7, x = var_987_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> transpose_0_perm_0 = const()[name = string("transpose_0_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_990_transpose_x_1 = const()[name = string("op_990_transpose_x_1"), val = bool(false)];
            bool var_990_transpose_y_1 = const()[name = string("op_990_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 96]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = q_1_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [1, 8, 12, 43]> var_990_cast_fp16 = matmul(transpose_x = var_990_transpose_x_1, transpose_y = var_990_transpose_y_1, x = transpose_0_cast_fp16, y = k_5_cast_fp16)[name = string("op_990_cast_fp16")];
            fp16 var_991_to_fp16 = const()[name = string("op_991_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 12, 43]> s_15_cast_fp16 = mul(x = var_990_cast_fp16, y = var_991_to_fp16)[name = string("s_15_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_29_cast_fp16 = add(x = s_15_cast_fp16, y = mask)[name = string("x_29_cast_fp16")];
            tensor<int32, [1]> var_995_axes_0 = const()[name = string("op_995_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 12, 1]> var_995_cast_fp16 = reduce_max(axes = var_995_axes_0, keep_dims = var_128, x = x_29_cast_fp16)[name = string("op_995_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_31_cast_fp16 = sub(x = x_29_cast_fp16, y = var_995_cast_fp16)[name = string("x_31_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_33_cast_fp16 = exp(x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<int32, [1]> var_999_axes_0 = const()[name = string("op_999_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_999_keep_dims_0 = const()[name = string("op_999_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 1]> var_999_cast_fp16 = reduce_sum(axes = var_999_axes_0, keep_dims = var_999_keep_dims_0, x = x_33_cast_fp16)[name = string("op_999_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> var_1000_cast_fp16 = real_div(x = x_33_cast_fp16, y = var_999_cast_fp16)[name = string("op_1000_cast_fp16")];
            bool o_1_transpose_x_0 = const()[name = string("o_1_transpose_x_0"), val = bool(false)];
            bool o_1_transpose_y_0 = const()[name = string("o_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 12, 96]> o_1_cast_fp16 = matmul(transpose_x = o_1_transpose_x_0, transpose_y = o_1_transpose_y_0, x = var_1000_cast_fp16, y = v_5_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<int32, [4]> o_3_perm_0 = const()[name = string("o_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [3]>([1, 12, 768])];
            tensor<fp16, [1, 12, 8, 96]> o_3_cast_fp16 = transpose(perm = o_3_perm_0, x = o_1_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [1, 12, 768]> input_47_cast_fp16 = reshape(shape = concat_8, x = o_3_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34934720)))];
            tensor<fp16, [1, 12, 768]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_o_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> var_1009_begin_0 = const()[name = string("op_1009_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_1009_end_0 = const()[name = string("op_1009_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_1009_end_mask_0 = const()[name = string("op_1009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1009_cast_fp16 = slice_by_index(begin = var_1009_begin_0, end = var_1009_end_0, end_mask = var_1009_end_mask_0, x = x_27_cast_fp16)[name = string("op_1009_cast_fp16")];
            tensor<int32, [4]> var_1012_begin_0 = const()[name = string("op_1012_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_1012_end_0 = const()[name = string("op_1012_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_1012_end_mask_0 = const()[name = string("op_1012_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = v_3_cast_fp16)[name = string("op_1012_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_35_cast_fp16 = add(x = x_19_cast_fp16, y = linear_13_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1015_cast_fp16 = abs(x = x_35_cast_fp16)[name = string("op_1015_cast_fp16")];
            tensor<int32, [1]> var_1017_axes_0 = const()[name = string("op_1017_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1017_cast_fp16 = reduce_max(axes = var_1017_axes_0, keep_dims = var_128, x = var_1015_cast_fp16)[name = string("op_1017_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_17_cast_fp16 = maximum(x = var_1017_cast_fp16, y = var_129_to_fp16)[name = string("s_17_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_19_cast_fp16 = real_div(x = x_35_cast_fp16, y = s_17_cast_fp16)[name = string("y_19_cast_fp16")];
            fp16 var_130_promoted_7_to_fp16 = const()[name = string("op_130_promoted_7_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1020_cast_fp16 = pow(x = y_19_cast_fp16, y = var_130_promoted_7_to_fp16)[name = string("op_1020_cast_fp16")];
            tensor<int32, [1]> var_1022_axes_0 = const()[name = string("op_1022_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1022_keep_dims_0 = const()[name = string("op_1022_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1022_cast_fp16 = reduce_mean(axes = var_1022_axes_0, keep_dims = var_1022_keep_dims_0, x = var_1020_cast_fp16)[name = string("op_1022_cast_fp16")];
            fp32 var_1023_epsilon_0 = const()[name = string("op_1023_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1023_cast_fp16 = inverse(epsilon = var_1023_epsilon_0, x = s_17_cast_fp16)[name = string("op_1023_cast_fp16")];
            fp16 var_1024_to_fp16 = const()[name = string("op_1024_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1025_cast_fp16 = mul(x = var_1023_cast_fp16, y = var_1024_to_fp16)[name = string("op_1025_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1026_cast_fp16 = real_div(x = var_1025_cast_fp16, y = s_17_cast_fp16)[name = string("op_1026_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1027_cast_fp16 = add(x = var_1022_cast_fp16, y = var_1026_cast_fp16)[name = string("op_1027_cast_fp16")];
            fp32 var_1028_epsilon_0 = const()[name = string("op_1028_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1028_cast_fp16 = rsqrt(epsilon = var_1028_epsilon_0, x = var_1027_cast_fp16)[name = string("op_1028_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1029_cast_fp16 = mul(x = y_19_cast_fp16, y = var_1028_cast_fp16)[name = string("op_1029_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_2_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36114432)))];
            tensor<fp16, [1, 12, 768]> input_49_cast_fp16 = mul(x = core_backbone_layers_2_ffn_norm_weight_to_fp16, y = var_1029_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_2_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36116032)))];
            tensor<fp16, [1, 12, 1792]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w1_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_2_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38868608)))];
            tensor<fp16, [1, 12, 1792]> linear_15_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w3_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_15_cast_fp16")];
            fp16 var_1038_to_fp16 = const()[name = string("op_1038_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1039_cast_fp16 = mul(x = linear_14_cast_fp16, y = var_1038_to_fp16)[name = string("op_1039_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1042_cast_fp16 = tanh(x = var_1039_cast_fp16)[name = string("op_1042_cast_fp16")];
            fp16 var_1043_to_fp16 = const()[name = string("op_1043_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1044_cast_fp16 = add(x = var_1042_cast_fp16, y = var_1043_to_fp16)[name = string("op_1044_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1045_cast_fp16 = mul(x = var_1039_cast_fp16, y = var_1044_cast_fp16)[name = string("op_1045_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_51_cast_fp16 = mul(x = var_1045_cast_fp16, y = linear_15_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_2_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41621184)))];
            tensor<fp16, [1, 12, 768]> linear_16_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w2_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_39_cast_fp16 = add(x = x_35_cast_fp16, y = linear_16_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1051_cast_fp16 = abs(x = x_39_cast_fp16)[name = string("op_1051_cast_fp16")];
            tensor<int32, [1]> var_1053_axes_0 = const()[name = string("op_1053_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1053_cast_fp16 = reduce_max(axes = var_1053_axes_0, keep_dims = var_128, x = var_1051_cast_fp16)[name = string("op_1053_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_19_cast_fp16 = maximum(x = var_1053_cast_fp16, y = var_129_to_fp16)[name = string("s_19_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_21_cast_fp16 = real_div(x = x_39_cast_fp16, y = s_19_cast_fp16)[name = string("y_21_cast_fp16")];
            fp16 var_130_promoted_8_to_fp16 = const()[name = string("op_130_promoted_8_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1056_cast_fp16 = pow(x = y_21_cast_fp16, y = var_130_promoted_8_to_fp16)[name = string("op_1056_cast_fp16")];
            tensor<int32, [1]> var_1058_axes_0 = const()[name = string("op_1058_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1058_keep_dims_0 = const()[name = string("op_1058_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1058_cast_fp16 = reduce_mean(axes = var_1058_axes_0, keep_dims = var_1058_keep_dims_0, x = var_1056_cast_fp16)[name = string("op_1058_cast_fp16")];
            fp32 var_1059_epsilon_0 = const()[name = string("op_1059_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1059_cast_fp16 = inverse(epsilon = var_1059_epsilon_0, x = s_19_cast_fp16)[name = string("op_1059_cast_fp16")];
            fp16 var_1060_to_fp16 = const()[name = string("op_1060_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1061_cast_fp16 = mul(x = var_1059_cast_fp16, y = var_1060_to_fp16)[name = string("op_1061_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1062_cast_fp16 = real_div(x = var_1061_cast_fp16, y = s_19_cast_fp16)[name = string("op_1062_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1063_cast_fp16 = add(x = var_1058_cast_fp16, y = var_1062_cast_fp16)[name = string("op_1063_cast_fp16")];
            fp32 var_1064_epsilon_0 = const()[name = string("op_1064_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1064_cast_fp16 = rsqrt(epsilon = var_1064_epsilon_0, x = var_1063_cast_fp16)[name = string("op_1064_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1065_cast_fp16 = mul(x = y_21_cast_fp16, y = var_1064_cast_fp16)[name = string("op_1065_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_3_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44373760)))];
            tensor<fp16, [1, 12, 768]> input_53_cast_fp16 = mul(x = core_backbone_layers_3_operator_norm_weight_to_fp16, y = var_1065_cast_fp16)[name = string("input_53_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_3_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44375360)))];
            tensor<fp16, [1, 12, 2304]> linear_17_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_3_conv_in_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<int32, [3]> bcx_5_perm_0 = const()[name = string("bcx_5_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_11_begin_0 = const()[name = string("b_11_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_11_end_0 = const()[name = string("b_11_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_11_end_mask_0 = const()[name = string("b_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_5_cast_fp16 = transpose(perm = bcx_5_perm_0, x = linear_17_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [1, 768, 12]> b_11_cast_fp16 = slice_by_index(begin = b_11_begin_0, end = b_11_end_0, end_mask = b_11_end_mask_0, x = bcx_5_cast_fp16)[name = string("b_11_cast_fp16")];
            tensor<int32, [3]> c_5_begin_0 = const()[name = string("c_5_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_5_end_0 = const()[name = string("c_5_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_5_end_mask_0 = const()[name = string("c_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_5_cast_fp16 = slice_by_index(begin = c_5_begin_0, end = c_5_end_0, end_mask = c_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("c_5_cast_fp16")];
            tensor<int32, [3]> xx_5_begin_0 = const()[name = string("xx_5_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_5_end_0 = const()[name = string("xx_5_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_5_end_mask_0 = const()[name = string("xx_5_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_5_cast_fp16 = slice_by_index(begin = xx_5_begin_0, end = xx_5_end_0, end_mask = xx_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("xx_5_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_9_cast_fp16 = mul(x = b_11_cast_fp16, y = xx_5_cast_fp16)[name = string("new_9_cast_fp16")];
            tensor<int32, [4]> input_57_begin_0 = const()[name = string("input_57_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])];
            tensor<int32, [4]> input_57_end_0 = const()[name = string("input_57_end_0"), val = tensor<int32, [4]>([3, 1, 768, 2])];
            tensor<bool, [4]> input_57_end_mask_0 = const()[name = string("input_57_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_57_squeeze_mask_0 = const()[name = string("input_57_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_57_cast_fp16 = slice_by_index(begin = input_57_begin_0, end = input_57_end_0, end_mask = input_57_end_mask_0, squeeze_mask = input_57_squeeze_mask_0, x = conv)[name = string("input_57_cast_fp16")];
            tensor<int32, [6]> var_1080_pad_0 = const()[name = string("op_1080_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1080_mode_0 = const()[name = string("op_1080_mode_0"), val = string("constant")];
            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1080_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = var_1080_mode_0, pad = var_1080_pad_0, x = new_9_cast_fp16)[name = string("op_1080_cast_fp16")];
            tensor<int32, [6]> var_1082_pad_0 = const()[name = string("op_1082_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_1082_mode_0 = const()[name = string("op_1082_mode_0"), val = string("constant")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1082_cast_fp16 = pad(constant_val = const_15_to_fp16, mode = var_1082_mode_0, pad = var_1082_pad_0, x = input_57_cast_fp16)[name = string("op_1082_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_9_cast_fp16 = add(x = var_1080_cast_fp16, y = var_1082_cast_fp16)[name = string("win_9_cast_fp16")];
            tensor<int32, [3]> var_1084_begin_0 = const()[name = string("op_1084_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_1084_end_0 = const()[name = string("op_1084_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_1084_end_mask_0 = const()[name = string("op_1084_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1084_cast_fp16 = slice_by_index(begin = var_1084_begin_0, end = var_1084_end_0, end_mask = var_1084_end_mask_0, x = win_9_cast_fp16)[name = string("op_1084_cast_fp16")];
            string var_1090_pad_type_0 = const()[name = string("op_1090_pad_type_0"), val = string("valid")];
            int32 var_1090_groups_0 = const()[name = string("op_1090_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1090_strides_0 = const()[name = string("op_1090_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1090_pad_0 = const()[name = string("op_1090_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1090_dilations_0 = const()[name = string("op_1090_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_3_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47914368)))];
            tensor<fp16, [1, 768, 12]> var_1090_cast_fp16 = conv(dilations = var_1090_dilations_0, groups = var_1090_groups_0, pad = var_1090_pad_0, pad_type = var_1090_pad_type_0, strides = var_1090_strides_0, weight = core_backbone_layers_3_conv_conv_weight_to_fp16, x = win_9_cast_fp16)[name = string("op_1090_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_1092_cast_fp16 = mul(x = c_5_cast_fp16, y = var_1090_cast_fp16)[name = string("op_1092_cast_fp16")];
            tensor<int32, [3]> input_59_perm_0 = const()[name = string("input_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_3_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47919040)))];
            tensor<fp16, [1, 12, 768]> input_59_cast_fp16 = transpose(perm = input_59_perm_0, x = var_1092_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [1, 12, 768]> linear_18_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_conv_out_proj_weight_to_fp16, x = input_59_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_41_cast_fp16 = add(x = x_39_cast_fp16, y = linear_18_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1098_cast_fp16 = abs(x = x_41_cast_fp16)[name = string("op_1098_cast_fp16")];
            tensor<int32, [1]> var_1100_axes_0 = const()[name = string("op_1100_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1100_cast_fp16 = reduce_max(axes = var_1100_axes_0, keep_dims = var_128, x = var_1098_cast_fp16)[name = string("op_1100_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_21_cast_fp16 = maximum(x = var_1100_cast_fp16, y = var_129_to_fp16)[name = string("s_21_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_25_cast_fp16 = real_div(x = x_41_cast_fp16, y = s_21_cast_fp16)[name = string("y_25_cast_fp16")];
            fp16 var_130_promoted_9_to_fp16 = const()[name = string("op_130_promoted_9_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1103_cast_fp16 = pow(x = y_25_cast_fp16, y = var_130_promoted_9_to_fp16)[name = string("op_1103_cast_fp16")];
            tensor<int32, [1]> var_1105_axes_0 = const()[name = string("op_1105_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1105_keep_dims_0 = const()[name = string("op_1105_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1105_cast_fp16 = reduce_mean(axes = var_1105_axes_0, keep_dims = var_1105_keep_dims_0, x = var_1103_cast_fp16)[name = string("op_1105_cast_fp16")];
            fp32 var_1106_epsilon_0 = const()[name = string("op_1106_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1106_cast_fp16 = inverse(epsilon = var_1106_epsilon_0, x = s_21_cast_fp16)[name = string("op_1106_cast_fp16")];
            fp16 var_1107_to_fp16 = const()[name = string("op_1107_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1108_cast_fp16 = mul(x = var_1106_cast_fp16, y = var_1107_to_fp16)[name = string("op_1108_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1109_cast_fp16 = real_div(x = var_1108_cast_fp16, y = s_21_cast_fp16)[name = string("op_1109_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1110_cast_fp16 = add(x = var_1105_cast_fp16, y = var_1109_cast_fp16)[name = string("op_1110_cast_fp16")];
            fp32 var_1111_epsilon_0 = const()[name = string("op_1111_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1111_cast_fp16 = rsqrt(epsilon = var_1111_epsilon_0, x = var_1110_cast_fp16)[name = string("op_1111_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1112_cast_fp16 = mul(x = y_25_cast_fp16, y = var_1111_cast_fp16)[name = string("op_1112_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_3_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49098752)))];
            tensor<fp16, [1, 12, 768]> input_61_cast_fp16 = mul(x = core_backbone_layers_3_ffn_norm_weight_to_fp16, y = var_1112_cast_fp16)[name = string("input_61_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_3_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49100352)))];
            tensor<fp16, [1, 12, 1792]> linear_19_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w1_weight_to_fp16, x = input_61_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_3_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51852928)))];
            tensor<fp16, [1, 12, 1792]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w3_weight_to_fp16, x = input_61_cast_fp16)[name = string("linear_20_cast_fp16")];
            fp16 var_1121_to_fp16 = const()[name = string("op_1121_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1122_cast_fp16 = mul(x = linear_19_cast_fp16, y = var_1121_to_fp16)[name = string("op_1122_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1125_cast_fp16 = tanh(x = var_1122_cast_fp16)[name = string("op_1125_cast_fp16")];
            fp16 var_1126_to_fp16 = const()[name = string("op_1126_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1127_cast_fp16 = add(x = var_1125_cast_fp16, y = var_1126_to_fp16)[name = string("op_1127_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1128_cast_fp16 = mul(x = var_1122_cast_fp16, y = var_1127_cast_fp16)[name = string("op_1128_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_63_cast_fp16 = mul(x = var_1128_cast_fp16, y = linear_20_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_3_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54605504)))];
            tensor<fp16, [1, 12, 768]> linear_21_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w2_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_45_cast_fp16 = add(x = x_41_cast_fp16, y = linear_21_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1134_cast_fp16 = abs(x = x_45_cast_fp16)[name = string("op_1134_cast_fp16")];
            tensor<int32, [1]> var_1136_axes_0 = const()[name = string("op_1136_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1136_cast_fp16 = reduce_max(axes = var_1136_axes_0, keep_dims = var_128, x = var_1134_cast_fp16)[name = string("op_1136_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_23_cast_fp16 = maximum(x = var_1136_cast_fp16, y = var_129_to_fp16)[name = string("s_23_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_27_cast_fp16 = real_div(x = x_45_cast_fp16, y = s_23_cast_fp16)[name = string("y_27_cast_fp16")];
            fp16 var_130_promoted_10_to_fp16 = const()[name = string("op_130_promoted_10_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1139_cast_fp16 = pow(x = y_27_cast_fp16, y = var_130_promoted_10_to_fp16)[name = string("op_1139_cast_fp16")];
            tensor<int32, [1]> var_1141_axes_0 = const()[name = string("op_1141_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1141_keep_dims_0 = const()[name = string("op_1141_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1141_cast_fp16 = reduce_mean(axes = var_1141_axes_0, keep_dims = var_1141_keep_dims_0, x = var_1139_cast_fp16)[name = string("op_1141_cast_fp16")];
            fp32 var_1142_epsilon_0 = const()[name = string("op_1142_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1142_cast_fp16 = inverse(epsilon = var_1142_epsilon_0, x = s_23_cast_fp16)[name = string("op_1142_cast_fp16")];
            fp16 var_1143_to_fp16 = const()[name = string("op_1143_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1144_cast_fp16 = mul(x = var_1142_cast_fp16, y = var_1143_to_fp16)[name = string("op_1144_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1145_cast_fp16 = real_div(x = var_1144_cast_fp16, y = s_23_cast_fp16)[name = string("op_1145_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1146_cast_fp16 = add(x = var_1141_cast_fp16, y = var_1145_cast_fp16)[name = string("op_1146_cast_fp16")];
            fp32 var_1147_epsilon_0 = const()[name = string("op_1147_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1147_cast_fp16 = rsqrt(epsilon = var_1147_epsilon_0, x = var_1146_cast_fp16)[name = string("op_1147_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1148_cast_fp16 = mul(x = y_27_cast_fp16, y = var_1147_cast_fp16)[name = string("op_1148_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_4_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57358080)))];
            tensor<fp16, [1, 12, 768]> input_65_cast_fp16 = mul(x = core_backbone_layers_4_operator_norm_weight_to_fp16, y = var_1148_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_4_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57359680)))];
            tensor<fp16, [1, 12, 2304]> linear_22_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_4_conv_in_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [3]> bcx_7_perm_0 = const()[name = string("bcx_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_15_begin_0 = const()[name = string("b_15_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_15_end_0 = const()[name = string("b_15_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_15_end_mask_0 = const()[name = string("b_15_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_7_cast_fp16 = transpose(perm = bcx_7_perm_0, x = linear_22_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [1, 768, 12]> b_15_cast_fp16 = slice_by_index(begin = b_15_begin_0, end = b_15_end_0, end_mask = b_15_end_mask_0, x = bcx_7_cast_fp16)[name = string("b_15_cast_fp16")];
            tensor<int32, [3]> c_7_begin_0 = const()[name = string("c_7_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_7_end_0 = const()[name = string("c_7_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_7_end_mask_0 = const()[name = string("c_7_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_7_cast_fp16 = slice_by_index(begin = c_7_begin_0, end = c_7_end_0, end_mask = c_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("c_7_cast_fp16")];
            tensor<int32, [3]> xx_7_begin_0 = const()[name = string("xx_7_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_7_end_0 = const()[name = string("xx_7_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_7_end_mask_0 = const()[name = string("xx_7_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_7_cast_fp16 = slice_by_index(begin = xx_7_begin_0, end = xx_7_end_0, end_mask = xx_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("xx_7_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_11_cast_fp16 = mul(x = b_15_cast_fp16, y = xx_7_cast_fp16)[name = string("new_11_cast_fp16")];
            tensor<int32, [4]> input_69_begin_0 = const()[name = string("input_69_begin_0"), val = tensor<int32, [4]>([3, 0, 0, 0])];
            tensor<int32, [4]> input_69_end_0 = const()[name = string("input_69_end_0"), val = tensor<int32, [4]>([4, 1, 768, 2])];
            tensor<bool, [4]> input_69_end_mask_0 = const()[name = string("input_69_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_69_squeeze_mask_0 = const()[name = string("input_69_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_69_cast_fp16 = slice_by_index(begin = input_69_begin_0, end = input_69_end_0, end_mask = input_69_end_mask_0, squeeze_mask = input_69_squeeze_mask_0, x = conv)[name = string("input_69_cast_fp16")];
            tensor<int32, [6]> var_1163_pad_0 = const()[name = string("op_1163_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1163_mode_0 = const()[name = string("op_1163_mode_0"), val = string("constant")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1163_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = var_1163_mode_0, pad = var_1163_pad_0, x = new_11_cast_fp16)[name = string("op_1163_cast_fp16")];
            tensor<int32, [6]> var_1165_pad_0 = const()[name = string("op_1165_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_1165_mode_0 = const()[name = string("op_1165_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1165_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = var_1165_mode_0, pad = var_1165_pad_0, x = input_69_cast_fp16)[name = string("op_1165_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_13_cast_fp16 = add(x = var_1163_cast_fp16, y = var_1165_cast_fp16)[name = string("win_13_cast_fp16")];
            tensor<int32, [3]> var_1167_begin_0 = const()[name = string("op_1167_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_1167_end_0 = const()[name = string("op_1167_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_1167_end_mask_0 = const()[name = string("op_1167_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1167_cast_fp16 = slice_by_index(begin = var_1167_begin_0, end = var_1167_end_0, end_mask = var_1167_end_mask_0, x = win_13_cast_fp16)[name = string("op_1167_cast_fp16")];
            string var_1173_pad_type_0 = const()[name = string("op_1173_pad_type_0"), val = string("valid")];
            int32 var_1173_groups_0 = const()[name = string("op_1173_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1173_strides_0 = const()[name = string("op_1173_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1173_pad_0 = const()[name = string("op_1173_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1173_dilations_0 = const()[name = string("op_1173_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_4_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60898688)))];
            tensor<fp16, [1, 768, 12]> var_1173_cast_fp16 = conv(dilations = var_1173_dilations_0, groups = var_1173_groups_0, pad = var_1173_pad_0, pad_type = var_1173_pad_type_0, strides = var_1173_strides_0, weight = core_backbone_layers_4_conv_conv_weight_to_fp16, x = win_13_cast_fp16)[name = string("op_1173_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_1175_cast_fp16 = mul(x = c_7_cast_fp16, y = var_1173_cast_fp16)[name = string("op_1175_cast_fp16")];
            tensor<int32, [3]> input_71_perm_0 = const()[name = string("input_71_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_4_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60903360)))];
            tensor<fp16, [1, 12, 768]> input_71_cast_fp16 = transpose(perm = input_71_perm_0, x = var_1175_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [1, 12, 768]> linear_23_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_4_conv_out_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_47_cast_fp16 = add(x = x_45_cast_fp16, y = linear_23_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1181_cast_fp16 = abs(x = x_47_cast_fp16)[name = string("op_1181_cast_fp16")];
            tensor<int32, [1]> var_1183_axes_0 = const()[name = string("op_1183_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1183_cast_fp16 = reduce_max(axes = var_1183_axes_0, keep_dims = var_128, x = var_1181_cast_fp16)[name = string("op_1183_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_25_cast_fp16 = maximum(x = var_1183_cast_fp16, y = var_129_to_fp16)[name = string("s_25_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_31_cast_fp16 = real_div(x = x_47_cast_fp16, y = s_25_cast_fp16)[name = string("y_31_cast_fp16")];
            fp16 var_130_promoted_11_to_fp16 = const()[name = string("op_130_promoted_11_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1186_cast_fp16 = pow(x = y_31_cast_fp16, y = var_130_promoted_11_to_fp16)[name = string("op_1186_cast_fp16")];
            tensor<int32, [1]> var_1188_axes_0 = const()[name = string("op_1188_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1188_keep_dims_0 = const()[name = string("op_1188_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1188_cast_fp16 = reduce_mean(axes = var_1188_axes_0, keep_dims = var_1188_keep_dims_0, x = var_1186_cast_fp16)[name = string("op_1188_cast_fp16")];
            fp32 var_1189_epsilon_0 = const()[name = string("op_1189_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1189_cast_fp16 = inverse(epsilon = var_1189_epsilon_0, x = s_25_cast_fp16)[name = string("op_1189_cast_fp16")];
            fp16 var_1190_to_fp16 = const()[name = string("op_1190_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1191_cast_fp16 = mul(x = var_1189_cast_fp16, y = var_1190_to_fp16)[name = string("op_1191_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1192_cast_fp16 = real_div(x = var_1191_cast_fp16, y = s_25_cast_fp16)[name = string("op_1192_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1193_cast_fp16 = add(x = var_1188_cast_fp16, y = var_1192_cast_fp16)[name = string("op_1193_cast_fp16")];
            fp32 var_1194_epsilon_0 = const()[name = string("op_1194_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1194_cast_fp16 = rsqrt(epsilon = var_1194_epsilon_0, x = var_1193_cast_fp16)[name = string("op_1194_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1195_cast_fp16 = mul(x = y_31_cast_fp16, y = var_1194_cast_fp16)[name = string("op_1195_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_4_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62083072)))];
            tensor<fp16, [1, 12, 768]> input_73_cast_fp16 = mul(x = core_backbone_layers_4_ffn_norm_weight_to_fp16, y = var_1195_cast_fp16)[name = string("input_73_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_4_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62084672)))];
            tensor<fp16, [1, 12, 1792]> linear_24_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w1_weight_to_fp16, x = input_73_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_4_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64837248)))];
            tensor<fp16, [1, 12, 1792]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w3_weight_to_fp16, x = input_73_cast_fp16)[name = string("linear_25_cast_fp16")];
            fp16 var_1204_to_fp16 = const()[name = string("op_1204_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1205_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_1204_to_fp16)[name = string("op_1205_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1208_cast_fp16 = tanh(x = var_1205_cast_fp16)[name = string("op_1208_cast_fp16")];
            fp16 var_1209_to_fp16 = const()[name = string("op_1209_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1210_cast_fp16 = add(x = var_1208_cast_fp16, y = var_1209_to_fp16)[name = string("op_1210_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1211_cast_fp16 = mul(x = var_1205_cast_fp16, y = var_1210_cast_fp16)[name = string("op_1211_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_75_cast_fp16 = mul(x = var_1211_cast_fp16, y = linear_25_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_4_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67589824)))];
            tensor<fp16, [1, 12, 768]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w2_weight_to_fp16, x = input_75_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_51_cast_fp16 = add(x = x_47_cast_fp16, y = linear_26_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1217_cast_fp16 = abs(x = x_51_cast_fp16)[name = string("op_1217_cast_fp16")];
            tensor<int32, [1]> var_1219_axes_0 = const()[name = string("op_1219_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1219_cast_fp16 = reduce_max(axes = var_1219_axes_0, keep_dims = var_128, x = var_1217_cast_fp16)[name = string("op_1219_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_27_cast_fp16 = maximum(x = var_1219_cast_fp16, y = var_129_to_fp16)[name = string("s_27_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_33_cast_fp16 = real_div(x = x_51_cast_fp16, y = s_27_cast_fp16)[name = string("y_33_cast_fp16")];
            fp16 var_130_promoted_12_to_fp16 = const()[name = string("op_130_promoted_12_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1222_cast_fp16 = pow(x = y_33_cast_fp16, y = var_130_promoted_12_to_fp16)[name = string("op_1222_cast_fp16")];
            tensor<int32, [1]> var_1224_axes_0 = const()[name = string("op_1224_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1224_keep_dims_0 = const()[name = string("op_1224_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1224_cast_fp16 = reduce_mean(axes = var_1224_axes_0, keep_dims = var_1224_keep_dims_0, x = var_1222_cast_fp16)[name = string("op_1224_cast_fp16")];
            fp32 var_1225_epsilon_0 = const()[name = string("op_1225_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1225_cast_fp16 = inverse(epsilon = var_1225_epsilon_0, x = s_27_cast_fp16)[name = string("op_1225_cast_fp16")];
            fp16 var_1226_to_fp16 = const()[name = string("op_1226_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1227_cast_fp16 = mul(x = var_1225_cast_fp16, y = var_1226_to_fp16)[name = string("op_1227_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1228_cast_fp16 = real_div(x = var_1227_cast_fp16, y = s_27_cast_fp16)[name = string("op_1228_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1229_cast_fp16 = add(x = var_1224_cast_fp16, y = var_1228_cast_fp16)[name = string("op_1229_cast_fp16")];
            fp32 var_1230_epsilon_0 = const()[name = string("op_1230_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1230_cast_fp16 = rsqrt(epsilon = var_1230_epsilon_0, x = var_1229_cast_fp16)[name = string("op_1230_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1231_cast_fp16 = mul(x = y_33_cast_fp16, y = var_1230_cast_fp16)[name = string("op_1231_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_5_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70342400)))];
            tensor<fp16, [1, 12, 768]> input_77_cast_fp16 = mul(x = core_backbone_layers_5_operator_norm_weight_to_fp16, y = var_1231_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70344000)))];
            tensor<fp16, [1, 12, 768]> linear_27_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_q_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_27_cast_fp16")];
            tensor<int32, [4]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [4]>([1, 12, 8, 96])];
            tensor<fp16, [1, 12, 8, 96]> x_53_cast_fp16 = reshape(shape = concat_9, x = linear_27_cast_fp16)[name = string("x_53_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1238_cast_fp16 = abs(x = x_53_cast_fp16)[name = string("op_1238_cast_fp16")];
            tensor<int32, [1]> var_1240_axes_0 = const()[name = string("op_1240_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 8, 1]> var_1240_cast_fp16 = reduce_max(axes = var_1240_axes_0, keep_dims = var_128, x = var_1238_cast_fp16)[name = string("op_1240_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> s_29_cast_fp16 = maximum(x = var_1240_cast_fp16, y = var_129_to_fp16)[name = string("s_29_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> y_35_cast_fp16 = real_div(x = x_53_cast_fp16, y = s_29_cast_fp16)[name = string("y_35_cast_fp16")];
            fp16 var_130_promoted_13_to_fp16 = const()[name = string("op_130_promoted_13_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 8, 96]> var_1243_cast_fp16 = pow(x = y_35_cast_fp16, y = var_130_promoted_13_to_fp16)[name = string("op_1243_cast_fp16")];
            tensor<int32, [1]> var_1245_axes_0 = const()[name = string("op_1245_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1245_keep_dims_0 = const()[name = string("op_1245_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 8, 1]> var_1245_cast_fp16 = reduce_mean(axes = var_1245_axes_0, keep_dims = var_1245_keep_dims_0, x = var_1243_cast_fp16)[name = string("op_1245_cast_fp16")];
            fp32 var_1246_epsilon_0 = const()[name = string("op_1246_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 8, 1]> var_1246_cast_fp16 = inverse(epsilon = var_1246_epsilon_0, x = s_29_cast_fp16)[name = string("op_1246_cast_fp16")];
            fp16 var_1247_to_fp16 = const()[name = string("op_1247_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 8, 1]> var_1248_cast_fp16 = mul(x = var_1246_cast_fp16, y = var_1247_to_fp16)[name = string("op_1248_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_1249_cast_fp16 = real_div(x = var_1248_cast_fp16, y = s_29_cast_fp16)[name = string("op_1249_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_1250_cast_fp16 = add(x = var_1245_cast_fp16, y = var_1249_cast_fp16)[name = string("op_1250_cast_fp16")];
            fp32 var_1251_epsilon_0 = const()[name = string("op_1251_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 8, 1]> var_1251_cast_fp16 = rsqrt(epsilon = var_1251_epsilon_0, x = var_1250_cast_fp16)[name = string("op_1251_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1252_cast_fp16 = mul(x = y_35_cast_fp16, y = var_1251_cast_fp16)[name = string("op_1252_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_5_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71523712)))];
            tensor<fp16, [1, 12, 8, 96]> var_1253_cast_fp16 = mul(x = core_backbone_layers_5_self_attn_q_norm_weight_to_fp16, y = var_1252_cast_fp16)[name = string("op_1253_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71523968)))];
            tensor<fp16, [1, 12, 384]> linear_28_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_k_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [4]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> x_55_cast_fp16 = reshape(shape = concat_10, x = linear_28_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_1260_cast_fp16 = abs(x = x_55_cast_fp16)[name = string("op_1260_cast_fp16")];
            tensor<int32, [1]> var_1262_axes_0 = const()[name = string("op_1262_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 4, 1]> var_1262_cast_fp16 = reduce_max(axes = var_1262_axes_0, keep_dims = var_128, x = var_1260_cast_fp16)[name = string("op_1262_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> s_31_cast_fp16 = maximum(x = var_1262_cast_fp16, y = var_129_to_fp16)[name = string("s_31_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> y_37_cast_fp16 = real_div(x = x_55_cast_fp16, y = s_31_cast_fp16)[name = string("y_37_cast_fp16")];
            fp16 var_130_promoted_14_to_fp16 = const()[name = string("op_130_promoted_14_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 4, 96]> var_1265_cast_fp16 = pow(x = y_37_cast_fp16, y = var_130_promoted_14_to_fp16)[name = string("op_1265_cast_fp16")];
            tensor<int32, [1]> var_1267_axes_0 = const()[name = string("op_1267_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1267_keep_dims_0 = const()[name = string("op_1267_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 4, 1]> var_1267_cast_fp16 = reduce_mean(axes = var_1267_axes_0, keep_dims = var_1267_keep_dims_0, x = var_1265_cast_fp16)[name = string("op_1267_cast_fp16")];
            fp32 var_1268_epsilon_0 = const()[name = string("op_1268_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 4, 1]> var_1268_cast_fp16 = inverse(epsilon = var_1268_epsilon_0, x = s_31_cast_fp16)[name = string("op_1268_cast_fp16")];
            fp16 var_1269_to_fp16 = const()[name = string("op_1269_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 4, 1]> var_1270_cast_fp16 = mul(x = var_1268_cast_fp16, y = var_1269_to_fp16)[name = string("op_1270_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_1271_cast_fp16 = real_div(x = var_1270_cast_fp16, y = s_31_cast_fp16)[name = string("op_1271_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_1272_cast_fp16 = add(x = var_1267_cast_fp16, y = var_1271_cast_fp16)[name = string("op_1272_cast_fp16")];
            fp32 var_1273_epsilon_0 = const()[name = string("op_1273_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 4, 1]> var_1273_cast_fp16 = rsqrt(epsilon = var_1273_epsilon_0, x = var_1272_cast_fp16)[name = string("op_1273_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_1274_cast_fp16 = mul(x = y_37_cast_fp16, y = var_1273_cast_fp16)[name = string("op_1274_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_5_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72113856)))];
            tensor<fp16, [1, 12, 4, 96]> var_1275_cast_fp16 = mul(x = core_backbone_layers_5_self_attn_k_norm_weight_to_fp16, y = var_1274_cast_fp16)[name = string("op_1275_cast_fp16")];
            tensor<int32, [4]> new_13_perm_0 = const()[name = string("new_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72114112)))];
            tensor<fp16, [1, 12, 384]> linear_29_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_v_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<int32, [4]> concat_11 = const()[name = string("concat_11"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> var_1280_cast_fp16 = reshape(shape = concat_11, x = linear_29_cast_fp16)[name = string("op_1280_cast_fp16")];
            tensor<int32, [4]> new_15_perm_0 = const()[name = string("new_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_81_begin_0 = const()[name = string("input_81_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_81_end_0 = const()[name = string("input_81_end_0"), val = tensor<int32, [5]>([2, 1, 4, 31, 96])];
            tensor<bool, [5]> input_81_end_mask_0 = const()[name = string("input_81_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_81_squeeze_mask_0 = const()[name = string("input_81_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_81_cast_fp16 = slice_by_index(begin = input_81_begin_0, end = input_81_end_0, end_mask = input_81_end_mask_0, squeeze_mask = input_81_squeeze_mask_0, x = k)[name = string("input_81_cast_fp16")];
            tensor<int32, [8]> var_1285_pad_0 = const()[name = string("op_1285_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1285_mode_0 = const()[name = string("op_1285_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_13_cast_fp16 = transpose(perm = new_13_perm_0, x = var_1275_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [1, 4, 43, 96]> var_1285_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = var_1285_mode_0, pad = var_1285_pad_0, x = new_13_cast_fp16)[name = string("op_1285_cast_fp16")];
            tensor<int32, [8]> var_1287_pad_0 = const()[name = string("op_1287_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_1287_mode_0 = const()[name = string("op_1287_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_1287_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = var_1287_mode_0, pad = var_1287_pad_0, x = input_81_cast_fp16)[name = string("op_1287_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> x_59_cast_fp16 = add(x = var_1285_cast_fp16, y = var_1287_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<int32, [5]> input_85_begin_0 = const()[name = string("input_85_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_85_end_0 = const()[name = string("input_85_end_0"), val = tensor<int32, [5]>([2, 1, 4, 31, 96])];
            tensor<bool, [5]> input_85_end_mask_0 = const()[name = string("input_85_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_85_squeeze_mask_0 = const()[name = string("input_85_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_85_cast_fp16 = slice_by_index(begin = input_85_begin_0, end = input_85_end_0, end_mask = input_85_end_mask_0, squeeze_mask = input_85_squeeze_mask_0, x = v)[name = string("input_85_cast_fp16")];
            tensor<int32, [8]> var_1292_pad_0 = const()[name = string("op_1292_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1292_mode_0 = const()[name = string("op_1292_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_15_cast_fp16 = transpose(perm = new_15_perm_0, x = var_1280_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [1, 4, 43, 96]> var_1292_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = var_1292_mode_0, pad = var_1292_pad_0, x = new_15_cast_fp16)[name = string("op_1292_cast_fp16")];
            tensor<int32, [8]> var_1294_pad_0 = const()[name = string("op_1294_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_1294_mode_0 = const()[name = string("op_1294_mode_0"), val = string("constant")];
            fp16 const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_1294_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = var_1294_mode_0, pad = var_1294_pad_0, x = input_85_cast_fp16)[name = string("op_1294_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> v_7_cast_fp16 = add(x = var_1292_cast_fp16, y = var_1294_cast_fp16)[name = string("v_7_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1296_cast_fp16 = mul(x = var_1253_cast_fp16, y = const_66_to_fp16)[name = string("op_1296_cast_fp16")];
            tensor<int32, [2]> var_1297_split_sizes_0 = const()[name = string("op_1297_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_71 = const()[name = string("const_71"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 48]> var_1297_cast_fp16_0, tensor<fp16, [1, 12, 8, 48]> var_1297_cast_fp16_1 = split(axis = const_71, split_sizes = var_1297_split_sizes_0, x = var_1253_cast_fp16)[name = string("op_1297_cast_fp16")];
            fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 12, 8, 48]> var_1299_cast_fp16 = mul(x = var_1297_cast_fp16_1, y = const_22_promoted_to_fp16)[name = string("op_1299_cast_fp16")];
            bool var_1301_interleave_0 = const()[name = string("op_1301_interleave_0"), val = bool(false)];
            int32 const_72 = const()[name = string("const_72"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 96]> var_1301_cast_fp16 = concat(axis = const_72, interleave = var_1301_interleave_0, values = (var_1299_cast_fp16, var_1297_cast_fp16_0))[name = string("op_1301_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1302_cast_fp16 = mul(x = var_1301_cast_fp16, y = const_69_to_fp16)[name = string("op_1302_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> q_3_cast_fp16 = add(x = var_1296_cast_fp16, y = var_1302_cast_fp16)[name = string("q_3_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_1304_cast_fp16 = mul(x = x_59_cast_fp16, y = const_0_to_fp16)[name = string("op_1304_cast_fp16")];
            tensor<int32, [2]> var_1305_split_sizes_0 = const()[name = string("op_1305_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1305_axis_0 = const()[name = string("op_1305_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 43, 48]> var_1305_cast_fp16_0, tensor<fp16, [1, 4, 43, 48]> var_1305_cast_fp16_1 = split(axis = var_1305_axis_0, split_sizes = var_1305_split_sizes_0, x = x_59_cast_fp16)[name = string("op_1305_cast_fp16")];
            fp16 const_23_promoted_to_fp16 = const()[name = string("const_23_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 43, 48]> var_1307_cast_fp16 = mul(x = var_1305_cast_fp16_1, y = const_23_promoted_to_fp16)[name = string("op_1307_cast_fp16")];
            bool var_1309_interleave_0 = const()[name = string("op_1309_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 43, 96]> var_1309_cast_fp16 = concat(axis = var_127, interleave = var_1309_interleave_0, values = (var_1307_cast_fp16, var_1305_cast_fp16_0))[name = string("op_1309_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_1310_cast_fp16 = mul(x = var_1309_cast_fp16, y = const_1_to_fp16)[name = string("op_1310_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> k_7_cast_fp16 = add(x = var_1304_cast_fp16, y = var_1310_cast_fp16)[name = string("k_7_cast_fp16")];
            tensor<int32, [1]> var_1312_axes_0 = const()[name = string("op_1312_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_1312_cast_fp16 = expand_dims(axes = var_1312_axes_0, x = k_7_cast_fp16)[name = string("op_1312_cast_fp16")];
            tensor<int32, [5]> var_1314_reps_0 = const()[name = string("op_1314_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_1314_cast_fp16 = tile(reps = var_1314_reps_0, x = var_1312_cast_fp16)[name = string("op_1314_cast_fp16")];
            tensor<int32, [4]> concat_13 = const()[name = string("concat_13"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> k_9_cast_fp16 = reshape(shape = concat_13, x = var_1314_cast_fp16)[name = string("k_9_cast_fp16")];
            tensor<int32, [1]> var_1316_axes_0 = const()[name = string("op_1316_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_1316_cast_fp16 = expand_dims(axes = var_1316_axes_0, x = v_7_cast_fp16)[name = string("op_1316_cast_fp16")];
            tensor<int32, [5]> var_1318_reps_0 = const()[name = string("op_1318_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_1318_cast_fp16 = tile(reps = var_1318_reps_0, x = var_1316_cast_fp16)[name = string("op_1318_cast_fp16")];
            tensor<int32, [4]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> v_9_cast_fp16 = reshape(shape = concat_15, x = var_1318_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> transpose_2_perm_0 = const()[name = string("transpose_2_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1321_transpose_x_1 = const()[name = string("op_1321_transpose_x_1"), val = bool(false)];
            bool var_1321_transpose_y_1 = const()[name = string("op_1321_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 96]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = q_3_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [1, 8, 12, 43]> var_1321_cast_fp16 = matmul(transpose_x = var_1321_transpose_x_1, transpose_y = var_1321_transpose_y_1, x = transpose_2_cast_fp16, y = k_9_cast_fp16)[name = string("op_1321_cast_fp16")];
            fp16 var_1322_to_fp16 = const()[name = string("op_1322_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 12, 43]> s_33_cast_fp16 = mul(x = var_1321_cast_fp16, y = var_1322_to_fp16)[name = string("s_33_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_61_cast_fp16 = add(x = s_33_cast_fp16, y = mask)[name = string("x_61_cast_fp16")];
            tensor<int32, [1]> var_1326_axes_0 = const()[name = string("op_1326_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 12, 1]> var_1326_cast_fp16 = reduce_max(axes = var_1326_axes_0, keep_dims = var_128, x = x_61_cast_fp16)[name = string("op_1326_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_63_cast_fp16 = sub(x = x_61_cast_fp16, y = var_1326_cast_fp16)[name = string("x_63_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_65_cast_fp16 = exp(x = x_63_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [1]> var_1330_axes_0 = const()[name = string("op_1330_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1330_keep_dims_0 = const()[name = string("op_1330_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 1]> var_1330_cast_fp16 = reduce_sum(axes = var_1330_axes_0, keep_dims = var_1330_keep_dims_0, x = x_65_cast_fp16)[name = string("op_1330_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> var_1331_cast_fp16 = real_div(x = x_65_cast_fp16, y = var_1330_cast_fp16)[name = string("op_1331_cast_fp16")];
            bool o_5_transpose_x_0 = const()[name = string("o_5_transpose_x_0"), val = bool(false)];
            bool o_5_transpose_y_0 = const()[name = string("o_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 12, 96]> o_5_cast_fp16 = matmul(transpose_x = o_5_transpose_x_0, transpose_y = o_5_transpose_y_0, x = var_1331_cast_fp16, y = v_9_cast_fp16)[name = string("o_5_cast_fp16")];
            tensor<int32, [4]> o_7_perm_0 = const()[name = string("o_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [3]>([1, 12, 768])];
            tensor<fp16, [1, 12, 8, 96]> o_7_cast_fp16 = transpose(perm = o_7_perm_0, x = o_5_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [1, 12, 768]> input_87_cast_fp16 = reshape(shape = concat_16, x = o_7_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72704000)))];
            tensor<fp16, [1, 12, 768]> linear_30_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_o_proj_weight_to_fp16, x = input_87_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> var_1340_begin_0 = const()[name = string("op_1340_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_1340_end_0 = const()[name = string("op_1340_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_1340_end_mask_0 = const()[name = string("op_1340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1340_cast_fp16 = slice_by_index(begin = var_1340_begin_0, end = var_1340_end_0, end_mask = var_1340_end_mask_0, x = x_59_cast_fp16)[name = string("op_1340_cast_fp16")];
            tensor<int32, [4]> var_1343_begin_0 = const()[name = string("op_1343_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_1343_end_0 = const()[name = string("op_1343_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_1343_end_mask_0 = const()[name = string("op_1343_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1343_cast_fp16 = slice_by_index(begin = var_1343_begin_0, end = var_1343_end_0, end_mask = var_1343_end_mask_0, x = v_7_cast_fp16)[name = string("op_1343_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_67_cast_fp16 = add(x = x_51_cast_fp16, y = linear_30_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1346_cast_fp16 = abs(x = x_67_cast_fp16)[name = string("op_1346_cast_fp16")];
            tensor<int32, [1]> var_1348_axes_0 = const()[name = string("op_1348_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1348_cast_fp16 = reduce_max(axes = var_1348_axes_0, keep_dims = var_128, x = var_1346_cast_fp16)[name = string("op_1348_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_35_cast_fp16 = maximum(x = var_1348_cast_fp16, y = var_129_to_fp16)[name = string("s_35_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_39_cast_fp16 = real_div(x = x_67_cast_fp16, y = s_35_cast_fp16)[name = string("y_39_cast_fp16")];
            fp16 var_130_promoted_15_to_fp16 = const()[name = string("op_130_promoted_15_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1351_cast_fp16 = pow(x = y_39_cast_fp16, y = var_130_promoted_15_to_fp16)[name = string("op_1351_cast_fp16")];
            tensor<int32, [1]> var_1353_axes_0 = const()[name = string("op_1353_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1353_keep_dims_0 = const()[name = string("op_1353_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1353_cast_fp16 = reduce_mean(axes = var_1353_axes_0, keep_dims = var_1353_keep_dims_0, x = var_1351_cast_fp16)[name = string("op_1353_cast_fp16")];
            fp32 var_1354_epsilon_0 = const()[name = string("op_1354_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1354_cast_fp16 = inverse(epsilon = var_1354_epsilon_0, x = s_35_cast_fp16)[name = string("op_1354_cast_fp16")];
            fp16 var_1355_to_fp16 = const()[name = string("op_1355_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1356_cast_fp16 = mul(x = var_1354_cast_fp16, y = var_1355_to_fp16)[name = string("op_1356_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1357_cast_fp16 = real_div(x = var_1356_cast_fp16, y = s_35_cast_fp16)[name = string("op_1357_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1358_cast_fp16 = add(x = var_1353_cast_fp16, y = var_1357_cast_fp16)[name = string("op_1358_cast_fp16")];
            fp32 var_1359_epsilon_0 = const()[name = string("op_1359_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1359_cast_fp16 = rsqrt(epsilon = var_1359_epsilon_0, x = var_1358_cast_fp16)[name = string("op_1359_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1360_cast_fp16 = mul(x = y_39_cast_fp16, y = var_1359_cast_fp16)[name = string("op_1360_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_5_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73883712)))];
            tensor<fp16, [1, 12, 768]> input_89_cast_fp16 = mul(x = core_backbone_layers_5_ffn_norm_weight_to_fp16, y = var_1360_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_5_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73885312)))];
            tensor<fp16, [1, 12, 1792]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w1_weight_to_fp16, x = input_89_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_5_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76637888)))];
            tensor<fp16, [1, 12, 1792]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w3_weight_to_fp16, x = input_89_cast_fp16)[name = string("linear_32_cast_fp16")];
            fp16 var_1369_to_fp16 = const()[name = string("op_1369_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1370_cast_fp16 = mul(x = linear_31_cast_fp16, y = var_1369_to_fp16)[name = string("op_1370_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1373_cast_fp16 = tanh(x = var_1370_cast_fp16)[name = string("op_1373_cast_fp16")];
            fp16 var_1374_to_fp16 = const()[name = string("op_1374_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1375_cast_fp16 = add(x = var_1373_cast_fp16, y = var_1374_to_fp16)[name = string("op_1375_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1376_cast_fp16 = mul(x = var_1370_cast_fp16, y = var_1375_cast_fp16)[name = string("op_1376_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_91_cast_fp16 = mul(x = var_1376_cast_fp16, y = linear_32_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_5_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79390464)))];
            tensor<fp16, [1, 12, 768]> linear_33_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w2_weight_to_fp16, x = input_91_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_71_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1382_cast_fp16 = abs(x = x_71_cast_fp16)[name = string("op_1382_cast_fp16")];
            tensor<int32, [1]> var_1384_axes_0 = const()[name = string("op_1384_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1384_cast_fp16 = reduce_max(axes = var_1384_axes_0, keep_dims = var_128, x = var_1382_cast_fp16)[name = string("op_1384_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_37_cast_fp16 = maximum(x = var_1384_cast_fp16, y = var_129_to_fp16)[name = string("s_37_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_41_cast_fp16 = real_div(x = x_71_cast_fp16, y = s_37_cast_fp16)[name = string("y_41_cast_fp16")];
            fp16 var_130_promoted_16_to_fp16 = const()[name = string("op_130_promoted_16_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1387_cast_fp16 = pow(x = y_41_cast_fp16, y = var_130_promoted_16_to_fp16)[name = string("op_1387_cast_fp16")];
            tensor<int32, [1]> var_1389_axes_0 = const()[name = string("op_1389_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1389_keep_dims_0 = const()[name = string("op_1389_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1389_cast_fp16 = reduce_mean(axes = var_1389_axes_0, keep_dims = var_1389_keep_dims_0, x = var_1387_cast_fp16)[name = string("op_1389_cast_fp16")];
            fp32 var_1390_epsilon_0 = const()[name = string("op_1390_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1390_cast_fp16 = inverse(epsilon = var_1390_epsilon_0, x = s_37_cast_fp16)[name = string("op_1390_cast_fp16")];
            fp16 var_1391_to_fp16 = const()[name = string("op_1391_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1392_cast_fp16 = mul(x = var_1390_cast_fp16, y = var_1391_to_fp16)[name = string("op_1392_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1393_cast_fp16 = real_div(x = var_1392_cast_fp16, y = s_37_cast_fp16)[name = string("op_1393_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1394_cast_fp16 = add(x = var_1389_cast_fp16, y = var_1393_cast_fp16)[name = string("op_1394_cast_fp16")];
            fp32 var_1395_epsilon_0 = const()[name = string("op_1395_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1395_cast_fp16 = rsqrt(epsilon = var_1395_epsilon_0, x = var_1394_cast_fp16)[name = string("op_1395_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1396_cast_fp16 = mul(x = y_41_cast_fp16, y = var_1395_cast_fp16)[name = string("op_1396_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_6_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82143040)))];
            tensor<fp16, [1, 12, 768]> input_93_cast_fp16 = mul(x = core_backbone_layers_6_operator_norm_weight_to_fp16, y = var_1396_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_6_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82144640)))];
            tensor<fp16, [1, 12, 2304]> linear_34_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_6_conv_in_proj_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<int32, [3]> bcx_9_perm_0 = const()[name = string("bcx_9_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_21_begin_0 = const()[name = string("b_21_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_21_end_0 = const()[name = string("b_21_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_21_end_mask_0 = const()[name = string("b_21_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_9_cast_fp16 = transpose(perm = bcx_9_perm_0, x = linear_34_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [1, 768, 12]> b_21_cast_fp16 = slice_by_index(begin = b_21_begin_0, end = b_21_end_0, end_mask = b_21_end_mask_0, x = bcx_9_cast_fp16)[name = string("b_21_cast_fp16")];
            tensor<int32, [3]> c_9_begin_0 = const()[name = string("c_9_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_9_end_0 = const()[name = string("c_9_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_9_end_mask_0 = const()[name = string("c_9_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_9_cast_fp16 = slice_by_index(begin = c_9_begin_0, end = c_9_end_0, end_mask = c_9_end_mask_0, x = bcx_9_cast_fp16)[name = string("c_9_cast_fp16")];
            tensor<int32, [3]> xx_9_begin_0 = const()[name = string("xx_9_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_9_end_0 = const()[name = string("xx_9_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_9_end_mask_0 = const()[name = string("xx_9_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_9_cast_fp16 = slice_by_index(begin = xx_9_begin_0, end = xx_9_end_0, end_mask = xx_9_end_mask_0, x = bcx_9_cast_fp16)[name = string("xx_9_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_17_cast_fp16 = mul(x = b_21_cast_fp16, y = xx_9_cast_fp16)[name = string("new_17_cast_fp16")];
            tensor<int32, [4]> input_97_begin_0 = const()[name = string("input_97_begin_0"), val = tensor<int32, [4]>([4, 0, 0, 0])];
            tensor<int32, [4]> input_97_end_0 = const()[name = string("input_97_end_0"), val = tensor<int32, [4]>([5, 1, 768, 2])];
            tensor<bool, [4]> input_97_end_mask_0 = const()[name = string("input_97_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_97_squeeze_mask_0 = const()[name = string("input_97_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_97_cast_fp16 = slice_by_index(begin = input_97_begin_0, end = input_97_end_0, end_mask = input_97_end_mask_0, squeeze_mask = input_97_squeeze_mask_0, x = conv)[name = string("input_97_cast_fp16")];
            tensor<int32, [6]> var_1411_pad_0 = const()[name = string("op_1411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1411_mode_0 = const()[name = string("op_1411_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1411_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_1411_mode_0, pad = var_1411_pad_0, x = new_17_cast_fp16)[name = string("op_1411_cast_fp16")];
            tensor<int32, [6]> var_1413_pad_0 = const()[name = string("op_1413_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_1413_mode_0 = const()[name = string("op_1413_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1413_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = var_1413_mode_0, pad = var_1413_pad_0, x = input_97_cast_fp16)[name = string("op_1413_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_17_cast_fp16 = add(x = var_1411_cast_fp16, y = var_1413_cast_fp16)[name = string("win_17_cast_fp16")];
            tensor<int32, [3]> var_1415_begin_0 = const()[name = string("op_1415_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_1415_end_0 = const()[name = string("op_1415_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_1415_end_mask_0 = const()[name = string("op_1415_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1415_cast_fp16 = slice_by_index(begin = var_1415_begin_0, end = var_1415_end_0, end_mask = var_1415_end_mask_0, x = win_17_cast_fp16)[name = string("op_1415_cast_fp16")];
            string var_1421_pad_type_0 = const()[name = string("op_1421_pad_type_0"), val = string("valid")];
            int32 var_1421_groups_0 = const()[name = string("op_1421_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1421_strides_0 = const()[name = string("op_1421_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1421_pad_0 = const()[name = string("op_1421_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1421_dilations_0 = const()[name = string("op_1421_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_6_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85683648)))];
            tensor<fp16, [1, 768, 12]> var_1421_cast_fp16 = conv(dilations = var_1421_dilations_0, groups = var_1421_groups_0, pad = var_1421_pad_0, pad_type = var_1421_pad_type_0, strides = var_1421_strides_0, weight = core_backbone_layers_6_conv_conv_weight_to_fp16, x = win_17_cast_fp16)[name = string("op_1421_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_1423_cast_fp16 = mul(x = c_9_cast_fp16, y = var_1421_cast_fp16)[name = string("op_1423_cast_fp16")];
            tensor<int32, [3]> input_99_perm_0 = const()[name = string("input_99_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_6_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85688320)))];
            tensor<fp16, [1, 12, 768]> input_99_cast_fp16 = transpose(perm = input_99_perm_0, x = var_1423_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [1, 12, 768]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_6_conv_out_proj_weight_to_fp16, x = input_99_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_73_cast_fp16 = add(x = x_71_cast_fp16, y = linear_35_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1429_cast_fp16 = abs(x = x_73_cast_fp16)[name = string("op_1429_cast_fp16")];
            tensor<int32, [1]> var_1431_axes_0 = const()[name = string("op_1431_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1431_cast_fp16 = reduce_max(axes = var_1431_axes_0, keep_dims = var_128, x = var_1429_cast_fp16)[name = string("op_1431_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_39_cast_fp16 = maximum(x = var_1431_cast_fp16, y = var_129_to_fp16)[name = string("s_39_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_45_cast_fp16 = real_div(x = x_73_cast_fp16, y = s_39_cast_fp16)[name = string("y_45_cast_fp16")];
            fp16 var_130_promoted_17_to_fp16 = const()[name = string("op_130_promoted_17_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1434_cast_fp16 = pow(x = y_45_cast_fp16, y = var_130_promoted_17_to_fp16)[name = string("op_1434_cast_fp16")];
            tensor<int32, [1]> var_1436_axes_0 = const()[name = string("op_1436_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1436_keep_dims_0 = const()[name = string("op_1436_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1436_cast_fp16 = reduce_mean(axes = var_1436_axes_0, keep_dims = var_1436_keep_dims_0, x = var_1434_cast_fp16)[name = string("op_1436_cast_fp16")];
            fp32 var_1437_epsilon_0 = const()[name = string("op_1437_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1437_cast_fp16 = inverse(epsilon = var_1437_epsilon_0, x = s_39_cast_fp16)[name = string("op_1437_cast_fp16")];
            fp16 var_1438_to_fp16 = const()[name = string("op_1438_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1439_cast_fp16 = mul(x = var_1437_cast_fp16, y = var_1438_to_fp16)[name = string("op_1439_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1440_cast_fp16 = real_div(x = var_1439_cast_fp16, y = s_39_cast_fp16)[name = string("op_1440_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1441_cast_fp16 = add(x = var_1436_cast_fp16, y = var_1440_cast_fp16)[name = string("op_1441_cast_fp16")];
            fp32 var_1442_epsilon_0 = const()[name = string("op_1442_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1442_cast_fp16 = rsqrt(epsilon = var_1442_epsilon_0, x = var_1441_cast_fp16)[name = string("op_1442_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1443_cast_fp16 = mul(x = y_45_cast_fp16, y = var_1442_cast_fp16)[name = string("op_1443_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_6_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86868032)))];
            tensor<fp16, [1, 12, 768]> input_101_cast_fp16 = mul(x = core_backbone_layers_6_ffn_norm_weight_to_fp16, y = var_1443_cast_fp16)[name = string("input_101_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_6_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86869632)))];
            tensor<fp16, [1, 12, 1792]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w1_weight_to_fp16, x = input_101_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_6_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89622208)))];
            tensor<fp16, [1, 12, 1792]> linear_37_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w3_weight_to_fp16, x = input_101_cast_fp16)[name = string("linear_37_cast_fp16")];
            fp16 var_1452_to_fp16 = const()[name = string("op_1452_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1453_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_1452_to_fp16)[name = string("op_1453_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1456_cast_fp16 = tanh(x = var_1453_cast_fp16)[name = string("op_1456_cast_fp16")];
            fp16 var_1457_to_fp16 = const()[name = string("op_1457_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1458_cast_fp16 = add(x = var_1456_cast_fp16, y = var_1457_to_fp16)[name = string("op_1458_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1459_cast_fp16 = mul(x = var_1453_cast_fp16, y = var_1458_cast_fp16)[name = string("op_1459_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_103_cast_fp16 = mul(x = var_1459_cast_fp16, y = linear_37_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_6_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92374784)))];
            tensor<fp16, [1, 12, 768]> linear_38_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w2_weight_to_fp16, x = input_103_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_77_cast_fp16 = add(x = x_73_cast_fp16, y = linear_38_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1465_cast_fp16 = abs(x = x_77_cast_fp16)[name = string("op_1465_cast_fp16")];
            tensor<int32, [1]> var_1467_axes_0 = const()[name = string("op_1467_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1467_cast_fp16 = reduce_max(axes = var_1467_axes_0, keep_dims = var_128, x = var_1465_cast_fp16)[name = string("op_1467_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_41_cast_fp16 = maximum(x = var_1467_cast_fp16, y = var_129_to_fp16)[name = string("s_41_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_47_cast_fp16 = real_div(x = x_77_cast_fp16, y = s_41_cast_fp16)[name = string("y_47_cast_fp16")];
            fp16 var_130_promoted_18_to_fp16 = const()[name = string("op_130_promoted_18_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1470_cast_fp16 = pow(x = y_47_cast_fp16, y = var_130_promoted_18_to_fp16)[name = string("op_1470_cast_fp16")];
            tensor<int32, [1]> var_1472_axes_0 = const()[name = string("op_1472_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1472_keep_dims_0 = const()[name = string("op_1472_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1472_cast_fp16 = reduce_mean(axes = var_1472_axes_0, keep_dims = var_1472_keep_dims_0, x = var_1470_cast_fp16)[name = string("op_1472_cast_fp16")];
            fp32 var_1473_epsilon_0 = const()[name = string("op_1473_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1473_cast_fp16 = inverse(epsilon = var_1473_epsilon_0, x = s_41_cast_fp16)[name = string("op_1473_cast_fp16")];
            fp16 var_1474_to_fp16 = const()[name = string("op_1474_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1475_cast_fp16 = mul(x = var_1473_cast_fp16, y = var_1474_to_fp16)[name = string("op_1475_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1476_cast_fp16 = real_div(x = var_1475_cast_fp16, y = s_41_cast_fp16)[name = string("op_1476_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1477_cast_fp16 = add(x = var_1472_cast_fp16, y = var_1476_cast_fp16)[name = string("op_1477_cast_fp16")];
            fp32 var_1478_epsilon_0 = const()[name = string("op_1478_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1478_cast_fp16 = rsqrt(epsilon = var_1478_epsilon_0, x = var_1477_cast_fp16)[name = string("op_1478_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1479_cast_fp16 = mul(x = y_47_cast_fp16, y = var_1478_cast_fp16)[name = string("op_1479_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_7_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95127360)))];
            tensor<fp16, [1, 12, 768]> input_105_cast_fp16 = mul(x = core_backbone_layers_7_operator_norm_weight_to_fp16, y = var_1479_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_7_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95128960)))];
            tensor<fp16, [1, 12, 2304]> linear_39_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_7_conv_in_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [3]> bcx_11_perm_0 = const()[name = string("bcx_11_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_25_begin_0 = const()[name = string("b_25_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_25_end_0 = const()[name = string("b_25_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_25_end_mask_0 = const()[name = string("b_25_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_11_cast_fp16 = transpose(perm = bcx_11_perm_0, x = linear_39_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [1, 768, 12]> b_25_cast_fp16 = slice_by_index(begin = b_25_begin_0, end = b_25_end_0, end_mask = b_25_end_mask_0, x = bcx_11_cast_fp16)[name = string("b_25_cast_fp16")];
            tensor<int32, [3]> c_11_begin_0 = const()[name = string("c_11_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_11_end_0 = const()[name = string("c_11_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_11_end_mask_0 = const()[name = string("c_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_11_cast_fp16 = slice_by_index(begin = c_11_begin_0, end = c_11_end_0, end_mask = c_11_end_mask_0, x = bcx_11_cast_fp16)[name = string("c_11_cast_fp16")];
            tensor<int32, [3]> xx_11_begin_0 = const()[name = string("xx_11_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_11_end_0 = const()[name = string("xx_11_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_11_end_mask_0 = const()[name = string("xx_11_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_11_cast_fp16 = slice_by_index(begin = xx_11_begin_0, end = xx_11_end_0, end_mask = xx_11_end_mask_0, x = bcx_11_cast_fp16)[name = string("xx_11_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_19_cast_fp16 = mul(x = b_25_cast_fp16, y = xx_11_cast_fp16)[name = string("new_19_cast_fp16")];
            tensor<int32, [4]> input_109_begin_0 = const()[name = string("input_109_begin_0"), val = tensor<int32, [4]>([5, 0, 0, 0])];
            tensor<int32, [4]> input_109_end_0 = const()[name = string("input_109_end_0"), val = tensor<int32, [4]>([6, 1, 768, 2])];
            tensor<bool, [4]> input_109_end_mask_0 = const()[name = string("input_109_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_109_squeeze_mask_0 = const()[name = string("input_109_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_109_cast_fp16 = slice_by_index(begin = input_109_begin_0, end = input_109_end_0, end_mask = input_109_end_mask_0, squeeze_mask = input_109_squeeze_mask_0, x = conv)[name = string("input_109_cast_fp16")];
            tensor<int32, [6]> var_1494_pad_0 = const()[name = string("op_1494_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1494_mode_0 = const()[name = string("op_1494_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1494_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = var_1494_mode_0, pad = var_1494_pad_0, x = new_19_cast_fp16)[name = string("op_1494_cast_fp16")];
            tensor<int32, [6]> var_1496_pad_0 = const()[name = string("op_1496_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_1496_mode_0 = const()[name = string("op_1496_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1496_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_1496_mode_0, pad = var_1496_pad_0, x = input_109_cast_fp16)[name = string("op_1496_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_21_cast_fp16 = add(x = var_1494_cast_fp16, y = var_1496_cast_fp16)[name = string("win_21_cast_fp16")];
            tensor<int32, [3]> var_1498_begin_0 = const()[name = string("op_1498_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_1498_end_0 = const()[name = string("op_1498_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_1498_end_mask_0 = const()[name = string("op_1498_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1498_cast_fp16 = slice_by_index(begin = var_1498_begin_0, end = var_1498_end_0, end_mask = var_1498_end_mask_0, x = win_21_cast_fp16)[name = string("op_1498_cast_fp16")];
            string var_1504_pad_type_0 = const()[name = string("op_1504_pad_type_0"), val = string("valid")];
            int32 var_1504_groups_0 = const()[name = string("op_1504_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1504_strides_0 = const()[name = string("op_1504_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1504_pad_0 = const()[name = string("op_1504_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1504_dilations_0 = const()[name = string("op_1504_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_7_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98667968)))];
            tensor<fp16, [1, 768, 12]> var_1504_cast_fp16 = conv(dilations = var_1504_dilations_0, groups = var_1504_groups_0, pad = var_1504_pad_0, pad_type = var_1504_pad_type_0, strides = var_1504_strides_0, weight = core_backbone_layers_7_conv_conv_weight_to_fp16, x = win_21_cast_fp16)[name = string("op_1504_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_1506_cast_fp16 = mul(x = c_11_cast_fp16, y = var_1504_cast_fp16)[name = string("op_1506_cast_fp16")];
            tensor<int32, [3]> input_111_perm_0 = const()[name = string("input_111_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_7_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98672640)))];
            tensor<fp16, [1, 12, 768]> input_111_cast_fp16 = transpose(perm = input_111_perm_0, x = var_1506_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [1, 12, 768]> linear_40_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_conv_out_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_79_cast_fp16 = add(x = x_77_cast_fp16, y = linear_40_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1512_cast_fp16 = abs(x = x_79_cast_fp16)[name = string("op_1512_cast_fp16")];
            tensor<int32, [1]> var_1514_axes_0 = const()[name = string("op_1514_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1514_cast_fp16 = reduce_max(axes = var_1514_axes_0, keep_dims = var_128, x = var_1512_cast_fp16)[name = string("op_1514_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_43_cast_fp16 = maximum(x = var_1514_cast_fp16, y = var_129_to_fp16)[name = string("s_43_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_51_cast_fp16 = real_div(x = x_79_cast_fp16, y = s_43_cast_fp16)[name = string("y_51_cast_fp16")];
            fp16 var_130_promoted_19_to_fp16 = const()[name = string("op_130_promoted_19_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1517_cast_fp16 = pow(x = y_51_cast_fp16, y = var_130_promoted_19_to_fp16)[name = string("op_1517_cast_fp16")];
            tensor<int32, [1]> var_1519_axes_0 = const()[name = string("op_1519_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1519_keep_dims_0 = const()[name = string("op_1519_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1519_cast_fp16 = reduce_mean(axes = var_1519_axes_0, keep_dims = var_1519_keep_dims_0, x = var_1517_cast_fp16)[name = string("op_1519_cast_fp16")];
            fp32 var_1520_epsilon_0 = const()[name = string("op_1520_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1520_cast_fp16 = inverse(epsilon = var_1520_epsilon_0, x = s_43_cast_fp16)[name = string("op_1520_cast_fp16")];
            fp16 var_1521_to_fp16 = const()[name = string("op_1521_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1522_cast_fp16 = mul(x = var_1520_cast_fp16, y = var_1521_to_fp16)[name = string("op_1522_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1523_cast_fp16 = real_div(x = var_1522_cast_fp16, y = s_43_cast_fp16)[name = string("op_1523_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1524_cast_fp16 = add(x = var_1519_cast_fp16, y = var_1523_cast_fp16)[name = string("op_1524_cast_fp16")];
            fp32 var_1525_epsilon_0 = const()[name = string("op_1525_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1525_cast_fp16 = rsqrt(epsilon = var_1525_epsilon_0, x = var_1524_cast_fp16)[name = string("op_1525_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1526_cast_fp16 = mul(x = y_51_cast_fp16, y = var_1525_cast_fp16)[name = string("op_1526_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_7_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99852352)))];
            tensor<fp16, [1, 12, 768]> input_113_cast_fp16 = mul(x = core_backbone_layers_7_ffn_norm_weight_to_fp16, y = var_1526_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_7_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99853952)))];
            tensor<fp16, [1, 12, 1792]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w1_weight_to_fp16, x = input_113_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_7_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102606528)))];
            tensor<fp16, [1, 12, 1792]> linear_42_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w3_weight_to_fp16, x = input_113_cast_fp16)[name = string("linear_42_cast_fp16")];
            fp16 var_1535_to_fp16 = const()[name = string("op_1535_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1536_cast_fp16 = mul(x = linear_41_cast_fp16, y = var_1535_to_fp16)[name = string("op_1536_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1539_cast_fp16 = tanh(x = var_1536_cast_fp16)[name = string("op_1539_cast_fp16")];
            fp16 var_1540_to_fp16 = const()[name = string("op_1540_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1541_cast_fp16 = add(x = var_1539_cast_fp16, y = var_1540_to_fp16)[name = string("op_1541_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1542_cast_fp16 = mul(x = var_1536_cast_fp16, y = var_1541_cast_fp16)[name = string("op_1542_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_115_cast_fp16 = mul(x = var_1542_cast_fp16, y = linear_42_cast_fp16)[name = string("input_115_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_7_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105359104)))];
            tensor<fp16, [1, 12, 768]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w2_weight_to_fp16, x = input_115_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_83_cast_fp16 = add(x = x_79_cast_fp16, y = linear_43_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1548_cast_fp16 = abs(x = x_83_cast_fp16)[name = string("op_1548_cast_fp16")];
            tensor<int32, [1]> var_1550_axes_0 = const()[name = string("op_1550_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1550_cast_fp16 = reduce_max(axes = var_1550_axes_0, keep_dims = var_128, x = var_1548_cast_fp16)[name = string("op_1550_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_45_cast_fp16 = maximum(x = var_1550_cast_fp16, y = var_129_to_fp16)[name = string("s_45_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_53_cast_fp16 = real_div(x = x_83_cast_fp16, y = s_45_cast_fp16)[name = string("y_53_cast_fp16")];
            fp16 var_130_promoted_20_to_fp16 = const()[name = string("op_130_promoted_20_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1553_cast_fp16 = pow(x = y_53_cast_fp16, y = var_130_promoted_20_to_fp16)[name = string("op_1553_cast_fp16")];
            tensor<int32, [1]> var_1555_axes_0 = const()[name = string("op_1555_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1555_keep_dims_0 = const()[name = string("op_1555_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1555_cast_fp16 = reduce_mean(axes = var_1555_axes_0, keep_dims = var_1555_keep_dims_0, x = var_1553_cast_fp16)[name = string("op_1555_cast_fp16")];
            fp32 var_1556_epsilon_0 = const()[name = string("op_1556_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1556_cast_fp16 = inverse(epsilon = var_1556_epsilon_0, x = s_45_cast_fp16)[name = string("op_1556_cast_fp16")];
            fp16 var_1557_to_fp16 = const()[name = string("op_1557_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1558_cast_fp16 = mul(x = var_1556_cast_fp16, y = var_1557_to_fp16)[name = string("op_1558_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1559_cast_fp16 = real_div(x = var_1558_cast_fp16, y = s_45_cast_fp16)[name = string("op_1559_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1560_cast_fp16 = add(x = var_1555_cast_fp16, y = var_1559_cast_fp16)[name = string("op_1560_cast_fp16")];
            fp32 var_1561_epsilon_0 = const()[name = string("op_1561_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1561_cast_fp16 = rsqrt(epsilon = var_1561_epsilon_0, x = var_1560_cast_fp16)[name = string("op_1561_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1562_cast_fp16 = mul(x = y_53_cast_fp16, y = var_1561_cast_fp16)[name = string("op_1562_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_8_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108111680)))];
            tensor<fp16, [1, 12, 768]> input_117_cast_fp16 = mul(x = core_backbone_layers_8_operator_norm_weight_to_fp16, y = var_1562_cast_fp16)[name = string("input_117_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108113280)))];
            tensor<fp16, [1, 12, 768]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_q_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<int32, [4]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [4]>([1, 12, 8, 96])];
            tensor<fp16, [1, 12, 8, 96]> x_85_cast_fp16 = reshape(shape = concat_17, x = linear_44_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1569_cast_fp16 = abs(x = x_85_cast_fp16)[name = string("op_1569_cast_fp16")];
            tensor<int32, [1]> var_1571_axes_0 = const()[name = string("op_1571_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 8, 1]> var_1571_cast_fp16 = reduce_max(axes = var_1571_axes_0, keep_dims = var_128, x = var_1569_cast_fp16)[name = string("op_1571_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> s_47_cast_fp16 = maximum(x = var_1571_cast_fp16, y = var_129_to_fp16)[name = string("s_47_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> y_55_cast_fp16 = real_div(x = x_85_cast_fp16, y = s_47_cast_fp16)[name = string("y_55_cast_fp16")];
            fp16 var_130_promoted_21_to_fp16 = const()[name = string("op_130_promoted_21_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 8, 96]> var_1574_cast_fp16 = pow(x = y_55_cast_fp16, y = var_130_promoted_21_to_fp16)[name = string("op_1574_cast_fp16")];
            tensor<int32, [1]> var_1576_axes_0 = const()[name = string("op_1576_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1576_keep_dims_0 = const()[name = string("op_1576_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 8, 1]> var_1576_cast_fp16 = reduce_mean(axes = var_1576_axes_0, keep_dims = var_1576_keep_dims_0, x = var_1574_cast_fp16)[name = string("op_1576_cast_fp16")];
            fp32 var_1577_epsilon_0 = const()[name = string("op_1577_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 8, 1]> var_1577_cast_fp16 = inverse(epsilon = var_1577_epsilon_0, x = s_47_cast_fp16)[name = string("op_1577_cast_fp16")];
            fp16 var_1578_to_fp16 = const()[name = string("op_1578_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 8, 1]> var_1579_cast_fp16 = mul(x = var_1577_cast_fp16, y = var_1578_to_fp16)[name = string("op_1579_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_1580_cast_fp16 = real_div(x = var_1579_cast_fp16, y = s_47_cast_fp16)[name = string("op_1580_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_1581_cast_fp16 = add(x = var_1576_cast_fp16, y = var_1580_cast_fp16)[name = string("op_1581_cast_fp16")];
            fp32 var_1582_epsilon_0 = const()[name = string("op_1582_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 8, 1]> var_1582_cast_fp16 = rsqrt(epsilon = var_1582_epsilon_0, x = var_1581_cast_fp16)[name = string("op_1582_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1583_cast_fp16 = mul(x = y_55_cast_fp16, y = var_1582_cast_fp16)[name = string("op_1583_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_8_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109292992)))];
            tensor<fp16, [1, 12, 8, 96]> var_1584_cast_fp16 = mul(x = core_backbone_layers_8_self_attn_q_norm_weight_to_fp16, y = var_1583_cast_fp16)[name = string("op_1584_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109293248)))];
            tensor<fp16, [1, 12, 384]> linear_45_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_k_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<int32, [4]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> x_87_cast_fp16 = reshape(shape = concat_18, x = linear_45_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_1591_cast_fp16 = abs(x = x_87_cast_fp16)[name = string("op_1591_cast_fp16")];
            tensor<int32, [1]> var_1593_axes_0 = const()[name = string("op_1593_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 4, 1]> var_1593_cast_fp16 = reduce_max(axes = var_1593_axes_0, keep_dims = var_128, x = var_1591_cast_fp16)[name = string("op_1593_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> s_49_cast_fp16 = maximum(x = var_1593_cast_fp16, y = var_129_to_fp16)[name = string("s_49_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> y_57_cast_fp16 = real_div(x = x_87_cast_fp16, y = s_49_cast_fp16)[name = string("y_57_cast_fp16")];
            fp16 var_130_promoted_22_to_fp16 = const()[name = string("op_130_promoted_22_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 4, 96]> var_1596_cast_fp16 = pow(x = y_57_cast_fp16, y = var_130_promoted_22_to_fp16)[name = string("op_1596_cast_fp16")];
            tensor<int32, [1]> var_1598_axes_0 = const()[name = string("op_1598_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1598_keep_dims_0 = const()[name = string("op_1598_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 4, 1]> var_1598_cast_fp16 = reduce_mean(axes = var_1598_axes_0, keep_dims = var_1598_keep_dims_0, x = var_1596_cast_fp16)[name = string("op_1598_cast_fp16")];
            fp32 var_1599_epsilon_0 = const()[name = string("op_1599_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 4, 1]> var_1599_cast_fp16 = inverse(epsilon = var_1599_epsilon_0, x = s_49_cast_fp16)[name = string("op_1599_cast_fp16")];
            fp16 var_1600_to_fp16 = const()[name = string("op_1600_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 4, 1]> var_1601_cast_fp16 = mul(x = var_1599_cast_fp16, y = var_1600_to_fp16)[name = string("op_1601_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_1602_cast_fp16 = real_div(x = var_1601_cast_fp16, y = s_49_cast_fp16)[name = string("op_1602_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_1603_cast_fp16 = add(x = var_1598_cast_fp16, y = var_1602_cast_fp16)[name = string("op_1603_cast_fp16")];
            fp32 var_1604_epsilon_0 = const()[name = string("op_1604_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 4, 1]> var_1604_cast_fp16 = rsqrt(epsilon = var_1604_epsilon_0, x = var_1603_cast_fp16)[name = string("op_1604_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_1605_cast_fp16 = mul(x = y_57_cast_fp16, y = var_1604_cast_fp16)[name = string("op_1605_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_8_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109883136)))];
            tensor<fp16, [1, 12, 4, 96]> var_1606_cast_fp16 = mul(x = core_backbone_layers_8_self_attn_k_norm_weight_to_fp16, y = var_1605_cast_fp16)[name = string("op_1606_cast_fp16")];
            tensor<int32, [4]> new_21_perm_0 = const()[name = string("new_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109883392)))];
            tensor<fp16, [1, 12, 384]> linear_46_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_v_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<int32, [4]> concat_19 = const()[name = string("concat_19"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> var_1611_cast_fp16 = reshape(shape = concat_19, x = linear_46_cast_fp16)[name = string("op_1611_cast_fp16")];
            tensor<int32, [4]> new_23_perm_0 = const()[name = string("new_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_121_begin_0 = const()[name = string("input_121_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_121_end_0 = const()[name = string("input_121_end_0"), val = tensor<int32, [5]>([3, 1, 4, 31, 96])];
            tensor<bool, [5]> input_121_end_mask_0 = const()[name = string("input_121_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_121_squeeze_mask_0 = const()[name = string("input_121_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_121_cast_fp16 = slice_by_index(begin = input_121_begin_0, end = input_121_end_0, end_mask = input_121_end_mask_0, squeeze_mask = input_121_squeeze_mask_0, x = k)[name = string("input_121_cast_fp16")];
            tensor<int32, [8]> var_1616_pad_0 = const()[name = string("op_1616_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1616_mode_0 = const()[name = string("op_1616_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_21_cast_fp16 = transpose(perm = new_21_perm_0, x = var_1606_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [1, 4, 43, 96]> var_1616_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = var_1616_mode_0, pad = var_1616_pad_0, x = new_21_cast_fp16)[name = string("op_1616_cast_fp16")];
            tensor<int32, [8]> var_1618_pad_0 = const()[name = string("op_1618_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_1618_mode_0 = const()[name = string("op_1618_mode_0"), val = string("constant")];
            fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_1618_cast_fp16 = pad(constant_val = const_29_to_fp16, mode = var_1618_mode_0, pad = var_1618_pad_0, x = input_121_cast_fp16)[name = string("op_1618_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> x_91_cast_fp16 = add(x = var_1616_cast_fp16, y = var_1618_cast_fp16)[name = string("x_91_cast_fp16")];
            tensor<int32, [5]> input_125_begin_0 = const()[name = string("input_125_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_125_end_0 = const()[name = string("input_125_end_0"), val = tensor<int32, [5]>([3, 1, 4, 31, 96])];
            tensor<bool, [5]> input_125_end_mask_0 = const()[name = string("input_125_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_125_squeeze_mask_0 = const()[name = string("input_125_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_125_cast_fp16 = slice_by_index(begin = input_125_begin_0, end = input_125_end_0, end_mask = input_125_end_mask_0, squeeze_mask = input_125_squeeze_mask_0, x = v)[name = string("input_125_cast_fp16")];
            tensor<int32, [8]> var_1623_pad_0 = const()[name = string("op_1623_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1623_mode_0 = const()[name = string("op_1623_mode_0"), val = string("constant")];
            fp16 const_30_to_fp16 = const()[name = string("const_30_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_23_cast_fp16 = transpose(perm = new_23_perm_0, x = var_1611_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [1, 4, 43, 96]> var_1623_cast_fp16 = pad(constant_val = const_30_to_fp16, mode = var_1623_mode_0, pad = var_1623_pad_0, x = new_23_cast_fp16)[name = string("op_1623_cast_fp16")];
            tensor<int32, [8]> var_1625_pad_0 = const()[name = string("op_1625_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_1625_mode_0 = const()[name = string("op_1625_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_1625_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = var_1625_mode_0, pad = var_1625_pad_0, x = input_125_cast_fp16)[name = string("op_1625_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> v_11_cast_fp16 = add(x = var_1623_cast_fp16, y = var_1625_cast_fp16)[name = string("v_11_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1627_cast_fp16 = mul(x = var_1584_cast_fp16, y = const_66_to_fp16)[name = string("op_1627_cast_fp16")];
            tensor<int32, [2]> var_1628_split_sizes_0 = const()[name = string("op_1628_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_75 = const()[name = string("const_75"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 48]> var_1628_cast_fp16_0, tensor<fp16, [1, 12, 8, 48]> var_1628_cast_fp16_1 = split(axis = const_75, split_sizes = var_1628_split_sizes_0, x = var_1584_cast_fp16)[name = string("op_1628_cast_fp16")];
            fp16 const_32_promoted_to_fp16 = const()[name = string("const_32_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 12, 8, 48]> var_1630_cast_fp16 = mul(x = var_1628_cast_fp16_1, y = const_32_promoted_to_fp16)[name = string("op_1630_cast_fp16")];
            bool var_1632_interleave_0 = const()[name = string("op_1632_interleave_0"), val = bool(false)];
            int32 const_76 = const()[name = string("const_76"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 96]> var_1632_cast_fp16 = concat(axis = const_76, interleave = var_1632_interleave_0, values = (var_1630_cast_fp16, var_1628_cast_fp16_0))[name = string("op_1632_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1633_cast_fp16 = mul(x = var_1632_cast_fp16, y = const_69_to_fp16)[name = string("op_1633_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> q_5_cast_fp16 = add(x = var_1627_cast_fp16, y = var_1633_cast_fp16)[name = string("q_5_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_1635_cast_fp16 = mul(x = x_91_cast_fp16, y = const_0_to_fp16)[name = string("op_1635_cast_fp16")];
            tensor<int32, [2]> var_1636_split_sizes_0 = const()[name = string("op_1636_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1636_axis_0 = const()[name = string("op_1636_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 43, 48]> var_1636_cast_fp16_0, tensor<fp16, [1, 4, 43, 48]> var_1636_cast_fp16_1 = split(axis = var_1636_axis_0, split_sizes = var_1636_split_sizes_0, x = x_91_cast_fp16)[name = string("op_1636_cast_fp16")];
            fp16 const_33_promoted_to_fp16 = const()[name = string("const_33_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 43, 48]> var_1638_cast_fp16 = mul(x = var_1636_cast_fp16_1, y = const_33_promoted_to_fp16)[name = string("op_1638_cast_fp16")];
            bool var_1640_interleave_0 = const()[name = string("op_1640_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 43, 96]> var_1640_cast_fp16 = concat(axis = var_127, interleave = var_1640_interleave_0, values = (var_1638_cast_fp16, var_1636_cast_fp16_0))[name = string("op_1640_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_1641_cast_fp16 = mul(x = var_1640_cast_fp16, y = const_1_to_fp16)[name = string("op_1641_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> k_11_cast_fp16 = add(x = var_1635_cast_fp16, y = var_1641_cast_fp16)[name = string("k_11_cast_fp16")];
            tensor<int32, [1]> var_1643_axes_0 = const()[name = string("op_1643_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_1643_cast_fp16 = expand_dims(axes = var_1643_axes_0, x = k_11_cast_fp16)[name = string("op_1643_cast_fp16")];
            tensor<int32, [5]> var_1645_reps_0 = const()[name = string("op_1645_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_1645_cast_fp16 = tile(reps = var_1645_reps_0, x = var_1643_cast_fp16)[name = string("op_1645_cast_fp16")];
            tensor<int32, [4]> concat_21 = const()[name = string("concat_21"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> k_13_cast_fp16 = reshape(shape = concat_21, x = var_1645_cast_fp16)[name = string("k_13_cast_fp16")];
            tensor<int32, [1]> var_1647_axes_0 = const()[name = string("op_1647_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_1647_cast_fp16 = expand_dims(axes = var_1647_axes_0, x = v_11_cast_fp16)[name = string("op_1647_cast_fp16")];
            tensor<int32, [5]> var_1649_reps_0 = const()[name = string("op_1649_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_1649_cast_fp16 = tile(reps = var_1649_reps_0, x = var_1647_cast_fp16)[name = string("op_1649_cast_fp16")];
            tensor<int32, [4]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> v_13_cast_fp16 = reshape(shape = concat_23, x = var_1649_cast_fp16)[name = string("v_13_cast_fp16")];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = string("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1652_transpose_x_1 = const()[name = string("op_1652_transpose_x_1"), val = bool(false)];
            bool var_1652_transpose_y_1 = const()[name = string("op_1652_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 96]> transpose_4_cast_fp16 = transpose(perm = transpose_4_perm_0, x = q_5_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [1, 8, 12, 43]> var_1652_cast_fp16 = matmul(transpose_x = var_1652_transpose_x_1, transpose_y = var_1652_transpose_y_1, x = transpose_4_cast_fp16, y = k_13_cast_fp16)[name = string("op_1652_cast_fp16")];
            fp16 var_1653_to_fp16 = const()[name = string("op_1653_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 12, 43]> s_51_cast_fp16 = mul(x = var_1652_cast_fp16, y = var_1653_to_fp16)[name = string("s_51_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_93_cast_fp16 = add(x = s_51_cast_fp16, y = mask)[name = string("x_93_cast_fp16")];
            tensor<int32, [1]> var_1657_axes_0 = const()[name = string("op_1657_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 12, 1]> var_1657_cast_fp16 = reduce_max(axes = var_1657_axes_0, keep_dims = var_128, x = x_93_cast_fp16)[name = string("op_1657_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_95_cast_fp16 = sub(x = x_93_cast_fp16, y = var_1657_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_97_cast_fp16 = exp(x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<int32, [1]> var_1661_axes_0 = const()[name = string("op_1661_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1661_keep_dims_0 = const()[name = string("op_1661_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 1]> var_1661_cast_fp16 = reduce_sum(axes = var_1661_axes_0, keep_dims = var_1661_keep_dims_0, x = x_97_cast_fp16)[name = string("op_1661_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> var_1662_cast_fp16 = real_div(x = x_97_cast_fp16, y = var_1661_cast_fp16)[name = string("op_1662_cast_fp16")];
            bool o_9_transpose_x_0 = const()[name = string("o_9_transpose_x_0"), val = bool(false)];
            bool o_9_transpose_y_0 = const()[name = string("o_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 12, 96]> o_9_cast_fp16 = matmul(transpose_x = o_9_transpose_x_0, transpose_y = o_9_transpose_y_0, x = var_1662_cast_fp16, y = v_13_cast_fp16)[name = string("o_9_cast_fp16")];
            tensor<int32, [4]> o_11_perm_0 = const()[name = string("o_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_24 = const()[name = string("concat_24"), val = tensor<int32, [3]>([1, 12, 768])];
            tensor<fp16, [1, 12, 8, 96]> o_11_cast_fp16 = transpose(perm = o_11_perm_0, x = o_9_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [1, 12, 768]> input_127_cast_fp16 = reshape(shape = concat_24, x = o_11_cast_fp16)[name = string("input_127_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110473280)))];
            tensor<fp16, [1, 12, 768]> linear_47_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_o_proj_weight_to_fp16, x = input_127_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<int32, [4]> var_1671_begin_0 = const()[name = string("op_1671_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_1671_end_0 = const()[name = string("op_1671_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_1671_end_mask_0 = const()[name = string("op_1671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1671_cast_fp16 = slice_by_index(begin = var_1671_begin_0, end = var_1671_end_0, end_mask = var_1671_end_mask_0, x = x_91_cast_fp16)[name = string("op_1671_cast_fp16")];
            tensor<int32, [4]> var_1674_begin_0 = const()[name = string("op_1674_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_1674_end_0 = const()[name = string("op_1674_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_1674_end_mask_0 = const()[name = string("op_1674_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1674_cast_fp16 = slice_by_index(begin = var_1674_begin_0, end = var_1674_end_0, end_mask = var_1674_end_mask_0, x = v_11_cast_fp16)[name = string("op_1674_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_99_cast_fp16 = add(x = x_83_cast_fp16, y = linear_47_cast_fp16)[name = string("x_99_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1677_cast_fp16 = abs(x = x_99_cast_fp16)[name = string("op_1677_cast_fp16")];
            tensor<int32, [1]> var_1679_axes_0 = const()[name = string("op_1679_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1679_cast_fp16 = reduce_max(axes = var_1679_axes_0, keep_dims = var_128, x = var_1677_cast_fp16)[name = string("op_1679_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_53_cast_fp16 = maximum(x = var_1679_cast_fp16, y = var_129_to_fp16)[name = string("s_53_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_59_cast_fp16 = real_div(x = x_99_cast_fp16, y = s_53_cast_fp16)[name = string("y_59_cast_fp16")];
            fp16 var_130_promoted_23_to_fp16 = const()[name = string("op_130_promoted_23_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1682_cast_fp16 = pow(x = y_59_cast_fp16, y = var_130_promoted_23_to_fp16)[name = string("op_1682_cast_fp16")];
            tensor<int32, [1]> var_1684_axes_0 = const()[name = string("op_1684_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1684_keep_dims_0 = const()[name = string("op_1684_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1684_cast_fp16 = reduce_mean(axes = var_1684_axes_0, keep_dims = var_1684_keep_dims_0, x = var_1682_cast_fp16)[name = string("op_1684_cast_fp16")];
            fp32 var_1685_epsilon_0 = const()[name = string("op_1685_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1685_cast_fp16 = inverse(epsilon = var_1685_epsilon_0, x = s_53_cast_fp16)[name = string("op_1685_cast_fp16")];
            fp16 var_1686_to_fp16 = const()[name = string("op_1686_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1687_cast_fp16 = mul(x = var_1685_cast_fp16, y = var_1686_to_fp16)[name = string("op_1687_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1688_cast_fp16 = real_div(x = var_1687_cast_fp16, y = s_53_cast_fp16)[name = string("op_1688_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1689_cast_fp16 = add(x = var_1684_cast_fp16, y = var_1688_cast_fp16)[name = string("op_1689_cast_fp16")];
            fp32 var_1690_epsilon_0 = const()[name = string("op_1690_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1690_cast_fp16 = rsqrt(epsilon = var_1690_epsilon_0, x = var_1689_cast_fp16)[name = string("op_1690_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1691_cast_fp16 = mul(x = y_59_cast_fp16, y = var_1690_cast_fp16)[name = string("op_1691_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_8_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111652992)))];
            tensor<fp16, [1, 12, 768]> input_129_cast_fp16 = mul(x = core_backbone_layers_8_ffn_norm_weight_to_fp16, y = var_1691_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_8_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111654592)))];
            tensor<fp16, [1, 12, 1792]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w1_weight_to_fp16, x = input_129_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_8_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114407168)))];
            tensor<fp16, [1, 12, 1792]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w3_weight_to_fp16, x = input_129_cast_fp16)[name = string("linear_49_cast_fp16")];
            fp16 var_1700_to_fp16 = const()[name = string("op_1700_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1701_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_1700_to_fp16)[name = string("op_1701_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1704_cast_fp16 = tanh(x = var_1701_cast_fp16)[name = string("op_1704_cast_fp16")];
            fp16 var_1705_to_fp16 = const()[name = string("op_1705_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1706_cast_fp16 = add(x = var_1704_cast_fp16, y = var_1705_to_fp16)[name = string("op_1706_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1707_cast_fp16 = mul(x = var_1701_cast_fp16, y = var_1706_cast_fp16)[name = string("op_1707_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_131_cast_fp16 = mul(x = var_1707_cast_fp16, y = linear_49_cast_fp16)[name = string("input_131_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_8_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117159744)))];
            tensor<fp16, [1, 12, 768]> linear_50_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w2_weight_to_fp16, x = input_131_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_103_cast_fp16 = add(x = x_99_cast_fp16, y = linear_50_cast_fp16)[name = string("x_103_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1713_cast_fp16 = abs(x = x_103_cast_fp16)[name = string("op_1713_cast_fp16")];
            tensor<int32, [1]> var_1715_axes_0 = const()[name = string("op_1715_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1715_cast_fp16 = reduce_max(axes = var_1715_axes_0, keep_dims = var_128, x = var_1713_cast_fp16)[name = string("op_1715_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_55_cast_fp16 = maximum(x = var_1715_cast_fp16, y = var_129_to_fp16)[name = string("s_55_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_61_cast_fp16 = real_div(x = x_103_cast_fp16, y = s_55_cast_fp16)[name = string("y_61_cast_fp16")];
            fp16 var_130_promoted_24_to_fp16 = const()[name = string("op_130_promoted_24_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1718_cast_fp16 = pow(x = y_61_cast_fp16, y = var_130_promoted_24_to_fp16)[name = string("op_1718_cast_fp16")];
            tensor<int32, [1]> var_1720_axes_0 = const()[name = string("op_1720_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1720_keep_dims_0 = const()[name = string("op_1720_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1720_cast_fp16 = reduce_mean(axes = var_1720_axes_0, keep_dims = var_1720_keep_dims_0, x = var_1718_cast_fp16)[name = string("op_1720_cast_fp16")];
            fp32 var_1721_epsilon_0 = const()[name = string("op_1721_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1721_cast_fp16 = inverse(epsilon = var_1721_epsilon_0, x = s_55_cast_fp16)[name = string("op_1721_cast_fp16")];
            fp16 var_1722_to_fp16 = const()[name = string("op_1722_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1723_cast_fp16 = mul(x = var_1721_cast_fp16, y = var_1722_to_fp16)[name = string("op_1723_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1724_cast_fp16 = real_div(x = var_1723_cast_fp16, y = s_55_cast_fp16)[name = string("op_1724_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1725_cast_fp16 = add(x = var_1720_cast_fp16, y = var_1724_cast_fp16)[name = string("op_1725_cast_fp16")];
            fp32 var_1726_epsilon_0 = const()[name = string("op_1726_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1726_cast_fp16 = rsqrt(epsilon = var_1726_epsilon_0, x = var_1725_cast_fp16)[name = string("op_1726_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1727_cast_fp16 = mul(x = y_61_cast_fp16, y = var_1726_cast_fp16)[name = string("op_1727_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_9_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_9_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119912320)))];
            tensor<fp16, [1, 12, 768]> input_133_cast_fp16 = mul(x = core_backbone_layers_9_operator_norm_weight_to_fp16, y = var_1727_cast_fp16)[name = string("input_133_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_9_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119913920)))];
            tensor<fp16, [1, 12, 2304]> linear_51_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_9_conv_in_proj_weight_to_fp16, x = input_133_cast_fp16)[name = string("linear_51_cast_fp16")];
            tensor<int32, [3]> bcx_13_perm_0 = const()[name = string("bcx_13_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_31_begin_0 = const()[name = string("b_31_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_31_end_0 = const()[name = string("b_31_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_31_end_mask_0 = const()[name = string("b_31_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_13_cast_fp16 = transpose(perm = bcx_13_perm_0, x = linear_51_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 768, 12]> b_31_cast_fp16 = slice_by_index(begin = b_31_begin_0, end = b_31_end_0, end_mask = b_31_end_mask_0, x = bcx_13_cast_fp16)[name = string("b_31_cast_fp16")];
            tensor<int32, [3]> c_13_begin_0 = const()[name = string("c_13_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_13_end_0 = const()[name = string("c_13_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_13_end_mask_0 = const()[name = string("c_13_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_13_cast_fp16 = slice_by_index(begin = c_13_begin_0, end = c_13_end_0, end_mask = c_13_end_mask_0, x = bcx_13_cast_fp16)[name = string("c_13_cast_fp16")];
            tensor<int32, [3]> xx_13_begin_0 = const()[name = string("xx_13_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_13_end_0 = const()[name = string("xx_13_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_13_end_mask_0 = const()[name = string("xx_13_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_13_cast_fp16 = slice_by_index(begin = xx_13_begin_0, end = xx_13_end_0, end_mask = xx_13_end_mask_0, x = bcx_13_cast_fp16)[name = string("xx_13_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_25_cast_fp16 = mul(x = b_31_cast_fp16, y = xx_13_cast_fp16)[name = string("new_25_cast_fp16")];
            tensor<int32, [4]> input_137_begin_0 = const()[name = string("input_137_begin_0"), val = tensor<int32, [4]>([6, 0, 0, 0])];
            tensor<int32, [4]> input_137_end_0 = const()[name = string("input_137_end_0"), val = tensor<int32, [4]>([7, 1, 768, 2])];
            tensor<bool, [4]> input_137_end_mask_0 = const()[name = string("input_137_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_137_squeeze_mask_0 = const()[name = string("input_137_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_137_cast_fp16 = slice_by_index(begin = input_137_begin_0, end = input_137_end_0, end_mask = input_137_end_mask_0, squeeze_mask = input_137_squeeze_mask_0, x = conv)[name = string("input_137_cast_fp16")];
            tensor<int32, [6]> var_1742_pad_0 = const()[name = string("op_1742_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1742_mode_0 = const()[name = string("op_1742_mode_0"), val = string("constant")];
            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1742_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = var_1742_mode_0, pad = var_1742_pad_0, x = new_25_cast_fp16)[name = string("op_1742_cast_fp16")];
            tensor<int32, [6]> var_1744_pad_0 = const()[name = string("op_1744_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_1744_mode_0 = const()[name = string("op_1744_mode_0"), val = string("constant")];
            fp16 const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1744_cast_fp16 = pad(constant_val = const_35_to_fp16, mode = var_1744_mode_0, pad = var_1744_pad_0, x = input_137_cast_fp16)[name = string("op_1744_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_25_cast_fp16 = add(x = var_1742_cast_fp16, y = var_1744_cast_fp16)[name = string("win_25_cast_fp16")];
            tensor<int32, [3]> var_1746_begin_0 = const()[name = string("op_1746_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_1746_end_0 = const()[name = string("op_1746_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_1746_end_mask_0 = const()[name = string("op_1746_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1746_cast_fp16 = slice_by_index(begin = var_1746_begin_0, end = var_1746_end_0, end_mask = var_1746_end_mask_0, x = win_25_cast_fp16)[name = string("op_1746_cast_fp16")];
            string var_1752_pad_type_0 = const()[name = string("op_1752_pad_type_0"), val = string("valid")];
            int32 var_1752_groups_0 = const()[name = string("op_1752_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1752_strides_0 = const()[name = string("op_1752_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1752_pad_0 = const()[name = string("op_1752_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1752_dilations_0 = const()[name = string("op_1752_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_9_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123452928)))];
            tensor<fp16, [1, 768, 12]> var_1752_cast_fp16 = conv(dilations = var_1752_dilations_0, groups = var_1752_groups_0, pad = var_1752_pad_0, pad_type = var_1752_pad_type_0, strides = var_1752_strides_0, weight = core_backbone_layers_9_conv_conv_weight_to_fp16, x = win_25_cast_fp16)[name = string("op_1752_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_1754_cast_fp16 = mul(x = c_13_cast_fp16, y = var_1752_cast_fp16)[name = string("op_1754_cast_fp16")];
            tensor<int32, [3]> input_139_perm_0 = const()[name = string("input_139_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_9_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123457600)))];
            tensor<fp16, [1, 12, 768]> input_139_cast_fp16 = transpose(perm = input_139_perm_0, x = var_1754_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 12, 768]> linear_52_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_9_conv_out_proj_weight_to_fp16, x = input_139_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_105_cast_fp16 = add(x = x_103_cast_fp16, y = linear_52_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1760_cast_fp16 = abs(x = x_105_cast_fp16)[name = string("op_1760_cast_fp16")];
            tensor<int32, [1]> var_1762_axes_0 = const()[name = string("op_1762_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1762_cast_fp16 = reduce_max(axes = var_1762_axes_0, keep_dims = var_128, x = var_1760_cast_fp16)[name = string("op_1762_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_57_cast_fp16 = maximum(x = var_1762_cast_fp16, y = var_129_to_fp16)[name = string("s_57_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_65_cast_fp16 = real_div(x = x_105_cast_fp16, y = s_57_cast_fp16)[name = string("y_65_cast_fp16")];
            fp16 var_130_promoted_25_to_fp16 = const()[name = string("op_130_promoted_25_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1765_cast_fp16 = pow(x = y_65_cast_fp16, y = var_130_promoted_25_to_fp16)[name = string("op_1765_cast_fp16")];
            tensor<int32, [1]> var_1767_axes_0 = const()[name = string("op_1767_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1767_keep_dims_0 = const()[name = string("op_1767_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1767_cast_fp16 = reduce_mean(axes = var_1767_axes_0, keep_dims = var_1767_keep_dims_0, x = var_1765_cast_fp16)[name = string("op_1767_cast_fp16")];
            fp32 var_1768_epsilon_0 = const()[name = string("op_1768_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1768_cast_fp16 = inverse(epsilon = var_1768_epsilon_0, x = s_57_cast_fp16)[name = string("op_1768_cast_fp16")];
            fp16 var_1769_to_fp16 = const()[name = string("op_1769_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1770_cast_fp16 = mul(x = var_1768_cast_fp16, y = var_1769_to_fp16)[name = string("op_1770_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1771_cast_fp16 = real_div(x = var_1770_cast_fp16, y = s_57_cast_fp16)[name = string("op_1771_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1772_cast_fp16 = add(x = var_1767_cast_fp16, y = var_1771_cast_fp16)[name = string("op_1772_cast_fp16")];
            fp32 var_1773_epsilon_0 = const()[name = string("op_1773_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1773_cast_fp16 = rsqrt(epsilon = var_1773_epsilon_0, x = var_1772_cast_fp16)[name = string("op_1773_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1774_cast_fp16 = mul(x = y_65_cast_fp16, y = var_1773_cast_fp16)[name = string("op_1774_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_9_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_9_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124637312)))];
            tensor<fp16, [1, 12, 768]> input_141_cast_fp16 = mul(x = core_backbone_layers_9_ffn_norm_weight_to_fp16, y = var_1774_cast_fp16)[name = string("input_141_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_9_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124638912)))];
            tensor<fp16, [1, 12, 1792]> linear_53_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w1_weight_to_fp16, x = input_141_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_9_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127391488)))];
            tensor<fp16, [1, 12, 1792]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w3_weight_to_fp16, x = input_141_cast_fp16)[name = string("linear_54_cast_fp16")];
            fp16 var_1783_to_fp16 = const()[name = string("op_1783_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1784_cast_fp16 = mul(x = linear_53_cast_fp16, y = var_1783_to_fp16)[name = string("op_1784_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1787_cast_fp16 = tanh(x = var_1784_cast_fp16)[name = string("op_1787_cast_fp16")];
            fp16 var_1788_to_fp16 = const()[name = string("op_1788_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1789_cast_fp16 = add(x = var_1787_cast_fp16, y = var_1788_to_fp16)[name = string("op_1789_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1790_cast_fp16 = mul(x = var_1784_cast_fp16, y = var_1789_cast_fp16)[name = string("op_1790_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_143_cast_fp16 = mul(x = var_1790_cast_fp16, y = linear_54_cast_fp16)[name = string("input_143_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_9_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130144064)))];
            tensor<fp16, [1, 12, 768]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w2_weight_to_fp16, x = input_143_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_109_cast_fp16 = add(x = x_105_cast_fp16, y = linear_55_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1796_cast_fp16 = abs(x = x_109_cast_fp16)[name = string("op_1796_cast_fp16")];
            tensor<int32, [1]> var_1798_axes_0 = const()[name = string("op_1798_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1798_cast_fp16 = reduce_max(axes = var_1798_axes_0, keep_dims = var_128, x = var_1796_cast_fp16)[name = string("op_1798_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_59_cast_fp16 = maximum(x = var_1798_cast_fp16, y = var_129_to_fp16)[name = string("s_59_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_67_cast_fp16 = real_div(x = x_109_cast_fp16, y = s_59_cast_fp16)[name = string("y_67_cast_fp16")];
            fp16 var_130_promoted_26_to_fp16 = const()[name = string("op_130_promoted_26_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1801_cast_fp16 = pow(x = y_67_cast_fp16, y = var_130_promoted_26_to_fp16)[name = string("op_1801_cast_fp16")];
            tensor<int32, [1]> var_1803_axes_0 = const()[name = string("op_1803_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1803_keep_dims_0 = const()[name = string("op_1803_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1803_cast_fp16 = reduce_mean(axes = var_1803_axes_0, keep_dims = var_1803_keep_dims_0, x = var_1801_cast_fp16)[name = string("op_1803_cast_fp16")];
            fp32 var_1804_epsilon_0 = const()[name = string("op_1804_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1804_cast_fp16 = inverse(epsilon = var_1804_epsilon_0, x = s_59_cast_fp16)[name = string("op_1804_cast_fp16")];
            fp16 var_1805_to_fp16 = const()[name = string("op_1805_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1806_cast_fp16 = mul(x = var_1804_cast_fp16, y = var_1805_to_fp16)[name = string("op_1806_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1807_cast_fp16 = real_div(x = var_1806_cast_fp16, y = s_59_cast_fp16)[name = string("op_1807_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1808_cast_fp16 = add(x = var_1803_cast_fp16, y = var_1807_cast_fp16)[name = string("op_1808_cast_fp16")];
            fp32 var_1809_epsilon_0 = const()[name = string("op_1809_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1809_cast_fp16 = rsqrt(epsilon = var_1809_epsilon_0, x = var_1808_cast_fp16)[name = string("op_1809_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1810_cast_fp16 = mul(x = y_67_cast_fp16, y = var_1809_cast_fp16)[name = string("op_1810_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_10_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_10_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132896640)))];
            tensor<fp16, [1, 12, 768]> input_145_cast_fp16 = mul(x = core_backbone_layers_10_operator_norm_weight_to_fp16, y = var_1810_cast_fp16)[name = string("input_145_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_10_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132898240)))];
            tensor<fp16, [1, 12, 2304]> linear_56_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_10_conv_in_proj_weight_to_fp16, x = input_145_cast_fp16)[name = string("linear_56_cast_fp16")];
            tensor<int32, [3]> bcx_15_perm_0 = const()[name = string("bcx_15_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_35_begin_0 = const()[name = string("b_35_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_35_end_0 = const()[name = string("b_35_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_35_end_mask_0 = const()[name = string("b_35_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_15_cast_fp16 = transpose(perm = bcx_15_perm_0, x = linear_56_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [1, 768, 12]> b_35_cast_fp16 = slice_by_index(begin = b_35_begin_0, end = b_35_end_0, end_mask = b_35_end_mask_0, x = bcx_15_cast_fp16)[name = string("b_35_cast_fp16")];
            tensor<int32, [3]> c_15_begin_0 = const()[name = string("c_15_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_15_end_0 = const()[name = string("c_15_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_15_end_mask_0 = const()[name = string("c_15_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_15_cast_fp16 = slice_by_index(begin = c_15_begin_0, end = c_15_end_0, end_mask = c_15_end_mask_0, x = bcx_15_cast_fp16)[name = string("c_15_cast_fp16")];
            tensor<int32, [3]> xx_15_begin_0 = const()[name = string("xx_15_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_15_end_0 = const()[name = string("xx_15_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_15_end_mask_0 = const()[name = string("xx_15_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_15_cast_fp16 = slice_by_index(begin = xx_15_begin_0, end = xx_15_end_0, end_mask = xx_15_end_mask_0, x = bcx_15_cast_fp16)[name = string("xx_15_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_27_cast_fp16 = mul(x = b_35_cast_fp16, y = xx_15_cast_fp16)[name = string("new_27_cast_fp16")];
            tensor<int32, [4]> input_149_begin_0 = const()[name = string("input_149_begin_0"), val = tensor<int32, [4]>([7, 0, 0, 0])];
            tensor<int32, [4]> input_149_end_0 = const()[name = string("input_149_end_0"), val = tensor<int32, [4]>([8, 1, 768, 2])];
            tensor<bool, [4]> input_149_end_mask_0 = const()[name = string("input_149_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_149_squeeze_mask_0 = const()[name = string("input_149_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_149_cast_fp16 = slice_by_index(begin = input_149_begin_0, end = input_149_end_0, end_mask = input_149_end_mask_0, squeeze_mask = input_149_squeeze_mask_0, x = conv)[name = string("input_149_cast_fp16")];
            tensor<int32, [6]> var_1825_pad_0 = const()[name = string("op_1825_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1825_mode_0 = const()[name = string("op_1825_mode_0"), val = string("constant")];
            fp16 const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1825_cast_fp16 = pad(constant_val = const_36_to_fp16, mode = var_1825_mode_0, pad = var_1825_pad_0, x = new_27_cast_fp16)[name = string("op_1825_cast_fp16")];
            tensor<int32, [6]> var_1827_pad_0 = const()[name = string("op_1827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_1827_mode_0 = const()[name = string("op_1827_mode_0"), val = string("constant")];
            fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_1827_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = var_1827_mode_0, pad = var_1827_pad_0, x = input_149_cast_fp16)[name = string("op_1827_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_29_cast_fp16 = add(x = var_1825_cast_fp16, y = var_1827_cast_fp16)[name = string("win_29_cast_fp16")];
            tensor<int32, [3]> var_1829_begin_0 = const()[name = string("op_1829_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_1829_end_0 = const()[name = string("op_1829_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_1829_end_mask_0 = const()[name = string("op_1829_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1829_cast_fp16 = slice_by_index(begin = var_1829_begin_0, end = var_1829_end_0, end_mask = var_1829_end_mask_0, x = win_29_cast_fp16)[name = string("op_1829_cast_fp16")];
            string var_1835_pad_type_0 = const()[name = string("op_1835_pad_type_0"), val = string("valid")];
            int32 var_1835_groups_0 = const()[name = string("op_1835_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1835_strides_0 = const()[name = string("op_1835_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1835_pad_0 = const()[name = string("op_1835_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1835_dilations_0 = const()[name = string("op_1835_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_10_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136437248)))];
            tensor<fp16, [1, 768, 12]> var_1835_cast_fp16 = conv(dilations = var_1835_dilations_0, groups = var_1835_groups_0, pad = var_1835_pad_0, pad_type = var_1835_pad_type_0, strides = var_1835_strides_0, weight = core_backbone_layers_10_conv_conv_weight_to_fp16, x = win_29_cast_fp16)[name = string("op_1835_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_1837_cast_fp16 = mul(x = c_15_cast_fp16, y = var_1835_cast_fp16)[name = string("op_1837_cast_fp16")];
            tensor<int32, [3]> input_151_perm_0 = const()[name = string("input_151_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_10_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136441920)))];
            tensor<fp16, [1, 12, 768]> input_151_cast_fp16 = transpose(perm = input_151_perm_0, x = var_1837_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 12, 768]> linear_57_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_10_conv_out_proj_weight_to_fp16, x = input_151_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_111_cast_fp16 = add(x = x_109_cast_fp16, y = linear_57_cast_fp16)[name = string("x_111_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1843_cast_fp16 = abs(x = x_111_cast_fp16)[name = string("op_1843_cast_fp16")];
            tensor<int32, [1]> var_1845_axes_0 = const()[name = string("op_1845_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1845_cast_fp16 = reduce_max(axes = var_1845_axes_0, keep_dims = var_128, x = var_1843_cast_fp16)[name = string("op_1845_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_61_cast_fp16 = maximum(x = var_1845_cast_fp16, y = var_129_to_fp16)[name = string("s_61_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_71_cast_fp16 = real_div(x = x_111_cast_fp16, y = s_61_cast_fp16)[name = string("y_71_cast_fp16")];
            fp16 var_130_promoted_27_to_fp16 = const()[name = string("op_130_promoted_27_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1848_cast_fp16 = pow(x = y_71_cast_fp16, y = var_130_promoted_27_to_fp16)[name = string("op_1848_cast_fp16")];
            tensor<int32, [1]> var_1850_axes_0 = const()[name = string("op_1850_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1850_keep_dims_0 = const()[name = string("op_1850_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1850_cast_fp16 = reduce_mean(axes = var_1850_axes_0, keep_dims = var_1850_keep_dims_0, x = var_1848_cast_fp16)[name = string("op_1850_cast_fp16")];
            fp32 var_1851_epsilon_0 = const()[name = string("op_1851_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1851_cast_fp16 = inverse(epsilon = var_1851_epsilon_0, x = s_61_cast_fp16)[name = string("op_1851_cast_fp16")];
            fp16 var_1852_to_fp16 = const()[name = string("op_1852_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1853_cast_fp16 = mul(x = var_1851_cast_fp16, y = var_1852_to_fp16)[name = string("op_1853_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1854_cast_fp16 = real_div(x = var_1853_cast_fp16, y = s_61_cast_fp16)[name = string("op_1854_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1855_cast_fp16 = add(x = var_1850_cast_fp16, y = var_1854_cast_fp16)[name = string("op_1855_cast_fp16")];
            fp32 var_1856_epsilon_0 = const()[name = string("op_1856_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1856_cast_fp16 = rsqrt(epsilon = var_1856_epsilon_0, x = var_1855_cast_fp16)[name = string("op_1856_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1857_cast_fp16 = mul(x = y_71_cast_fp16, y = var_1856_cast_fp16)[name = string("op_1857_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_10_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_10_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137621632)))];
            tensor<fp16, [1, 12, 768]> input_153_cast_fp16 = mul(x = core_backbone_layers_10_ffn_norm_weight_to_fp16, y = var_1857_cast_fp16)[name = string("input_153_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_10_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137623232)))];
            tensor<fp16, [1, 12, 1792]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w1_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_58_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_10_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140375808)))];
            tensor<fp16, [1, 12, 1792]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w3_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_59_cast_fp16")];
            fp16 var_1866_to_fp16 = const()[name = string("op_1866_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_1867_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_1866_to_fp16)[name = string("op_1867_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1870_cast_fp16 = tanh(x = var_1867_cast_fp16)[name = string("op_1870_cast_fp16")];
            fp16 var_1871_to_fp16 = const()[name = string("op_1871_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_1872_cast_fp16 = add(x = var_1870_cast_fp16, y = var_1871_to_fp16)[name = string("op_1872_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_1873_cast_fp16 = mul(x = var_1867_cast_fp16, y = var_1872_cast_fp16)[name = string("op_1873_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_155_cast_fp16 = mul(x = var_1873_cast_fp16, y = linear_59_cast_fp16)[name = string("input_155_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_10_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143128384)))];
            tensor<fp16, [1, 12, 768]> linear_60_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w2_weight_to_fp16, x = input_155_cast_fp16)[name = string("linear_60_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_115_cast_fp16 = add(x = x_111_cast_fp16, y = linear_60_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1879_cast_fp16 = abs(x = x_115_cast_fp16)[name = string("op_1879_cast_fp16")];
            tensor<int32, [1]> var_1881_axes_0 = const()[name = string("op_1881_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_1881_cast_fp16 = reduce_max(axes = var_1881_axes_0, keep_dims = var_128, x = var_1879_cast_fp16)[name = string("op_1881_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_63_cast_fp16 = maximum(x = var_1881_cast_fp16, y = var_129_to_fp16)[name = string("s_63_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_73_cast_fp16 = real_div(x = x_115_cast_fp16, y = s_63_cast_fp16)[name = string("y_73_cast_fp16")];
            fp16 var_130_promoted_28_to_fp16 = const()[name = string("op_130_promoted_28_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_1884_cast_fp16 = pow(x = y_73_cast_fp16, y = var_130_promoted_28_to_fp16)[name = string("op_1884_cast_fp16")];
            tensor<int32, [1]> var_1886_axes_0 = const()[name = string("op_1886_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1886_keep_dims_0 = const()[name = string("op_1886_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_1886_cast_fp16 = reduce_mean(axes = var_1886_axes_0, keep_dims = var_1886_keep_dims_0, x = var_1884_cast_fp16)[name = string("op_1886_cast_fp16")];
            fp32 var_1887_epsilon_0 = const()[name = string("op_1887_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_1887_cast_fp16 = inverse(epsilon = var_1887_epsilon_0, x = s_63_cast_fp16)[name = string("op_1887_cast_fp16")];
            fp16 var_1888_to_fp16 = const()[name = string("op_1888_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_1889_cast_fp16 = mul(x = var_1887_cast_fp16, y = var_1888_to_fp16)[name = string("op_1889_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1890_cast_fp16 = real_div(x = var_1889_cast_fp16, y = s_63_cast_fp16)[name = string("op_1890_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_1891_cast_fp16 = add(x = var_1886_cast_fp16, y = var_1890_cast_fp16)[name = string("op_1891_cast_fp16")];
            fp32 var_1892_epsilon_0 = const()[name = string("op_1892_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_1892_cast_fp16 = rsqrt(epsilon = var_1892_epsilon_0, x = var_1891_cast_fp16)[name = string("op_1892_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_1893_cast_fp16 = mul(x = y_73_cast_fp16, y = var_1892_cast_fp16)[name = string("op_1893_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_11_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145880960)))];
            tensor<fp16, [1, 12, 768]> input_157_cast_fp16 = mul(x = core_backbone_layers_11_operator_norm_weight_to_fp16, y = var_1893_cast_fp16)[name = string("input_157_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145882560)))];
            tensor<fp16, [1, 12, 768]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_q_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<int32, [4]> concat_25 = const()[name = string("concat_25"), val = tensor<int32, [4]>([1, 12, 8, 96])];
            tensor<fp16, [1, 12, 8, 96]> x_117_cast_fp16 = reshape(shape = concat_25, x = linear_61_cast_fp16)[name = string("x_117_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1900_cast_fp16 = abs(x = x_117_cast_fp16)[name = string("op_1900_cast_fp16")];
            tensor<int32, [1]> var_1902_axes_0 = const()[name = string("op_1902_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 8, 1]> var_1902_cast_fp16 = reduce_max(axes = var_1902_axes_0, keep_dims = var_128, x = var_1900_cast_fp16)[name = string("op_1902_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> s_65_cast_fp16 = maximum(x = var_1902_cast_fp16, y = var_129_to_fp16)[name = string("s_65_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> y_75_cast_fp16 = real_div(x = x_117_cast_fp16, y = s_65_cast_fp16)[name = string("y_75_cast_fp16")];
            fp16 var_130_promoted_29_to_fp16 = const()[name = string("op_130_promoted_29_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 8, 96]> var_1905_cast_fp16 = pow(x = y_75_cast_fp16, y = var_130_promoted_29_to_fp16)[name = string("op_1905_cast_fp16")];
            tensor<int32, [1]> var_1907_axes_0 = const()[name = string("op_1907_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1907_keep_dims_0 = const()[name = string("op_1907_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 8, 1]> var_1907_cast_fp16 = reduce_mean(axes = var_1907_axes_0, keep_dims = var_1907_keep_dims_0, x = var_1905_cast_fp16)[name = string("op_1907_cast_fp16")];
            fp32 var_1908_epsilon_0 = const()[name = string("op_1908_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 8, 1]> var_1908_cast_fp16 = inverse(epsilon = var_1908_epsilon_0, x = s_65_cast_fp16)[name = string("op_1908_cast_fp16")];
            fp16 var_1909_to_fp16 = const()[name = string("op_1909_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 8, 1]> var_1910_cast_fp16 = mul(x = var_1908_cast_fp16, y = var_1909_to_fp16)[name = string("op_1910_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_1911_cast_fp16 = real_div(x = var_1910_cast_fp16, y = s_65_cast_fp16)[name = string("op_1911_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_1912_cast_fp16 = add(x = var_1907_cast_fp16, y = var_1911_cast_fp16)[name = string("op_1912_cast_fp16")];
            fp32 var_1913_epsilon_0 = const()[name = string("op_1913_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 8, 1]> var_1913_cast_fp16 = rsqrt(epsilon = var_1913_epsilon_0, x = var_1912_cast_fp16)[name = string("op_1913_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1914_cast_fp16 = mul(x = y_75_cast_fp16, y = var_1913_cast_fp16)[name = string("op_1914_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_11_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147062272)))];
            tensor<fp16, [1, 12, 8, 96]> var_1915_cast_fp16 = mul(x = core_backbone_layers_11_self_attn_q_norm_weight_to_fp16, y = var_1914_cast_fp16)[name = string("op_1915_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147062528)))];
            tensor<fp16, [1, 12, 384]> linear_62_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_k_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_62_cast_fp16")];
            tensor<int32, [4]> concat_26 = const()[name = string("concat_26"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> x_119_cast_fp16 = reshape(shape = concat_26, x = linear_62_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_1922_cast_fp16 = abs(x = x_119_cast_fp16)[name = string("op_1922_cast_fp16")];
            tensor<int32, [1]> var_1924_axes_0 = const()[name = string("op_1924_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 4, 1]> var_1924_cast_fp16 = reduce_max(axes = var_1924_axes_0, keep_dims = var_128, x = var_1922_cast_fp16)[name = string("op_1924_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> s_67_cast_fp16 = maximum(x = var_1924_cast_fp16, y = var_129_to_fp16)[name = string("s_67_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> y_77_cast_fp16 = real_div(x = x_119_cast_fp16, y = s_67_cast_fp16)[name = string("y_77_cast_fp16")];
            fp16 var_130_promoted_30_to_fp16 = const()[name = string("op_130_promoted_30_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 4, 96]> var_1927_cast_fp16 = pow(x = y_77_cast_fp16, y = var_130_promoted_30_to_fp16)[name = string("op_1927_cast_fp16")];
            tensor<int32, [1]> var_1929_axes_0 = const()[name = string("op_1929_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1929_keep_dims_0 = const()[name = string("op_1929_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 4, 1]> var_1929_cast_fp16 = reduce_mean(axes = var_1929_axes_0, keep_dims = var_1929_keep_dims_0, x = var_1927_cast_fp16)[name = string("op_1929_cast_fp16")];
            fp32 var_1930_epsilon_0 = const()[name = string("op_1930_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 4, 1]> var_1930_cast_fp16 = inverse(epsilon = var_1930_epsilon_0, x = s_67_cast_fp16)[name = string("op_1930_cast_fp16")];
            fp16 var_1931_to_fp16 = const()[name = string("op_1931_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 4, 1]> var_1932_cast_fp16 = mul(x = var_1930_cast_fp16, y = var_1931_to_fp16)[name = string("op_1932_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_1933_cast_fp16 = real_div(x = var_1932_cast_fp16, y = s_67_cast_fp16)[name = string("op_1933_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_1934_cast_fp16 = add(x = var_1929_cast_fp16, y = var_1933_cast_fp16)[name = string("op_1934_cast_fp16")];
            fp32 var_1935_epsilon_0 = const()[name = string("op_1935_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 4, 1]> var_1935_cast_fp16 = rsqrt(epsilon = var_1935_epsilon_0, x = var_1934_cast_fp16)[name = string("op_1935_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_1936_cast_fp16 = mul(x = y_77_cast_fp16, y = var_1935_cast_fp16)[name = string("op_1936_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_11_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147652416)))];
            tensor<fp16, [1, 12, 4, 96]> var_1937_cast_fp16 = mul(x = core_backbone_layers_11_self_attn_k_norm_weight_to_fp16, y = var_1936_cast_fp16)[name = string("op_1937_cast_fp16")];
            tensor<int32, [4]> new_29_perm_0 = const()[name = string("new_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147652672)))];
            tensor<fp16, [1, 12, 384]> linear_63_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_v_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_63_cast_fp16")];
            tensor<int32, [4]> concat_27 = const()[name = string("concat_27"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> var_1942_cast_fp16 = reshape(shape = concat_27, x = linear_63_cast_fp16)[name = string("op_1942_cast_fp16")];
            tensor<int32, [4]> new_31_perm_0 = const()[name = string("new_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_161_begin_0 = const()[name = string("input_161_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])];
            tensor<int32, [5]> input_161_end_0 = const()[name = string("input_161_end_0"), val = tensor<int32, [5]>([4, 1, 4, 31, 96])];
            tensor<bool, [5]> input_161_end_mask_0 = const()[name = string("input_161_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_161_squeeze_mask_0 = const()[name = string("input_161_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_161_cast_fp16 = slice_by_index(begin = input_161_begin_0, end = input_161_end_0, end_mask = input_161_end_mask_0, squeeze_mask = input_161_squeeze_mask_0, x = k)[name = string("input_161_cast_fp16")];
            tensor<int32, [8]> var_1947_pad_0 = const()[name = string("op_1947_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1947_mode_0 = const()[name = string("op_1947_mode_0"), val = string("constant")];
            fp16 const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_29_cast_fp16 = transpose(perm = new_29_perm_0, x = var_1937_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [1, 4, 43, 96]> var_1947_cast_fp16 = pad(constant_val = const_38_to_fp16, mode = var_1947_mode_0, pad = var_1947_pad_0, x = new_29_cast_fp16)[name = string("op_1947_cast_fp16")];
            tensor<int32, [8]> var_1949_pad_0 = const()[name = string("op_1949_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_1949_mode_0 = const()[name = string("op_1949_mode_0"), val = string("constant")];
            fp16 const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_1949_cast_fp16 = pad(constant_val = const_39_to_fp16, mode = var_1949_mode_0, pad = var_1949_pad_0, x = input_161_cast_fp16)[name = string("op_1949_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> x_123_cast_fp16 = add(x = var_1947_cast_fp16, y = var_1949_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<int32, [5]> input_165_begin_0 = const()[name = string("input_165_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])];
            tensor<int32, [5]> input_165_end_0 = const()[name = string("input_165_end_0"), val = tensor<int32, [5]>([4, 1, 4, 31, 96])];
            tensor<bool, [5]> input_165_end_mask_0 = const()[name = string("input_165_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_165_squeeze_mask_0 = const()[name = string("input_165_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_165_cast_fp16 = slice_by_index(begin = input_165_begin_0, end = input_165_end_0, end_mask = input_165_end_mask_0, squeeze_mask = input_165_squeeze_mask_0, x = v)[name = string("input_165_cast_fp16")];
            tensor<int32, [8]> var_1954_pad_0 = const()[name = string("op_1954_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1954_mode_0 = const()[name = string("op_1954_mode_0"), val = string("constant")];
            fp16 const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_31_cast_fp16 = transpose(perm = new_31_perm_0, x = var_1942_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [1, 4, 43, 96]> var_1954_cast_fp16 = pad(constant_val = const_40_to_fp16, mode = var_1954_mode_0, pad = var_1954_pad_0, x = new_31_cast_fp16)[name = string("op_1954_cast_fp16")];
            tensor<int32, [8]> var_1956_pad_0 = const()[name = string("op_1956_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_1956_mode_0 = const()[name = string("op_1956_mode_0"), val = string("constant")];
            fp16 const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_1956_cast_fp16 = pad(constant_val = const_41_to_fp16, mode = var_1956_mode_0, pad = var_1956_pad_0, x = input_165_cast_fp16)[name = string("op_1956_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> v_15_cast_fp16 = add(x = var_1954_cast_fp16, y = var_1956_cast_fp16)[name = string("v_15_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1958_cast_fp16 = mul(x = var_1915_cast_fp16, y = const_66_to_fp16)[name = string("op_1958_cast_fp16")];
            tensor<int32, [2]> var_1959_split_sizes_0 = const()[name = string("op_1959_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_79 = const()[name = string("const_79"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 48]> var_1959_cast_fp16_0, tensor<fp16, [1, 12, 8, 48]> var_1959_cast_fp16_1 = split(axis = const_79, split_sizes = var_1959_split_sizes_0, x = var_1915_cast_fp16)[name = string("op_1959_cast_fp16")];
            fp16 const_42_promoted_to_fp16 = const()[name = string("const_42_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 12, 8, 48]> var_1961_cast_fp16 = mul(x = var_1959_cast_fp16_1, y = const_42_promoted_to_fp16)[name = string("op_1961_cast_fp16")];
            bool var_1963_interleave_0 = const()[name = string("op_1963_interleave_0"), val = bool(false)];
            int32 const_80 = const()[name = string("const_80"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 96]> var_1963_cast_fp16 = concat(axis = const_80, interleave = var_1963_interleave_0, values = (var_1961_cast_fp16, var_1959_cast_fp16_0))[name = string("op_1963_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_1964_cast_fp16 = mul(x = var_1963_cast_fp16, y = const_69_to_fp16)[name = string("op_1964_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> q_7_cast_fp16 = add(x = var_1958_cast_fp16, y = var_1964_cast_fp16)[name = string("q_7_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_1966_cast_fp16 = mul(x = x_123_cast_fp16, y = const_0_to_fp16)[name = string("op_1966_cast_fp16")];
            tensor<int32, [2]> var_1967_split_sizes_0 = const()[name = string("op_1967_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1967_axis_0 = const()[name = string("op_1967_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 43, 48]> var_1967_cast_fp16_0, tensor<fp16, [1, 4, 43, 48]> var_1967_cast_fp16_1 = split(axis = var_1967_axis_0, split_sizes = var_1967_split_sizes_0, x = x_123_cast_fp16)[name = string("op_1967_cast_fp16")];
            fp16 const_43_promoted_to_fp16 = const()[name = string("const_43_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 43, 48]> var_1969_cast_fp16 = mul(x = var_1967_cast_fp16_1, y = const_43_promoted_to_fp16)[name = string("op_1969_cast_fp16")];
            bool var_1971_interleave_0 = const()[name = string("op_1971_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 43, 96]> var_1971_cast_fp16 = concat(axis = var_127, interleave = var_1971_interleave_0, values = (var_1969_cast_fp16, var_1967_cast_fp16_0))[name = string("op_1971_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_1972_cast_fp16 = mul(x = var_1971_cast_fp16, y = const_1_to_fp16)[name = string("op_1972_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> k_15_cast_fp16 = add(x = var_1966_cast_fp16, y = var_1972_cast_fp16)[name = string("k_15_cast_fp16")];
            tensor<int32, [1]> var_1974_axes_0 = const()[name = string("op_1974_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_1974_cast_fp16 = expand_dims(axes = var_1974_axes_0, x = k_15_cast_fp16)[name = string("op_1974_cast_fp16")];
            tensor<int32, [5]> var_1976_reps_0 = const()[name = string("op_1976_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_1976_cast_fp16 = tile(reps = var_1976_reps_0, x = var_1974_cast_fp16)[name = string("op_1976_cast_fp16")];
            tensor<int32, [4]> concat_29 = const()[name = string("concat_29"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> k_17_cast_fp16 = reshape(shape = concat_29, x = var_1976_cast_fp16)[name = string("k_17_cast_fp16")];
            tensor<int32, [1]> var_1978_axes_0 = const()[name = string("op_1978_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_1978_cast_fp16 = expand_dims(axes = var_1978_axes_0, x = v_15_cast_fp16)[name = string("op_1978_cast_fp16")];
            tensor<int32, [5]> var_1980_reps_0 = const()[name = string("op_1980_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_1980_cast_fp16 = tile(reps = var_1980_reps_0, x = var_1978_cast_fp16)[name = string("op_1980_cast_fp16")];
            tensor<int32, [4]> concat_31 = const()[name = string("concat_31"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> v_17_cast_fp16 = reshape(shape = concat_31, x = var_1980_cast_fp16)[name = string("v_17_cast_fp16")];
            tensor<int32, [4]> transpose_6_perm_0 = const()[name = string("transpose_6_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1983_transpose_x_1 = const()[name = string("op_1983_transpose_x_1"), val = bool(false)];
            bool var_1983_transpose_y_1 = const()[name = string("op_1983_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 96]> transpose_6_cast_fp16 = transpose(perm = transpose_6_perm_0, x = q_7_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 8, 12, 43]> var_1983_cast_fp16 = matmul(transpose_x = var_1983_transpose_x_1, transpose_y = var_1983_transpose_y_1, x = transpose_6_cast_fp16, y = k_17_cast_fp16)[name = string("op_1983_cast_fp16")];
            fp16 var_1984_to_fp16 = const()[name = string("op_1984_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 12, 43]> s_69_cast_fp16 = mul(x = var_1983_cast_fp16, y = var_1984_to_fp16)[name = string("s_69_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_125_cast_fp16 = add(x = s_69_cast_fp16, y = mask)[name = string("x_125_cast_fp16")];
            tensor<int32, [1]> var_1988_axes_0 = const()[name = string("op_1988_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 12, 1]> var_1988_cast_fp16 = reduce_max(axes = var_1988_axes_0, keep_dims = var_128, x = x_125_cast_fp16)[name = string("op_1988_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_127_cast_fp16 = sub(x = x_125_cast_fp16, y = var_1988_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_129_cast_fp16 = exp(x = x_127_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<int32, [1]> var_1992_axes_0 = const()[name = string("op_1992_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1992_keep_dims_0 = const()[name = string("op_1992_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 1]> var_1992_cast_fp16 = reduce_sum(axes = var_1992_axes_0, keep_dims = var_1992_keep_dims_0, x = x_129_cast_fp16)[name = string("op_1992_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> var_1993_cast_fp16 = real_div(x = x_129_cast_fp16, y = var_1992_cast_fp16)[name = string("op_1993_cast_fp16")];
            bool o_13_transpose_x_0 = const()[name = string("o_13_transpose_x_0"), val = bool(false)];
            bool o_13_transpose_y_0 = const()[name = string("o_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 12, 96]> o_13_cast_fp16 = matmul(transpose_x = o_13_transpose_x_0, transpose_y = o_13_transpose_y_0, x = var_1993_cast_fp16, y = v_17_cast_fp16)[name = string("o_13_cast_fp16")];
            tensor<int32, [4]> o_15_perm_0 = const()[name = string("o_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_32 = const()[name = string("concat_32"), val = tensor<int32, [3]>([1, 12, 768])];
            tensor<fp16, [1, 12, 8, 96]> o_15_cast_fp16 = transpose(perm = o_15_perm_0, x = o_13_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 12, 768]> input_167_cast_fp16 = reshape(shape = concat_32, x = o_15_cast_fp16)[name = string("input_167_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148242560)))];
            tensor<fp16, [1, 12, 768]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_o_proj_weight_to_fp16, x = input_167_cast_fp16)[name = string("linear_64_cast_fp16")];
            tensor<int32, [4]> var_2002_begin_0 = const()[name = string("op_2002_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_2002_end_0 = const()[name = string("op_2002_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_2002_end_mask_0 = const()[name = string("op_2002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2002_cast_fp16 = slice_by_index(begin = var_2002_begin_0, end = var_2002_end_0, end_mask = var_2002_end_mask_0, x = x_123_cast_fp16)[name = string("op_2002_cast_fp16")];
            tensor<int32, [4]> var_2005_begin_0 = const()[name = string("op_2005_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_2005_end_0 = const()[name = string("op_2005_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_2005_end_mask_0 = const()[name = string("op_2005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2005_cast_fp16 = slice_by_index(begin = var_2005_begin_0, end = var_2005_end_0, end_mask = var_2005_end_mask_0, x = v_15_cast_fp16)[name = string("op_2005_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_131_cast_fp16 = add(x = x_115_cast_fp16, y = linear_64_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2008_cast_fp16 = abs(x = x_131_cast_fp16)[name = string("op_2008_cast_fp16")];
            tensor<int32, [1]> var_2010_axes_0 = const()[name = string("op_2010_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2010_cast_fp16 = reduce_max(axes = var_2010_axes_0, keep_dims = var_128, x = var_2008_cast_fp16)[name = string("op_2010_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_71_cast_fp16 = maximum(x = var_2010_cast_fp16, y = var_129_to_fp16)[name = string("s_71_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_79_cast_fp16 = real_div(x = x_131_cast_fp16, y = s_71_cast_fp16)[name = string("y_79_cast_fp16")];
            fp16 var_130_promoted_31_to_fp16 = const()[name = string("op_130_promoted_31_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2013_cast_fp16 = pow(x = y_79_cast_fp16, y = var_130_promoted_31_to_fp16)[name = string("op_2013_cast_fp16")];
            tensor<int32, [1]> var_2015_axes_0 = const()[name = string("op_2015_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2015_keep_dims_0 = const()[name = string("op_2015_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2015_cast_fp16 = reduce_mean(axes = var_2015_axes_0, keep_dims = var_2015_keep_dims_0, x = var_2013_cast_fp16)[name = string("op_2015_cast_fp16")];
            fp32 var_2016_epsilon_0 = const()[name = string("op_2016_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2016_cast_fp16 = inverse(epsilon = var_2016_epsilon_0, x = s_71_cast_fp16)[name = string("op_2016_cast_fp16")];
            fp16 var_2017_to_fp16 = const()[name = string("op_2017_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2018_cast_fp16 = mul(x = var_2016_cast_fp16, y = var_2017_to_fp16)[name = string("op_2018_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2019_cast_fp16 = real_div(x = var_2018_cast_fp16, y = s_71_cast_fp16)[name = string("op_2019_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2020_cast_fp16 = add(x = var_2015_cast_fp16, y = var_2019_cast_fp16)[name = string("op_2020_cast_fp16")];
            fp32 var_2021_epsilon_0 = const()[name = string("op_2021_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2021_cast_fp16 = rsqrt(epsilon = var_2021_epsilon_0, x = var_2020_cast_fp16)[name = string("op_2021_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2022_cast_fp16 = mul(x = y_79_cast_fp16, y = var_2021_cast_fp16)[name = string("op_2022_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_11_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149422272)))];
            tensor<fp16, [1, 12, 768]> input_169_cast_fp16 = mul(x = core_backbone_layers_11_ffn_norm_weight_to_fp16, y = var_2022_cast_fp16)[name = string("input_169_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_11_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149423872)))];
            tensor<fp16, [1, 12, 1792]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w1_weight_to_fp16, x = input_169_cast_fp16)[name = string("linear_65_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_11_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152176448)))];
            tensor<fp16, [1, 12, 1792]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w3_weight_to_fp16, x = input_169_cast_fp16)[name = string("linear_66_cast_fp16")];
            fp16 var_2031_to_fp16 = const()[name = string("op_2031_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_2032_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_2031_to_fp16)[name = string("op_2032_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2035_cast_fp16 = tanh(x = var_2032_cast_fp16)[name = string("op_2035_cast_fp16")];
            fp16 var_2036_to_fp16 = const()[name = string("op_2036_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_2037_cast_fp16 = add(x = var_2035_cast_fp16, y = var_2036_to_fp16)[name = string("op_2037_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2038_cast_fp16 = mul(x = var_2032_cast_fp16, y = var_2037_cast_fp16)[name = string("op_2038_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_171_cast_fp16 = mul(x = var_2038_cast_fp16, y = linear_66_cast_fp16)[name = string("input_171_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_11_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154929024)))];
            tensor<fp16, [1, 12, 768]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w2_weight_to_fp16, x = input_171_cast_fp16)[name = string("linear_67_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_135_cast_fp16 = add(x = x_131_cast_fp16, y = linear_67_cast_fp16)[name = string("x_135_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2044_cast_fp16 = abs(x = x_135_cast_fp16)[name = string("op_2044_cast_fp16")];
            tensor<int32, [1]> var_2046_axes_0 = const()[name = string("op_2046_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2046_cast_fp16 = reduce_max(axes = var_2046_axes_0, keep_dims = var_128, x = var_2044_cast_fp16)[name = string("op_2046_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_73_cast_fp16 = maximum(x = var_2046_cast_fp16, y = var_129_to_fp16)[name = string("s_73_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_81_cast_fp16 = real_div(x = x_135_cast_fp16, y = s_73_cast_fp16)[name = string("y_81_cast_fp16")];
            fp16 var_130_promoted_32_to_fp16 = const()[name = string("op_130_promoted_32_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2049_cast_fp16 = pow(x = y_81_cast_fp16, y = var_130_promoted_32_to_fp16)[name = string("op_2049_cast_fp16")];
            tensor<int32, [1]> var_2051_axes_0 = const()[name = string("op_2051_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2051_keep_dims_0 = const()[name = string("op_2051_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2051_cast_fp16 = reduce_mean(axes = var_2051_axes_0, keep_dims = var_2051_keep_dims_0, x = var_2049_cast_fp16)[name = string("op_2051_cast_fp16")];
            fp32 var_2052_epsilon_0 = const()[name = string("op_2052_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2052_cast_fp16 = inverse(epsilon = var_2052_epsilon_0, x = s_73_cast_fp16)[name = string("op_2052_cast_fp16")];
            fp16 var_2053_to_fp16 = const()[name = string("op_2053_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2054_cast_fp16 = mul(x = var_2052_cast_fp16, y = var_2053_to_fp16)[name = string("op_2054_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2055_cast_fp16 = real_div(x = var_2054_cast_fp16, y = s_73_cast_fp16)[name = string("op_2055_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2056_cast_fp16 = add(x = var_2051_cast_fp16, y = var_2055_cast_fp16)[name = string("op_2056_cast_fp16")];
            fp32 var_2057_epsilon_0 = const()[name = string("op_2057_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2057_cast_fp16 = rsqrt(epsilon = var_2057_epsilon_0, x = var_2056_cast_fp16)[name = string("op_2057_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2058_cast_fp16 = mul(x = y_81_cast_fp16, y = var_2057_cast_fp16)[name = string("op_2058_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_12_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_12_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157681600)))];
            tensor<fp16, [1, 12, 768]> input_173_cast_fp16 = mul(x = core_backbone_layers_12_operator_norm_weight_to_fp16, y = var_2058_cast_fp16)[name = string("input_173_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_12_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157683200)))];
            tensor<fp16, [1, 12, 2304]> linear_68_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_12_conv_in_proj_weight_to_fp16, x = input_173_cast_fp16)[name = string("linear_68_cast_fp16")];
            tensor<int32, [3]> bcx_17_perm_0 = const()[name = string("bcx_17_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_41_begin_0 = const()[name = string("b_41_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_41_end_0 = const()[name = string("b_41_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_41_end_mask_0 = const()[name = string("b_41_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_17_cast_fp16 = transpose(perm = bcx_17_perm_0, x = linear_68_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 768, 12]> b_41_cast_fp16 = slice_by_index(begin = b_41_begin_0, end = b_41_end_0, end_mask = b_41_end_mask_0, x = bcx_17_cast_fp16)[name = string("b_41_cast_fp16")];
            tensor<int32, [3]> c_17_begin_0 = const()[name = string("c_17_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_17_end_0 = const()[name = string("c_17_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_17_end_mask_0 = const()[name = string("c_17_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_17_cast_fp16 = slice_by_index(begin = c_17_begin_0, end = c_17_end_0, end_mask = c_17_end_mask_0, x = bcx_17_cast_fp16)[name = string("c_17_cast_fp16")];
            tensor<int32, [3]> xx_17_begin_0 = const()[name = string("xx_17_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_17_end_0 = const()[name = string("xx_17_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_17_end_mask_0 = const()[name = string("xx_17_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_17_cast_fp16 = slice_by_index(begin = xx_17_begin_0, end = xx_17_end_0, end_mask = xx_17_end_mask_0, x = bcx_17_cast_fp16)[name = string("xx_17_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_33_cast_fp16 = mul(x = b_41_cast_fp16, y = xx_17_cast_fp16)[name = string("new_33_cast_fp16")];
            tensor<int32, [4]> input_177_begin_0 = const()[name = string("input_177_begin_0"), val = tensor<int32, [4]>([8, 0, 0, 0])];
            tensor<int32, [4]> input_177_end_0 = const()[name = string("input_177_end_0"), val = tensor<int32, [4]>([9, 1, 768, 2])];
            tensor<bool, [4]> input_177_end_mask_0 = const()[name = string("input_177_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_177_squeeze_mask_0 = const()[name = string("input_177_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_177_cast_fp16 = slice_by_index(begin = input_177_begin_0, end = input_177_end_0, end_mask = input_177_end_mask_0, squeeze_mask = input_177_squeeze_mask_0, x = conv)[name = string("input_177_cast_fp16")];
            tensor<int32, [6]> var_2073_pad_0 = const()[name = string("op_2073_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_2073_mode_0 = const()[name = string("op_2073_mode_0"), val = string("constant")];
            fp16 const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_2073_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = var_2073_mode_0, pad = var_2073_pad_0, x = new_33_cast_fp16)[name = string("op_2073_cast_fp16")];
            tensor<int32, [6]> var_2075_pad_0 = const()[name = string("op_2075_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_2075_mode_0 = const()[name = string("op_2075_mode_0"), val = string("constant")];
            fp16 const_45_to_fp16 = const()[name = string("const_45_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_2075_cast_fp16 = pad(constant_val = const_45_to_fp16, mode = var_2075_mode_0, pad = var_2075_pad_0, x = input_177_cast_fp16)[name = string("op_2075_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_33_cast_fp16 = add(x = var_2073_cast_fp16, y = var_2075_cast_fp16)[name = string("win_33_cast_fp16")];
            tensor<int32, [3]> var_2077_begin_0 = const()[name = string("op_2077_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_2077_end_0 = const()[name = string("op_2077_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_2077_end_mask_0 = const()[name = string("op_2077_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_2077_cast_fp16 = slice_by_index(begin = var_2077_begin_0, end = var_2077_end_0, end_mask = var_2077_end_mask_0, x = win_33_cast_fp16)[name = string("op_2077_cast_fp16")];
            string var_2083_pad_type_0 = const()[name = string("op_2083_pad_type_0"), val = string("valid")];
            int32 var_2083_groups_0 = const()[name = string("op_2083_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_2083_strides_0 = const()[name = string("op_2083_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2083_pad_0 = const()[name = string("op_2083_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2083_dilations_0 = const()[name = string("op_2083_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_12_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161222208)))];
            tensor<fp16, [1, 768, 12]> var_2083_cast_fp16 = conv(dilations = var_2083_dilations_0, groups = var_2083_groups_0, pad = var_2083_pad_0, pad_type = var_2083_pad_type_0, strides = var_2083_strides_0, weight = core_backbone_layers_12_conv_conv_weight_to_fp16, x = win_33_cast_fp16)[name = string("op_2083_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_2085_cast_fp16 = mul(x = c_17_cast_fp16, y = var_2083_cast_fp16)[name = string("op_2085_cast_fp16")];
            tensor<int32, [3]> input_179_perm_0 = const()[name = string("input_179_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_12_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161226880)))];
            tensor<fp16, [1, 12, 768]> input_179_cast_fp16 = transpose(perm = input_179_perm_0, x = var_2085_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [1, 12, 768]> linear_69_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_12_conv_out_proj_weight_to_fp16, x = input_179_cast_fp16)[name = string("linear_69_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_137_cast_fp16 = add(x = x_135_cast_fp16, y = linear_69_cast_fp16)[name = string("x_137_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2091_cast_fp16 = abs(x = x_137_cast_fp16)[name = string("op_2091_cast_fp16")];
            tensor<int32, [1]> var_2093_axes_0 = const()[name = string("op_2093_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2093_cast_fp16 = reduce_max(axes = var_2093_axes_0, keep_dims = var_128, x = var_2091_cast_fp16)[name = string("op_2093_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_75_cast_fp16 = maximum(x = var_2093_cast_fp16, y = var_129_to_fp16)[name = string("s_75_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_85_cast_fp16 = real_div(x = x_137_cast_fp16, y = s_75_cast_fp16)[name = string("y_85_cast_fp16")];
            fp16 var_130_promoted_33_to_fp16 = const()[name = string("op_130_promoted_33_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2096_cast_fp16 = pow(x = y_85_cast_fp16, y = var_130_promoted_33_to_fp16)[name = string("op_2096_cast_fp16")];
            tensor<int32, [1]> var_2098_axes_0 = const()[name = string("op_2098_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2098_keep_dims_0 = const()[name = string("op_2098_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2098_cast_fp16 = reduce_mean(axes = var_2098_axes_0, keep_dims = var_2098_keep_dims_0, x = var_2096_cast_fp16)[name = string("op_2098_cast_fp16")];
            fp32 var_2099_epsilon_0 = const()[name = string("op_2099_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2099_cast_fp16 = inverse(epsilon = var_2099_epsilon_0, x = s_75_cast_fp16)[name = string("op_2099_cast_fp16")];
            fp16 var_2100_to_fp16 = const()[name = string("op_2100_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2101_cast_fp16 = mul(x = var_2099_cast_fp16, y = var_2100_to_fp16)[name = string("op_2101_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2102_cast_fp16 = real_div(x = var_2101_cast_fp16, y = s_75_cast_fp16)[name = string("op_2102_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2103_cast_fp16 = add(x = var_2098_cast_fp16, y = var_2102_cast_fp16)[name = string("op_2103_cast_fp16")];
            fp32 var_2104_epsilon_0 = const()[name = string("op_2104_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2104_cast_fp16 = rsqrt(epsilon = var_2104_epsilon_0, x = var_2103_cast_fp16)[name = string("op_2104_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2105_cast_fp16 = mul(x = y_85_cast_fp16, y = var_2104_cast_fp16)[name = string("op_2105_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_12_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_12_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162406592)))];
            tensor<fp16, [1, 12, 768]> input_181_cast_fp16 = mul(x = core_backbone_layers_12_ffn_norm_weight_to_fp16, y = var_2105_cast_fp16)[name = string("input_181_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_12_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162408192)))];
            tensor<fp16, [1, 12, 1792]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w1_weight_to_fp16, x = input_181_cast_fp16)[name = string("linear_70_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_12_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165160768)))];
            tensor<fp16, [1, 12, 1792]> linear_71_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w3_weight_to_fp16, x = input_181_cast_fp16)[name = string("linear_71_cast_fp16")];
            fp16 var_2114_to_fp16 = const()[name = string("op_2114_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_2115_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_2114_to_fp16)[name = string("op_2115_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2118_cast_fp16 = tanh(x = var_2115_cast_fp16)[name = string("op_2118_cast_fp16")];
            fp16 var_2119_to_fp16 = const()[name = string("op_2119_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_2120_cast_fp16 = add(x = var_2118_cast_fp16, y = var_2119_to_fp16)[name = string("op_2120_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2121_cast_fp16 = mul(x = var_2115_cast_fp16, y = var_2120_cast_fp16)[name = string("op_2121_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_183_cast_fp16 = mul(x = var_2121_cast_fp16, y = linear_71_cast_fp16)[name = string("input_183_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_12_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167913344)))];
            tensor<fp16, [1, 12, 768]> linear_72_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w2_weight_to_fp16, x = input_183_cast_fp16)[name = string("linear_72_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_141_cast_fp16 = add(x = x_137_cast_fp16, y = linear_72_cast_fp16)[name = string("x_141_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2127_cast_fp16 = abs(x = x_141_cast_fp16)[name = string("op_2127_cast_fp16")];
            tensor<int32, [1]> var_2129_axes_0 = const()[name = string("op_2129_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2129_cast_fp16 = reduce_max(axes = var_2129_axes_0, keep_dims = var_128, x = var_2127_cast_fp16)[name = string("op_2129_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_77_cast_fp16 = maximum(x = var_2129_cast_fp16, y = var_129_to_fp16)[name = string("s_77_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_87_cast_fp16 = real_div(x = x_141_cast_fp16, y = s_77_cast_fp16)[name = string("y_87_cast_fp16")];
            fp16 var_130_promoted_34_to_fp16 = const()[name = string("op_130_promoted_34_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2132_cast_fp16 = pow(x = y_87_cast_fp16, y = var_130_promoted_34_to_fp16)[name = string("op_2132_cast_fp16")];
            tensor<int32, [1]> var_2134_axes_0 = const()[name = string("op_2134_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2134_keep_dims_0 = const()[name = string("op_2134_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2134_cast_fp16 = reduce_mean(axes = var_2134_axes_0, keep_dims = var_2134_keep_dims_0, x = var_2132_cast_fp16)[name = string("op_2134_cast_fp16")];
            fp32 var_2135_epsilon_0 = const()[name = string("op_2135_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2135_cast_fp16 = inverse(epsilon = var_2135_epsilon_0, x = s_77_cast_fp16)[name = string("op_2135_cast_fp16")];
            fp16 var_2136_to_fp16 = const()[name = string("op_2136_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2137_cast_fp16 = mul(x = var_2135_cast_fp16, y = var_2136_to_fp16)[name = string("op_2137_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2138_cast_fp16 = real_div(x = var_2137_cast_fp16, y = s_77_cast_fp16)[name = string("op_2138_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2139_cast_fp16 = add(x = var_2134_cast_fp16, y = var_2138_cast_fp16)[name = string("op_2139_cast_fp16")];
            fp32 var_2140_epsilon_0 = const()[name = string("op_2140_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2140_cast_fp16 = rsqrt(epsilon = var_2140_epsilon_0, x = var_2139_cast_fp16)[name = string("op_2140_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2141_cast_fp16 = mul(x = y_87_cast_fp16, y = var_2140_cast_fp16)[name = string("op_2141_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_13_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_13_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170665920)))];
            tensor<fp16, [1, 12, 768]> input_185_cast_fp16 = mul(x = core_backbone_layers_13_operator_norm_weight_to_fp16, y = var_2141_cast_fp16)[name = string("input_185_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_13_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170667520)))];
            tensor<fp16, [1, 12, 2304]> linear_73_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_13_conv_in_proj_weight_to_fp16, x = input_185_cast_fp16)[name = string("linear_73_cast_fp16")];
            tensor<int32, [3]> bcx_perm_0 = const()[name = string("bcx_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_45_begin_0 = const()[name = string("b_45_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_45_end_0 = const()[name = string("b_45_end_0"), val = tensor<int32, [3]>([1, 768, 12])];
            tensor<bool, [3]> b_45_end_mask_0 = const()[name = string("b_45_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 12]> bcx_cast_fp16 = transpose(perm = bcx_perm_0, x = linear_73_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 768, 12]> b_45_cast_fp16 = slice_by_index(begin = b_45_begin_0, end = b_45_end_0, end_mask = b_45_end_mask_0, x = bcx_cast_fp16)[name = string("b_45_cast_fp16")];
            tensor<int32, [3]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [3]>([1, 1536, 12])];
            tensor<bool, [3]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 12]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = bcx_cast_fp16)[name = string("c_cast_fp16")];
            tensor<int32, [3]> xx_begin_0 = const()[name = string("xx_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_end_0 = const()[name = string("xx_end_0"), val = tensor<int32, [3]>([1, 2304, 12])];
            tensor<bool, [3]> xx_end_mask_0 = const()[name = string("xx_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 12]> xx_cast_fp16 = slice_by_index(begin = xx_begin_0, end = xx_end_0, end_mask = xx_end_mask_0, x = bcx_cast_fp16)[name = string("xx_cast_fp16")];
            tensor<fp16, [1, 768, 12]> new_35_cast_fp16 = mul(x = b_45_cast_fp16, y = xx_cast_fp16)[name = string("new_35_cast_fp16")];
            tensor<int32, [4]> input_189_begin_0 = const()[name = string("input_189_begin_0"), val = tensor<int32, [4]>([9, 0, 0, 0])];
            tensor<int32, [4]> input_189_end_0 = const()[name = string("input_189_end_0"), val = tensor<int32, [4]>([10, 1, 768, 2])];
            tensor<bool, [4]> input_189_end_mask_0 = const()[name = string("input_189_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_189_squeeze_mask_0 = const()[name = string("input_189_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_189_cast_fp16 = slice_by_index(begin = input_189_begin_0, end = input_189_end_0, end_mask = input_189_end_mask_0, squeeze_mask = input_189_squeeze_mask_0, x = conv)[name = string("input_189_cast_fp16")];
            tensor<int32, [6]> var_2156_pad_0 = const()[name = string("op_2156_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_2156_mode_0 = const()[name = string("op_2156_mode_0"), val = string("constant")];
            fp16 const_46_to_fp16 = const()[name = string("const_46_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_2156_cast_fp16 = pad(constant_val = const_46_to_fp16, mode = var_2156_mode_0, pad = var_2156_pad_0, x = new_35_cast_fp16)[name = string("op_2156_cast_fp16")];
            tensor<int32, [6]> var_2158_pad_0 = const()[name = string("op_2158_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 12])];
            string var_2158_mode_0 = const()[name = string("op_2158_mode_0"), val = string("constant")];
            fp16 const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 14]> var_2158_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = var_2158_mode_0, pad = var_2158_pad_0, x = input_189_cast_fp16)[name = string("op_2158_cast_fp16")];
            tensor<fp16, [1, 768, 14]> win_37_cast_fp16 = add(x = var_2156_cast_fp16, y = var_2158_cast_fp16)[name = string("win_37_cast_fp16")];
            tensor<int32, [3]> var_2160_begin_0 = const()[name = string("op_2160_begin_0"), val = tensor<int32, [3]>([0, 0, 12])];
            tensor<int32, [3]> var_2160_end_0 = const()[name = string("op_2160_end_0"), val = tensor<int32, [3]>([1, 768, 14])];
            tensor<bool, [3]> var_2160_end_mask_0 = const()[name = string("op_2160_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_2160_cast_fp16 = slice_by_index(begin = var_2160_begin_0, end = var_2160_end_0, end_mask = var_2160_end_mask_0, x = win_37_cast_fp16)[name = string("op_2160_cast_fp16")];
            string var_2166_pad_type_0 = const()[name = string("op_2166_pad_type_0"), val = string("valid")];
            int32 var_2166_groups_0 = const()[name = string("op_2166_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_2166_strides_0 = const()[name = string("op_2166_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2166_pad_0 = const()[name = string("op_2166_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2166_dilations_0 = const()[name = string("op_2166_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_13_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174206528)))];
            tensor<fp16, [1, 768, 12]> var_2166_cast_fp16 = conv(dilations = var_2166_dilations_0, groups = var_2166_groups_0, pad = var_2166_pad_0, pad_type = var_2166_pad_type_0, strides = var_2166_strides_0, weight = core_backbone_layers_13_conv_conv_weight_to_fp16, x = win_37_cast_fp16)[name = string("op_2166_cast_fp16")];
            tensor<fp16, [1, 768, 12]> var_2168_cast_fp16 = mul(x = c_cast_fp16, y = var_2166_cast_fp16)[name = string("op_2168_cast_fp16")];
            tensor<int32, [3]> input_191_perm_0 = const()[name = string("input_191_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_13_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174211200)))];
            tensor<fp16, [1, 12, 768]> input_191_cast_fp16 = transpose(perm = input_191_perm_0, x = var_2168_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 12, 768]> linear_74_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_13_conv_out_proj_weight_to_fp16, x = input_191_cast_fp16)[name = string("linear_74_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_143_cast_fp16 = add(x = x_141_cast_fp16, y = linear_74_cast_fp16)[name = string("x_143_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2174_cast_fp16 = abs(x = x_143_cast_fp16)[name = string("op_2174_cast_fp16")];
            tensor<int32, [1]> var_2176_axes_0 = const()[name = string("op_2176_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2176_cast_fp16 = reduce_max(axes = var_2176_axes_0, keep_dims = var_128, x = var_2174_cast_fp16)[name = string("op_2176_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_79_cast_fp16 = maximum(x = var_2176_cast_fp16, y = var_129_to_fp16)[name = string("s_79_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_91_cast_fp16 = real_div(x = x_143_cast_fp16, y = s_79_cast_fp16)[name = string("y_91_cast_fp16")];
            fp16 var_130_promoted_35_to_fp16 = const()[name = string("op_130_promoted_35_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2179_cast_fp16 = pow(x = y_91_cast_fp16, y = var_130_promoted_35_to_fp16)[name = string("op_2179_cast_fp16")];
            tensor<int32, [1]> var_2181_axes_0 = const()[name = string("op_2181_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2181_keep_dims_0 = const()[name = string("op_2181_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2181_cast_fp16 = reduce_mean(axes = var_2181_axes_0, keep_dims = var_2181_keep_dims_0, x = var_2179_cast_fp16)[name = string("op_2181_cast_fp16")];
            fp32 var_2182_epsilon_0 = const()[name = string("op_2182_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2182_cast_fp16 = inverse(epsilon = var_2182_epsilon_0, x = s_79_cast_fp16)[name = string("op_2182_cast_fp16")];
            fp16 var_2183_to_fp16 = const()[name = string("op_2183_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2184_cast_fp16 = mul(x = var_2182_cast_fp16, y = var_2183_to_fp16)[name = string("op_2184_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2185_cast_fp16 = real_div(x = var_2184_cast_fp16, y = s_79_cast_fp16)[name = string("op_2185_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2186_cast_fp16 = add(x = var_2181_cast_fp16, y = var_2185_cast_fp16)[name = string("op_2186_cast_fp16")];
            fp32 var_2187_epsilon_0 = const()[name = string("op_2187_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2187_cast_fp16 = rsqrt(epsilon = var_2187_epsilon_0, x = var_2186_cast_fp16)[name = string("op_2187_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2188_cast_fp16 = mul(x = y_91_cast_fp16, y = var_2187_cast_fp16)[name = string("op_2188_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_13_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_13_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175390912)))];
            tensor<fp16, [1, 12, 768]> input_193_cast_fp16 = mul(x = core_backbone_layers_13_ffn_norm_weight_to_fp16, y = var_2188_cast_fp16)[name = string("input_193_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_13_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175392512)))];
            tensor<fp16, [1, 12, 1792]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w1_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_75_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_13_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178145088)))];
            tensor<fp16, [1, 12, 1792]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w3_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_76_cast_fp16")];
            fp16 var_2197_to_fp16 = const()[name = string("op_2197_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_2198_cast_fp16 = mul(x = linear_75_cast_fp16, y = var_2197_to_fp16)[name = string("op_2198_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2201_cast_fp16 = tanh(x = var_2198_cast_fp16)[name = string("op_2201_cast_fp16")];
            fp16 var_2202_to_fp16 = const()[name = string("op_2202_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_2203_cast_fp16 = add(x = var_2201_cast_fp16, y = var_2202_to_fp16)[name = string("op_2203_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2204_cast_fp16 = mul(x = var_2198_cast_fp16, y = var_2203_cast_fp16)[name = string("op_2204_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_195_cast_fp16 = mul(x = var_2204_cast_fp16, y = linear_76_cast_fp16)[name = string("input_195_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_13_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180897664)))];
            tensor<fp16, [1, 12, 768]> linear_77_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w2_weight_to_fp16, x = input_195_cast_fp16)[name = string("linear_77_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_147_cast_fp16 = add(x = x_143_cast_fp16, y = linear_77_cast_fp16)[name = string("x_147_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2210_cast_fp16 = abs(x = x_147_cast_fp16)[name = string("op_2210_cast_fp16")];
            tensor<int32, [1]> var_2212_axes_0 = const()[name = string("op_2212_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2212_cast_fp16 = reduce_max(axes = var_2212_axes_0, keep_dims = var_128, x = var_2210_cast_fp16)[name = string("op_2212_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_81_cast_fp16 = maximum(x = var_2212_cast_fp16, y = var_129_to_fp16)[name = string("s_81_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_93_cast_fp16 = real_div(x = x_147_cast_fp16, y = s_81_cast_fp16)[name = string("y_93_cast_fp16")];
            fp16 var_130_promoted_36_to_fp16 = const()[name = string("op_130_promoted_36_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2215_cast_fp16 = pow(x = y_93_cast_fp16, y = var_130_promoted_36_to_fp16)[name = string("op_2215_cast_fp16")];
            tensor<int32, [1]> var_2217_axes_0 = const()[name = string("op_2217_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2217_keep_dims_0 = const()[name = string("op_2217_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2217_cast_fp16 = reduce_mean(axes = var_2217_axes_0, keep_dims = var_2217_keep_dims_0, x = var_2215_cast_fp16)[name = string("op_2217_cast_fp16")];
            fp32 var_2218_epsilon_0 = const()[name = string("op_2218_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2218_cast_fp16 = inverse(epsilon = var_2218_epsilon_0, x = s_81_cast_fp16)[name = string("op_2218_cast_fp16")];
            fp16 var_2219_to_fp16 = const()[name = string("op_2219_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2220_cast_fp16 = mul(x = var_2218_cast_fp16, y = var_2219_to_fp16)[name = string("op_2220_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2221_cast_fp16 = real_div(x = var_2220_cast_fp16, y = s_81_cast_fp16)[name = string("op_2221_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2222_cast_fp16 = add(x = var_2217_cast_fp16, y = var_2221_cast_fp16)[name = string("op_2222_cast_fp16")];
            fp32 var_2223_epsilon_0 = const()[name = string("op_2223_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2223_cast_fp16 = rsqrt(epsilon = var_2223_epsilon_0, x = var_2222_cast_fp16)[name = string("op_2223_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2224_cast_fp16 = mul(x = y_93_cast_fp16, y = var_2223_cast_fp16)[name = string("op_2224_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_14_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183650240)))];
            tensor<fp16, [1, 12, 768]> input_197_cast_fp16 = mul(x = core_backbone_layers_14_operator_norm_weight_to_fp16, y = var_2224_cast_fp16)[name = string("input_197_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183651840)))];
            tensor<fp16, [1, 12, 768]> linear_78_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_q_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_78_cast_fp16")];
            tensor<int32, [4]> concat_33 = const()[name = string("concat_33"), val = tensor<int32, [4]>([1, 12, 8, 96])];
            tensor<fp16, [1, 12, 8, 96]> x_149_cast_fp16 = reshape(shape = concat_33, x = linear_78_cast_fp16)[name = string("x_149_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2231_cast_fp16 = abs(x = x_149_cast_fp16)[name = string("op_2231_cast_fp16")];
            tensor<int32, [1]> var_2233_axes_0 = const()[name = string("op_2233_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 8, 1]> var_2233_cast_fp16 = reduce_max(axes = var_2233_axes_0, keep_dims = var_128, x = var_2231_cast_fp16)[name = string("op_2233_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> s_83_cast_fp16 = maximum(x = var_2233_cast_fp16, y = var_129_to_fp16)[name = string("s_83_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> y_95_cast_fp16 = real_div(x = x_149_cast_fp16, y = s_83_cast_fp16)[name = string("y_95_cast_fp16")];
            fp16 var_130_promoted_37_to_fp16 = const()[name = string("op_130_promoted_37_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 8, 96]> var_2236_cast_fp16 = pow(x = y_95_cast_fp16, y = var_130_promoted_37_to_fp16)[name = string("op_2236_cast_fp16")];
            tensor<int32, [1]> var_2238_axes_0 = const()[name = string("op_2238_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2238_keep_dims_0 = const()[name = string("op_2238_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 8, 1]> var_2238_cast_fp16 = reduce_mean(axes = var_2238_axes_0, keep_dims = var_2238_keep_dims_0, x = var_2236_cast_fp16)[name = string("op_2238_cast_fp16")];
            fp32 var_2239_epsilon_0 = const()[name = string("op_2239_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 8, 1]> var_2239_cast_fp16 = inverse(epsilon = var_2239_epsilon_0, x = s_83_cast_fp16)[name = string("op_2239_cast_fp16")];
            fp16 var_2240_to_fp16 = const()[name = string("op_2240_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 8, 1]> var_2241_cast_fp16 = mul(x = var_2239_cast_fp16, y = var_2240_to_fp16)[name = string("op_2241_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_2242_cast_fp16 = real_div(x = var_2241_cast_fp16, y = s_83_cast_fp16)[name = string("op_2242_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_2243_cast_fp16 = add(x = var_2238_cast_fp16, y = var_2242_cast_fp16)[name = string("op_2243_cast_fp16")];
            fp32 var_2244_epsilon_0 = const()[name = string("op_2244_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 8, 1]> var_2244_cast_fp16 = rsqrt(epsilon = var_2244_epsilon_0, x = var_2243_cast_fp16)[name = string("op_2244_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2245_cast_fp16 = mul(x = y_95_cast_fp16, y = var_2244_cast_fp16)[name = string("op_2245_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_14_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184831552)))];
            tensor<fp16, [1, 12, 8, 96]> var_2246_cast_fp16 = mul(x = core_backbone_layers_14_self_attn_q_norm_weight_to_fp16, y = var_2245_cast_fp16)[name = string("op_2246_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184831808)))];
            tensor<fp16, [1, 12, 384]> linear_79_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_k_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_79_cast_fp16")];
            tensor<int32, [4]> concat_34 = const()[name = string("concat_34"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> x_151_cast_fp16 = reshape(shape = concat_34, x = linear_79_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_2253_cast_fp16 = abs(x = x_151_cast_fp16)[name = string("op_2253_cast_fp16")];
            tensor<int32, [1]> var_2255_axes_0 = const()[name = string("op_2255_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 4, 1]> var_2255_cast_fp16 = reduce_max(axes = var_2255_axes_0, keep_dims = var_128, x = var_2253_cast_fp16)[name = string("op_2255_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> s_85_cast_fp16 = maximum(x = var_2255_cast_fp16, y = var_129_to_fp16)[name = string("s_85_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> y_97_cast_fp16 = real_div(x = x_151_cast_fp16, y = s_85_cast_fp16)[name = string("y_97_cast_fp16")];
            fp16 var_130_promoted_38_to_fp16 = const()[name = string("op_130_promoted_38_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 4, 96]> var_2258_cast_fp16 = pow(x = y_97_cast_fp16, y = var_130_promoted_38_to_fp16)[name = string("op_2258_cast_fp16")];
            tensor<int32, [1]> var_2260_axes_0 = const()[name = string("op_2260_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2260_keep_dims_0 = const()[name = string("op_2260_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 4, 1]> var_2260_cast_fp16 = reduce_mean(axes = var_2260_axes_0, keep_dims = var_2260_keep_dims_0, x = var_2258_cast_fp16)[name = string("op_2260_cast_fp16")];
            fp32 var_2261_epsilon_0 = const()[name = string("op_2261_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 4, 1]> var_2261_cast_fp16 = inverse(epsilon = var_2261_epsilon_0, x = s_85_cast_fp16)[name = string("op_2261_cast_fp16")];
            fp16 var_2262_to_fp16 = const()[name = string("op_2262_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 4, 1]> var_2263_cast_fp16 = mul(x = var_2261_cast_fp16, y = var_2262_to_fp16)[name = string("op_2263_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_2264_cast_fp16 = real_div(x = var_2263_cast_fp16, y = s_85_cast_fp16)[name = string("op_2264_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_2265_cast_fp16 = add(x = var_2260_cast_fp16, y = var_2264_cast_fp16)[name = string("op_2265_cast_fp16")];
            fp32 var_2266_epsilon_0 = const()[name = string("op_2266_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 4, 1]> var_2266_cast_fp16 = rsqrt(epsilon = var_2266_epsilon_0, x = var_2265_cast_fp16)[name = string("op_2266_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_2267_cast_fp16 = mul(x = y_97_cast_fp16, y = var_2266_cast_fp16)[name = string("op_2267_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_14_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185421696)))];
            tensor<fp16, [1, 12, 4, 96]> var_2268_cast_fp16 = mul(x = core_backbone_layers_14_self_attn_k_norm_weight_to_fp16, y = var_2267_cast_fp16)[name = string("op_2268_cast_fp16")];
            tensor<int32, [4]> new_37_perm_0 = const()[name = string("new_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185421952)))];
            tensor<fp16, [1, 12, 384]> linear_80_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_v_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_80_cast_fp16")];
            tensor<int32, [4]> concat_35 = const()[name = string("concat_35"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> var_2273_cast_fp16 = reshape(shape = concat_35, x = linear_80_cast_fp16)[name = string("op_2273_cast_fp16")];
            tensor<int32, [4]> new_39_perm_0 = const()[name = string("new_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_201_begin_0 = const()[name = string("input_201_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])];
            tensor<int32, [5]> input_201_end_0 = const()[name = string("input_201_end_0"), val = tensor<int32, [5]>([5, 1, 4, 31, 96])];
            tensor<bool, [5]> input_201_end_mask_0 = const()[name = string("input_201_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_201_squeeze_mask_0 = const()[name = string("input_201_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_201_cast_fp16 = slice_by_index(begin = input_201_begin_0, end = input_201_end_0, end_mask = input_201_end_mask_0, squeeze_mask = input_201_squeeze_mask_0, x = k)[name = string("input_201_cast_fp16")];
            tensor<int32, [8]> var_2278_pad_0 = const()[name = string("op_2278_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2278_mode_0 = const()[name = string("op_2278_mode_0"), val = string("constant")];
            fp16 const_48_to_fp16 = const()[name = string("const_48_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_37_cast_fp16 = transpose(perm = new_37_perm_0, x = var_2268_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 4, 43, 96]> var_2278_cast_fp16 = pad(constant_val = const_48_to_fp16, mode = var_2278_mode_0, pad = var_2278_pad_0, x = new_37_cast_fp16)[name = string("op_2278_cast_fp16")];
            tensor<int32, [8]> var_2280_pad_0 = const()[name = string("op_2280_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_2280_mode_0 = const()[name = string("op_2280_mode_0"), val = string("constant")];
            fp16 const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_2280_cast_fp16 = pad(constant_val = const_49_to_fp16, mode = var_2280_mode_0, pad = var_2280_pad_0, x = input_201_cast_fp16)[name = string("op_2280_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> x_155_cast_fp16 = add(x = var_2278_cast_fp16, y = var_2280_cast_fp16)[name = string("x_155_cast_fp16")];
            tensor<int32, [5]> input_205_begin_0 = const()[name = string("input_205_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])];
            tensor<int32, [5]> input_205_end_0 = const()[name = string("input_205_end_0"), val = tensor<int32, [5]>([5, 1, 4, 31, 96])];
            tensor<bool, [5]> input_205_end_mask_0 = const()[name = string("input_205_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_205_squeeze_mask_0 = const()[name = string("input_205_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_205_cast_fp16 = slice_by_index(begin = input_205_begin_0, end = input_205_end_0, end_mask = input_205_end_mask_0, squeeze_mask = input_205_squeeze_mask_0, x = v)[name = string("input_205_cast_fp16")];
            tensor<int32, [8]> var_2285_pad_0 = const()[name = string("op_2285_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2285_mode_0 = const()[name = string("op_2285_mode_0"), val = string("constant")];
            fp16 const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_39_cast_fp16 = transpose(perm = new_39_perm_0, x = var_2273_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 4, 43, 96]> var_2285_cast_fp16 = pad(constant_val = const_50_to_fp16, mode = var_2285_mode_0, pad = var_2285_pad_0, x = new_39_cast_fp16)[name = string("op_2285_cast_fp16")];
            tensor<int32, [8]> var_2287_pad_0 = const()[name = string("op_2287_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_2287_mode_0 = const()[name = string("op_2287_mode_0"), val = string("constant")];
            fp16 const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_2287_cast_fp16 = pad(constant_val = const_51_to_fp16, mode = var_2287_mode_0, pad = var_2287_pad_0, x = input_205_cast_fp16)[name = string("op_2287_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> v_19_cast_fp16 = add(x = var_2285_cast_fp16, y = var_2287_cast_fp16)[name = string("v_19_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2289_cast_fp16 = mul(x = var_2246_cast_fp16, y = const_66_to_fp16)[name = string("op_2289_cast_fp16")];
            tensor<int32, [2]> var_2290_split_sizes_0 = const()[name = string("op_2290_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_83 = const()[name = string("const_83"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 48]> var_2290_cast_fp16_0, tensor<fp16, [1, 12, 8, 48]> var_2290_cast_fp16_1 = split(axis = const_83, split_sizes = var_2290_split_sizes_0, x = var_2246_cast_fp16)[name = string("op_2290_cast_fp16")];
            fp16 const_52_promoted_to_fp16 = const()[name = string("const_52_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 12, 8, 48]> var_2292_cast_fp16 = mul(x = var_2290_cast_fp16_1, y = const_52_promoted_to_fp16)[name = string("op_2292_cast_fp16")];
            bool var_2294_interleave_0 = const()[name = string("op_2294_interleave_0"), val = bool(false)];
            int32 const_84 = const()[name = string("const_84"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 96]> var_2294_cast_fp16 = concat(axis = const_84, interleave = var_2294_interleave_0, values = (var_2292_cast_fp16, var_2290_cast_fp16_0))[name = string("op_2294_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2295_cast_fp16 = mul(x = var_2294_cast_fp16, y = const_69_to_fp16)[name = string("op_2295_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> q_9_cast_fp16 = add(x = var_2289_cast_fp16, y = var_2295_cast_fp16)[name = string("q_9_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_2297_cast_fp16 = mul(x = x_155_cast_fp16, y = const_0_to_fp16)[name = string("op_2297_cast_fp16")];
            tensor<int32, [2]> var_2298_split_sizes_0 = const()[name = string("op_2298_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2298_axis_0 = const()[name = string("op_2298_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 43, 48]> var_2298_cast_fp16_0, tensor<fp16, [1, 4, 43, 48]> var_2298_cast_fp16_1 = split(axis = var_2298_axis_0, split_sizes = var_2298_split_sizes_0, x = x_155_cast_fp16)[name = string("op_2298_cast_fp16")];
            fp16 const_53_promoted_to_fp16 = const()[name = string("const_53_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 43, 48]> var_2300_cast_fp16 = mul(x = var_2298_cast_fp16_1, y = const_53_promoted_to_fp16)[name = string("op_2300_cast_fp16")];
            bool var_2302_interleave_0 = const()[name = string("op_2302_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 43, 96]> var_2302_cast_fp16 = concat(axis = var_127, interleave = var_2302_interleave_0, values = (var_2300_cast_fp16, var_2298_cast_fp16_0))[name = string("op_2302_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_2303_cast_fp16 = mul(x = var_2302_cast_fp16, y = const_1_to_fp16)[name = string("op_2303_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> k_19_cast_fp16 = add(x = var_2297_cast_fp16, y = var_2303_cast_fp16)[name = string("k_19_cast_fp16")];
            tensor<int32, [1]> var_2305_axes_0 = const()[name = string("op_2305_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_2305_cast_fp16 = expand_dims(axes = var_2305_axes_0, x = k_19_cast_fp16)[name = string("op_2305_cast_fp16")];
            tensor<int32, [5]> var_2307_reps_0 = const()[name = string("op_2307_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_2307_cast_fp16 = tile(reps = var_2307_reps_0, x = var_2305_cast_fp16)[name = string("op_2307_cast_fp16")];
            tensor<int32, [4]> concat_37 = const()[name = string("concat_37"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> k_21_cast_fp16 = reshape(shape = concat_37, x = var_2307_cast_fp16)[name = string("k_21_cast_fp16")];
            tensor<int32, [1]> var_2309_axes_0 = const()[name = string("op_2309_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_2309_cast_fp16 = expand_dims(axes = var_2309_axes_0, x = v_19_cast_fp16)[name = string("op_2309_cast_fp16")];
            tensor<int32, [5]> var_2311_reps_0 = const()[name = string("op_2311_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_2311_cast_fp16 = tile(reps = var_2311_reps_0, x = var_2309_cast_fp16)[name = string("op_2311_cast_fp16")];
            tensor<int32, [4]> concat_39 = const()[name = string("concat_39"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> v_21_cast_fp16 = reshape(shape = concat_39, x = var_2311_cast_fp16)[name = string("v_21_cast_fp16")];
            tensor<int32, [4]> transpose_8_perm_0 = const()[name = string("transpose_8_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2314_transpose_x_1 = const()[name = string("op_2314_transpose_x_1"), val = bool(false)];
            bool var_2314_transpose_y_1 = const()[name = string("op_2314_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 96]> transpose_8_cast_fp16 = transpose(perm = transpose_8_perm_0, x = q_9_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 8, 12, 43]> var_2314_cast_fp16 = matmul(transpose_x = var_2314_transpose_x_1, transpose_y = var_2314_transpose_y_1, x = transpose_8_cast_fp16, y = k_21_cast_fp16)[name = string("op_2314_cast_fp16")];
            fp16 var_2315_to_fp16 = const()[name = string("op_2315_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 12, 43]> s_87_cast_fp16 = mul(x = var_2314_cast_fp16, y = var_2315_to_fp16)[name = string("s_87_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_157_cast_fp16 = add(x = s_87_cast_fp16, y = mask)[name = string("x_157_cast_fp16")];
            tensor<int32, [1]> var_2319_axes_0 = const()[name = string("op_2319_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 12, 1]> var_2319_cast_fp16 = reduce_max(axes = var_2319_axes_0, keep_dims = var_128, x = x_157_cast_fp16)[name = string("op_2319_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_159_cast_fp16 = sub(x = x_157_cast_fp16, y = var_2319_cast_fp16)[name = string("x_159_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_161_cast_fp16 = exp(x = x_159_cast_fp16)[name = string("x_161_cast_fp16")];
            tensor<int32, [1]> var_2323_axes_0 = const()[name = string("op_2323_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2323_keep_dims_0 = const()[name = string("op_2323_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 1]> var_2323_cast_fp16 = reduce_sum(axes = var_2323_axes_0, keep_dims = var_2323_keep_dims_0, x = x_161_cast_fp16)[name = string("op_2323_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> var_2324_cast_fp16 = real_div(x = x_161_cast_fp16, y = var_2323_cast_fp16)[name = string("op_2324_cast_fp16")];
            bool o_17_transpose_x_0 = const()[name = string("o_17_transpose_x_0"), val = bool(false)];
            bool o_17_transpose_y_0 = const()[name = string("o_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 12, 96]> o_17_cast_fp16 = matmul(transpose_x = o_17_transpose_x_0, transpose_y = o_17_transpose_y_0, x = var_2324_cast_fp16, y = v_21_cast_fp16)[name = string("o_17_cast_fp16")];
            tensor<int32, [4]> o_19_perm_0 = const()[name = string("o_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_40 = const()[name = string("concat_40"), val = tensor<int32, [3]>([1, 12, 768])];
            tensor<fp16, [1, 12, 8, 96]> o_19_cast_fp16 = transpose(perm = o_19_perm_0, x = o_17_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 12, 768]> input_207_cast_fp16 = reshape(shape = concat_40, x = o_19_cast_fp16)[name = string("input_207_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186011840)))];
            tensor<fp16, [1, 12, 768]> linear_81_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_o_proj_weight_to_fp16, x = input_207_cast_fp16)[name = string("linear_81_cast_fp16")];
            tensor<int32, [4]> var_2333_begin_0 = const()[name = string("op_2333_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_2333_end_0 = const()[name = string("op_2333_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_2333_end_mask_0 = const()[name = string("op_2333_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2333_cast_fp16 = slice_by_index(begin = var_2333_begin_0, end = var_2333_end_0, end_mask = var_2333_end_mask_0, x = x_155_cast_fp16)[name = string("op_2333_cast_fp16")];
            tensor<int32, [4]> var_2336_begin_0 = const()[name = string("op_2336_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_2336_end_0 = const()[name = string("op_2336_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_2336_end_mask_0 = const()[name = string("op_2336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2336_cast_fp16 = slice_by_index(begin = var_2336_begin_0, end = var_2336_end_0, end_mask = var_2336_end_mask_0, x = v_19_cast_fp16)[name = string("op_2336_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_163_cast_fp16 = add(x = x_147_cast_fp16, y = linear_81_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2339_cast_fp16 = abs(x = x_163_cast_fp16)[name = string("op_2339_cast_fp16")];
            tensor<int32, [1]> var_2341_axes_0 = const()[name = string("op_2341_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2341_cast_fp16 = reduce_max(axes = var_2341_axes_0, keep_dims = var_128, x = var_2339_cast_fp16)[name = string("op_2341_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_89_cast_fp16 = maximum(x = var_2341_cast_fp16, y = var_129_to_fp16)[name = string("s_89_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_99_cast_fp16 = real_div(x = x_163_cast_fp16, y = s_89_cast_fp16)[name = string("y_99_cast_fp16")];
            fp16 var_130_promoted_39_to_fp16 = const()[name = string("op_130_promoted_39_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2344_cast_fp16 = pow(x = y_99_cast_fp16, y = var_130_promoted_39_to_fp16)[name = string("op_2344_cast_fp16")];
            tensor<int32, [1]> var_2346_axes_0 = const()[name = string("op_2346_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2346_keep_dims_0 = const()[name = string("op_2346_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2346_cast_fp16 = reduce_mean(axes = var_2346_axes_0, keep_dims = var_2346_keep_dims_0, x = var_2344_cast_fp16)[name = string("op_2346_cast_fp16")];
            fp32 var_2347_epsilon_0 = const()[name = string("op_2347_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2347_cast_fp16 = inverse(epsilon = var_2347_epsilon_0, x = s_89_cast_fp16)[name = string("op_2347_cast_fp16")];
            fp16 var_2348_to_fp16 = const()[name = string("op_2348_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2349_cast_fp16 = mul(x = var_2347_cast_fp16, y = var_2348_to_fp16)[name = string("op_2349_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2350_cast_fp16 = real_div(x = var_2349_cast_fp16, y = s_89_cast_fp16)[name = string("op_2350_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2351_cast_fp16 = add(x = var_2346_cast_fp16, y = var_2350_cast_fp16)[name = string("op_2351_cast_fp16")];
            fp32 var_2352_epsilon_0 = const()[name = string("op_2352_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2352_cast_fp16 = rsqrt(epsilon = var_2352_epsilon_0, x = var_2351_cast_fp16)[name = string("op_2352_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2353_cast_fp16 = mul(x = y_99_cast_fp16, y = var_2352_cast_fp16)[name = string("op_2353_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_14_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187191552)))];
            tensor<fp16, [1, 12, 768]> input_209_cast_fp16 = mul(x = core_backbone_layers_14_ffn_norm_weight_to_fp16, y = var_2353_cast_fp16)[name = string("input_209_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_14_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187193152)))];
            tensor<fp16, [1, 12, 1792]> linear_82_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w1_weight_to_fp16, x = input_209_cast_fp16)[name = string("linear_82_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_14_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189945728)))];
            tensor<fp16, [1, 12, 1792]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w3_weight_to_fp16, x = input_209_cast_fp16)[name = string("linear_83_cast_fp16")];
            fp16 var_2362_to_fp16 = const()[name = string("op_2362_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_2363_cast_fp16 = mul(x = linear_82_cast_fp16, y = var_2362_to_fp16)[name = string("op_2363_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2366_cast_fp16 = tanh(x = var_2363_cast_fp16)[name = string("op_2366_cast_fp16")];
            fp16 var_2367_to_fp16 = const()[name = string("op_2367_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_2368_cast_fp16 = add(x = var_2366_cast_fp16, y = var_2367_to_fp16)[name = string("op_2368_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2369_cast_fp16 = mul(x = var_2363_cast_fp16, y = var_2368_cast_fp16)[name = string("op_2369_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_211_cast_fp16 = mul(x = var_2369_cast_fp16, y = linear_83_cast_fp16)[name = string("input_211_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_14_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192698304)))];
            tensor<fp16, [1, 12, 768]> linear_84_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w2_weight_to_fp16, x = input_211_cast_fp16)[name = string("linear_84_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_167_cast_fp16 = add(x = x_163_cast_fp16, y = linear_84_cast_fp16)[name = string("x_167_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2375_cast_fp16 = abs(x = x_167_cast_fp16)[name = string("op_2375_cast_fp16")];
            tensor<int32, [1]> var_2377_axes_0 = const()[name = string("op_2377_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2377_cast_fp16 = reduce_max(axes = var_2377_axes_0, keep_dims = var_128, x = var_2375_cast_fp16)[name = string("op_2377_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_91_cast_fp16 = maximum(x = var_2377_cast_fp16, y = var_129_to_fp16)[name = string("s_91_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_101_cast_fp16 = real_div(x = x_167_cast_fp16, y = s_91_cast_fp16)[name = string("y_101_cast_fp16")];
            fp16 var_130_promoted_40_to_fp16 = const()[name = string("op_130_promoted_40_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2380_cast_fp16 = pow(x = y_101_cast_fp16, y = var_130_promoted_40_to_fp16)[name = string("op_2380_cast_fp16")];
            tensor<int32, [1]> var_2382_axes_0 = const()[name = string("op_2382_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2382_keep_dims_0 = const()[name = string("op_2382_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2382_cast_fp16 = reduce_mean(axes = var_2382_axes_0, keep_dims = var_2382_keep_dims_0, x = var_2380_cast_fp16)[name = string("op_2382_cast_fp16")];
            fp32 var_2383_epsilon_0 = const()[name = string("op_2383_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2383_cast_fp16 = inverse(epsilon = var_2383_epsilon_0, x = s_91_cast_fp16)[name = string("op_2383_cast_fp16")];
            fp16 var_2384_to_fp16 = const()[name = string("op_2384_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2385_cast_fp16 = mul(x = var_2383_cast_fp16, y = var_2384_to_fp16)[name = string("op_2385_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2386_cast_fp16 = real_div(x = var_2385_cast_fp16, y = s_91_cast_fp16)[name = string("op_2386_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2387_cast_fp16 = add(x = var_2382_cast_fp16, y = var_2386_cast_fp16)[name = string("op_2387_cast_fp16")];
            fp32 var_2388_epsilon_0 = const()[name = string("op_2388_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2388_cast_fp16 = rsqrt(epsilon = var_2388_epsilon_0, x = var_2387_cast_fp16)[name = string("op_2388_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2389_cast_fp16 = mul(x = y_101_cast_fp16, y = var_2388_cast_fp16)[name = string("op_2389_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_15_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195450880)))];
            tensor<fp16, [1, 12, 768]> input_213_cast_fp16 = mul(x = core_backbone_layers_15_operator_norm_weight_to_fp16, y = var_2389_cast_fp16)[name = string("input_213_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195452480)))];
            tensor<fp16, [1, 12, 768]> linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_q_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_85_cast_fp16")];
            tensor<int32, [4]> concat_41 = const()[name = string("concat_41"), val = tensor<int32, [4]>([1, 12, 8, 96])];
            tensor<fp16, [1, 12, 8, 96]> x_169_cast_fp16 = reshape(shape = concat_41, x = linear_85_cast_fp16)[name = string("x_169_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2396_cast_fp16 = abs(x = x_169_cast_fp16)[name = string("op_2396_cast_fp16")];
            tensor<int32, [1]> var_2398_axes_0 = const()[name = string("op_2398_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 8, 1]> var_2398_cast_fp16 = reduce_max(axes = var_2398_axes_0, keep_dims = var_128, x = var_2396_cast_fp16)[name = string("op_2398_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> s_93_cast_fp16 = maximum(x = var_2398_cast_fp16, y = var_129_to_fp16)[name = string("s_93_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> y_103_cast_fp16 = real_div(x = x_169_cast_fp16, y = s_93_cast_fp16)[name = string("y_103_cast_fp16")];
            fp16 var_130_promoted_41_to_fp16 = const()[name = string("op_130_promoted_41_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 8, 96]> var_2401_cast_fp16 = pow(x = y_103_cast_fp16, y = var_130_promoted_41_to_fp16)[name = string("op_2401_cast_fp16")];
            tensor<int32, [1]> var_2403_axes_0 = const()[name = string("op_2403_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2403_keep_dims_0 = const()[name = string("op_2403_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 8, 1]> var_2403_cast_fp16 = reduce_mean(axes = var_2403_axes_0, keep_dims = var_2403_keep_dims_0, x = var_2401_cast_fp16)[name = string("op_2403_cast_fp16")];
            fp32 var_2404_epsilon_0 = const()[name = string("op_2404_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 8, 1]> var_2404_cast_fp16 = inverse(epsilon = var_2404_epsilon_0, x = s_93_cast_fp16)[name = string("op_2404_cast_fp16")];
            fp16 var_2405_to_fp16 = const()[name = string("op_2405_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 8, 1]> var_2406_cast_fp16 = mul(x = var_2404_cast_fp16, y = var_2405_to_fp16)[name = string("op_2406_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_2407_cast_fp16 = real_div(x = var_2406_cast_fp16, y = s_93_cast_fp16)[name = string("op_2407_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_2408_cast_fp16 = add(x = var_2403_cast_fp16, y = var_2407_cast_fp16)[name = string("op_2408_cast_fp16")];
            fp32 var_2409_epsilon_0 = const()[name = string("op_2409_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 8, 1]> var_2409_cast_fp16 = rsqrt(epsilon = var_2409_epsilon_0, x = var_2408_cast_fp16)[name = string("op_2409_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2410_cast_fp16 = mul(x = y_103_cast_fp16, y = var_2409_cast_fp16)[name = string("op_2410_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_15_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196632192)))];
            tensor<fp16, [1, 12, 8, 96]> var_2411_cast_fp16 = mul(x = core_backbone_layers_15_self_attn_q_norm_weight_to_fp16, y = var_2410_cast_fp16)[name = string("op_2411_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196632448)))];
            tensor<fp16, [1, 12, 384]> linear_86_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_k_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_86_cast_fp16")];
            tensor<int32, [4]> concat_42 = const()[name = string("concat_42"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> x_171_cast_fp16 = reshape(shape = concat_42, x = linear_86_cast_fp16)[name = string("x_171_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_2418_cast_fp16 = abs(x = x_171_cast_fp16)[name = string("op_2418_cast_fp16")];
            tensor<int32, [1]> var_2420_axes_0 = const()[name = string("op_2420_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 4, 1]> var_2420_cast_fp16 = reduce_max(axes = var_2420_axes_0, keep_dims = var_128, x = var_2418_cast_fp16)[name = string("op_2420_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> s_95_cast_fp16 = maximum(x = var_2420_cast_fp16, y = var_129_to_fp16)[name = string("s_95_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> y_105_cast_fp16 = real_div(x = x_171_cast_fp16, y = s_95_cast_fp16)[name = string("y_105_cast_fp16")];
            fp16 var_130_promoted_42_to_fp16 = const()[name = string("op_130_promoted_42_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 4, 96]> var_2423_cast_fp16 = pow(x = y_105_cast_fp16, y = var_130_promoted_42_to_fp16)[name = string("op_2423_cast_fp16")];
            tensor<int32, [1]> var_2425_axes_0 = const()[name = string("op_2425_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2425_keep_dims_0 = const()[name = string("op_2425_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 4, 1]> var_2425_cast_fp16 = reduce_mean(axes = var_2425_axes_0, keep_dims = var_2425_keep_dims_0, x = var_2423_cast_fp16)[name = string("op_2425_cast_fp16")];
            fp32 var_2426_epsilon_0 = const()[name = string("op_2426_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 4, 1]> var_2426_cast_fp16 = inverse(epsilon = var_2426_epsilon_0, x = s_95_cast_fp16)[name = string("op_2426_cast_fp16")];
            fp16 var_2427_to_fp16 = const()[name = string("op_2427_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 4, 1]> var_2428_cast_fp16 = mul(x = var_2426_cast_fp16, y = var_2427_to_fp16)[name = string("op_2428_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_2429_cast_fp16 = real_div(x = var_2428_cast_fp16, y = s_95_cast_fp16)[name = string("op_2429_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_2430_cast_fp16 = add(x = var_2425_cast_fp16, y = var_2429_cast_fp16)[name = string("op_2430_cast_fp16")];
            fp32 var_2431_epsilon_0 = const()[name = string("op_2431_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 4, 1]> var_2431_cast_fp16 = rsqrt(epsilon = var_2431_epsilon_0, x = var_2430_cast_fp16)[name = string("op_2431_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_2432_cast_fp16 = mul(x = y_105_cast_fp16, y = var_2431_cast_fp16)[name = string("op_2432_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_15_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197222336)))];
            tensor<fp16, [1, 12, 4, 96]> var_2433_cast_fp16 = mul(x = core_backbone_layers_15_self_attn_k_norm_weight_to_fp16, y = var_2432_cast_fp16)[name = string("op_2433_cast_fp16")];
            tensor<int32, [4]> new_41_perm_0 = const()[name = string("new_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197222592)))];
            tensor<fp16, [1, 12, 384]> linear_87_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_v_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_87_cast_fp16")];
            tensor<int32, [4]> concat_43 = const()[name = string("concat_43"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> var_2438_cast_fp16 = reshape(shape = concat_43, x = linear_87_cast_fp16)[name = string("op_2438_cast_fp16")];
            tensor<int32, [4]> new_43_perm_0 = const()[name = string("new_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_217_begin_0 = const()[name = string("input_217_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])];
            tensor<int32, [5]> input_217_end_0 = const()[name = string("input_217_end_0"), val = tensor<int32, [5]>([6, 1, 4, 31, 96])];
            tensor<bool, [5]> input_217_end_mask_0 = const()[name = string("input_217_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_217_squeeze_mask_0 = const()[name = string("input_217_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_217_cast_fp16 = slice_by_index(begin = input_217_begin_0, end = input_217_end_0, end_mask = input_217_end_mask_0, squeeze_mask = input_217_squeeze_mask_0, x = k)[name = string("input_217_cast_fp16")];
            tensor<int32, [8]> var_2443_pad_0 = const()[name = string("op_2443_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2443_mode_0 = const()[name = string("op_2443_mode_0"), val = string("constant")];
            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_41_cast_fp16 = transpose(perm = new_41_perm_0, x = var_2433_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 4, 43, 96]> var_2443_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = var_2443_mode_0, pad = var_2443_pad_0, x = new_41_cast_fp16)[name = string("op_2443_cast_fp16")];
            tensor<int32, [8]> var_2445_pad_0 = const()[name = string("op_2445_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_2445_mode_0 = const()[name = string("op_2445_mode_0"), val = string("constant")];
            fp16 const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_2445_cast_fp16 = pad(constant_val = const_55_to_fp16, mode = var_2445_mode_0, pad = var_2445_pad_0, x = input_217_cast_fp16)[name = string("op_2445_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> x_175_cast_fp16 = add(x = var_2443_cast_fp16, y = var_2445_cast_fp16)[name = string("x_175_cast_fp16")];
            tensor<int32, [5]> input_221_begin_0 = const()[name = string("input_221_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])];
            tensor<int32, [5]> input_221_end_0 = const()[name = string("input_221_end_0"), val = tensor<int32, [5]>([6, 1, 4, 31, 96])];
            tensor<bool, [5]> input_221_end_mask_0 = const()[name = string("input_221_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_221_squeeze_mask_0 = const()[name = string("input_221_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_221_cast_fp16 = slice_by_index(begin = input_221_begin_0, end = input_221_end_0, end_mask = input_221_end_mask_0, squeeze_mask = input_221_squeeze_mask_0, x = v)[name = string("input_221_cast_fp16")];
            tensor<int32, [8]> var_2450_pad_0 = const()[name = string("op_2450_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2450_mode_0 = const()[name = string("op_2450_mode_0"), val = string("constant")];
            fp16 const_56_to_fp16 = const()[name = string("const_56_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_43_cast_fp16 = transpose(perm = new_43_perm_0, x = var_2438_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 4, 43, 96]> var_2450_cast_fp16 = pad(constant_val = const_56_to_fp16, mode = var_2450_mode_0, pad = var_2450_pad_0, x = new_43_cast_fp16)[name = string("op_2450_cast_fp16")];
            tensor<int32, [8]> var_2452_pad_0 = const()[name = string("op_2452_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_2452_mode_0 = const()[name = string("op_2452_mode_0"), val = string("constant")];
            fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_2452_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = var_2452_mode_0, pad = var_2452_pad_0, x = input_221_cast_fp16)[name = string("op_2452_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> v_23_cast_fp16 = add(x = var_2450_cast_fp16, y = var_2452_cast_fp16)[name = string("v_23_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2454_cast_fp16 = mul(x = var_2411_cast_fp16, y = const_66_to_fp16)[name = string("op_2454_cast_fp16")];
            tensor<int32, [2]> var_2455_split_sizes_0 = const()[name = string("op_2455_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_87 = const()[name = string("const_87"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 48]> var_2455_cast_fp16_0, tensor<fp16, [1, 12, 8, 48]> var_2455_cast_fp16_1 = split(axis = const_87, split_sizes = var_2455_split_sizes_0, x = var_2411_cast_fp16)[name = string("op_2455_cast_fp16")];
            fp16 const_58_promoted_to_fp16 = const()[name = string("const_58_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 12, 8, 48]> var_2457_cast_fp16 = mul(x = var_2455_cast_fp16_1, y = const_58_promoted_to_fp16)[name = string("op_2457_cast_fp16")];
            bool var_2459_interleave_0 = const()[name = string("op_2459_interleave_0"), val = bool(false)];
            int32 const_88 = const()[name = string("const_88"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 96]> var_2459_cast_fp16 = concat(axis = const_88, interleave = var_2459_interleave_0, values = (var_2457_cast_fp16, var_2455_cast_fp16_0))[name = string("op_2459_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2460_cast_fp16 = mul(x = var_2459_cast_fp16, y = const_69_to_fp16)[name = string("op_2460_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> q_11_cast_fp16 = add(x = var_2454_cast_fp16, y = var_2460_cast_fp16)[name = string("q_11_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_2462_cast_fp16 = mul(x = x_175_cast_fp16, y = const_0_to_fp16)[name = string("op_2462_cast_fp16")];
            tensor<int32, [2]> var_2463_split_sizes_0 = const()[name = string("op_2463_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2463_axis_0 = const()[name = string("op_2463_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 43, 48]> var_2463_cast_fp16_0, tensor<fp16, [1, 4, 43, 48]> var_2463_cast_fp16_1 = split(axis = var_2463_axis_0, split_sizes = var_2463_split_sizes_0, x = x_175_cast_fp16)[name = string("op_2463_cast_fp16")];
            fp16 const_59_promoted_to_fp16 = const()[name = string("const_59_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 43, 48]> var_2465_cast_fp16 = mul(x = var_2463_cast_fp16_1, y = const_59_promoted_to_fp16)[name = string("op_2465_cast_fp16")];
            bool var_2467_interleave_0 = const()[name = string("op_2467_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 43, 96]> var_2467_cast_fp16 = concat(axis = var_127, interleave = var_2467_interleave_0, values = (var_2465_cast_fp16, var_2463_cast_fp16_0))[name = string("op_2467_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_2468_cast_fp16 = mul(x = var_2467_cast_fp16, y = const_1_to_fp16)[name = string("op_2468_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> k_23_cast_fp16 = add(x = var_2462_cast_fp16, y = var_2468_cast_fp16)[name = string("k_23_cast_fp16")];
            tensor<int32, [1]> var_2470_axes_0 = const()[name = string("op_2470_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_2470_cast_fp16 = expand_dims(axes = var_2470_axes_0, x = k_23_cast_fp16)[name = string("op_2470_cast_fp16")];
            tensor<int32, [5]> var_2472_reps_0 = const()[name = string("op_2472_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_2472_cast_fp16 = tile(reps = var_2472_reps_0, x = var_2470_cast_fp16)[name = string("op_2472_cast_fp16")];
            tensor<int32, [4]> concat_45 = const()[name = string("concat_45"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> k_25_cast_fp16 = reshape(shape = concat_45, x = var_2472_cast_fp16)[name = string("k_25_cast_fp16")];
            tensor<int32, [1]> var_2474_axes_0 = const()[name = string("op_2474_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_2474_cast_fp16 = expand_dims(axes = var_2474_axes_0, x = v_23_cast_fp16)[name = string("op_2474_cast_fp16")];
            tensor<int32, [5]> var_2476_reps_0 = const()[name = string("op_2476_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_2476_cast_fp16 = tile(reps = var_2476_reps_0, x = var_2474_cast_fp16)[name = string("op_2476_cast_fp16")];
            tensor<int32, [4]> concat_47 = const()[name = string("concat_47"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> v_25_cast_fp16 = reshape(shape = concat_47, x = var_2476_cast_fp16)[name = string("v_25_cast_fp16")];
            tensor<int32, [4]> transpose_10_perm_0 = const()[name = string("transpose_10_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2479_transpose_x_1 = const()[name = string("op_2479_transpose_x_1"), val = bool(false)];
            bool var_2479_transpose_y_1 = const()[name = string("op_2479_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 96]> transpose_10_cast_fp16 = transpose(perm = transpose_10_perm_0, x = q_11_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 8, 12, 43]> var_2479_cast_fp16 = matmul(transpose_x = var_2479_transpose_x_1, transpose_y = var_2479_transpose_y_1, x = transpose_10_cast_fp16, y = k_25_cast_fp16)[name = string("op_2479_cast_fp16")];
            fp16 var_2480_to_fp16 = const()[name = string("op_2480_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 12, 43]> s_97_cast_fp16 = mul(x = var_2479_cast_fp16, y = var_2480_to_fp16)[name = string("s_97_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_177_cast_fp16 = add(x = s_97_cast_fp16, y = mask)[name = string("x_177_cast_fp16")];
            tensor<int32, [1]> var_2484_axes_0 = const()[name = string("op_2484_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 12, 1]> var_2484_cast_fp16 = reduce_max(axes = var_2484_axes_0, keep_dims = var_128, x = x_177_cast_fp16)[name = string("op_2484_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_179_cast_fp16 = sub(x = x_177_cast_fp16, y = var_2484_cast_fp16)[name = string("x_179_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_181_cast_fp16 = exp(x = x_179_cast_fp16)[name = string("x_181_cast_fp16")];
            tensor<int32, [1]> var_2488_axes_0 = const()[name = string("op_2488_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2488_keep_dims_0 = const()[name = string("op_2488_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 1]> var_2488_cast_fp16 = reduce_sum(axes = var_2488_axes_0, keep_dims = var_2488_keep_dims_0, x = x_181_cast_fp16)[name = string("op_2488_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> var_2489_cast_fp16 = real_div(x = x_181_cast_fp16, y = var_2488_cast_fp16)[name = string("op_2489_cast_fp16")];
            bool o_21_transpose_x_0 = const()[name = string("o_21_transpose_x_0"), val = bool(false)];
            bool o_21_transpose_y_0 = const()[name = string("o_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 12, 96]> o_21_cast_fp16 = matmul(transpose_x = o_21_transpose_x_0, transpose_y = o_21_transpose_y_0, x = var_2489_cast_fp16, y = v_25_cast_fp16)[name = string("o_21_cast_fp16")];
            tensor<int32, [4]> o_23_perm_0 = const()[name = string("o_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_48 = const()[name = string("concat_48"), val = tensor<int32, [3]>([1, 12, 768])];
            tensor<fp16, [1, 12, 8, 96]> o_23_cast_fp16 = transpose(perm = o_23_perm_0, x = o_21_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 12, 768]> input_223_cast_fp16 = reshape(shape = concat_48, x = o_23_cast_fp16)[name = string("input_223_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197812480)))];
            tensor<fp16, [1, 12, 768]> linear_88_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_o_proj_weight_to_fp16, x = input_223_cast_fp16)[name = string("linear_88_cast_fp16")];
            tensor<int32, [4]> var_2498_begin_0 = const()[name = string("op_2498_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_2498_end_0 = const()[name = string("op_2498_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_2498_end_mask_0 = const()[name = string("op_2498_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2498_cast_fp16 = slice_by_index(begin = var_2498_begin_0, end = var_2498_end_0, end_mask = var_2498_end_mask_0, x = x_175_cast_fp16)[name = string("op_2498_cast_fp16")];
            tensor<int32, [4]> var_2501_begin_0 = const()[name = string("op_2501_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_2501_end_0 = const()[name = string("op_2501_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_2501_end_mask_0 = const()[name = string("op_2501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2501_cast_fp16 = slice_by_index(begin = var_2501_begin_0, end = var_2501_end_0, end_mask = var_2501_end_mask_0, x = v_23_cast_fp16)[name = string("op_2501_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_183_cast_fp16 = add(x = x_167_cast_fp16, y = linear_88_cast_fp16)[name = string("x_183_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2504_cast_fp16 = abs(x = x_183_cast_fp16)[name = string("op_2504_cast_fp16")];
            tensor<int32, [1]> var_2506_axes_0 = const()[name = string("op_2506_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2506_cast_fp16 = reduce_max(axes = var_2506_axes_0, keep_dims = var_128, x = var_2504_cast_fp16)[name = string("op_2506_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_99_cast_fp16 = maximum(x = var_2506_cast_fp16, y = var_129_to_fp16)[name = string("s_99_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_107_cast_fp16 = real_div(x = x_183_cast_fp16, y = s_99_cast_fp16)[name = string("y_107_cast_fp16")];
            fp16 var_130_promoted_43_to_fp16 = const()[name = string("op_130_promoted_43_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2509_cast_fp16 = pow(x = y_107_cast_fp16, y = var_130_promoted_43_to_fp16)[name = string("op_2509_cast_fp16")];
            tensor<int32, [1]> var_2511_axes_0 = const()[name = string("op_2511_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2511_keep_dims_0 = const()[name = string("op_2511_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2511_cast_fp16 = reduce_mean(axes = var_2511_axes_0, keep_dims = var_2511_keep_dims_0, x = var_2509_cast_fp16)[name = string("op_2511_cast_fp16")];
            fp32 var_2512_epsilon_0 = const()[name = string("op_2512_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2512_cast_fp16 = inverse(epsilon = var_2512_epsilon_0, x = s_99_cast_fp16)[name = string("op_2512_cast_fp16")];
            fp16 var_2513_to_fp16 = const()[name = string("op_2513_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2514_cast_fp16 = mul(x = var_2512_cast_fp16, y = var_2513_to_fp16)[name = string("op_2514_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2515_cast_fp16 = real_div(x = var_2514_cast_fp16, y = s_99_cast_fp16)[name = string("op_2515_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2516_cast_fp16 = add(x = var_2511_cast_fp16, y = var_2515_cast_fp16)[name = string("op_2516_cast_fp16")];
            fp32 var_2517_epsilon_0 = const()[name = string("op_2517_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2517_cast_fp16 = rsqrt(epsilon = var_2517_epsilon_0, x = var_2516_cast_fp16)[name = string("op_2517_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2518_cast_fp16 = mul(x = y_107_cast_fp16, y = var_2517_cast_fp16)[name = string("op_2518_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_15_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198992192)))];
            tensor<fp16, [1, 12, 768]> input_225_cast_fp16 = mul(x = core_backbone_layers_15_ffn_norm_weight_to_fp16, y = var_2518_cast_fp16)[name = string("input_225_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_15_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198993792)))];
            tensor<fp16, [1, 12, 1792]> linear_89_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w1_weight_to_fp16, x = input_225_cast_fp16)[name = string("linear_89_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_15_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201746368)))];
            tensor<fp16, [1, 12, 1792]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w3_weight_to_fp16, x = input_225_cast_fp16)[name = string("linear_90_cast_fp16")];
            fp16 var_2527_to_fp16 = const()[name = string("op_2527_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_2528_cast_fp16 = mul(x = linear_89_cast_fp16, y = var_2527_to_fp16)[name = string("op_2528_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2531_cast_fp16 = tanh(x = var_2528_cast_fp16)[name = string("op_2531_cast_fp16")];
            fp16 var_2532_to_fp16 = const()[name = string("op_2532_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_2533_cast_fp16 = add(x = var_2531_cast_fp16, y = var_2532_to_fp16)[name = string("op_2533_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2534_cast_fp16 = mul(x = var_2528_cast_fp16, y = var_2533_cast_fp16)[name = string("op_2534_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_227_cast_fp16 = mul(x = var_2534_cast_fp16, y = linear_90_cast_fp16)[name = string("input_227_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_15_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204498944)))];
            tensor<fp16, [1, 12, 768]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w2_weight_to_fp16, x = input_227_cast_fp16)[name = string("linear_91_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_187_cast_fp16 = add(x = x_183_cast_fp16, y = linear_91_cast_fp16)[name = string("x_187_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2540_cast_fp16 = abs(x = x_187_cast_fp16)[name = string("op_2540_cast_fp16")];
            tensor<int32, [1]> var_2542_axes_0 = const()[name = string("op_2542_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2542_cast_fp16 = reduce_max(axes = var_2542_axes_0, keep_dims = var_128, x = var_2540_cast_fp16)[name = string("op_2542_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_101_cast_fp16 = maximum(x = var_2542_cast_fp16, y = var_129_to_fp16)[name = string("s_101_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_109_cast_fp16 = real_div(x = x_187_cast_fp16, y = s_101_cast_fp16)[name = string("y_109_cast_fp16")];
            fp16 var_130_promoted_44_to_fp16 = const()[name = string("op_130_promoted_44_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2545_cast_fp16 = pow(x = y_109_cast_fp16, y = var_130_promoted_44_to_fp16)[name = string("op_2545_cast_fp16")];
            tensor<int32, [1]> var_2547_axes_0 = const()[name = string("op_2547_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2547_keep_dims_0 = const()[name = string("op_2547_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2547_cast_fp16 = reduce_mean(axes = var_2547_axes_0, keep_dims = var_2547_keep_dims_0, x = var_2545_cast_fp16)[name = string("op_2547_cast_fp16")];
            fp32 var_2548_epsilon_0 = const()[name = string("op_2548_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2548_cast_fp16 = inverse(epsilon = var_2548_epsilon_0, x = s_101_cast_fp16)[name = string("op_2548_cast_fp16")];
            fp16 var_2549_to_fp16 = const()[name = string("op_2549_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2550_cast_fp16 = mul(x = var_2548_cast_fp16, y = var_2549_to_fp16)[name = string("op_2550_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2551_cast_fp16 = real_div(x = var_2550_cast_fp16, y = s_101_cast_fp16)[name = string("op_2551_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2552_cast_fp16 = add(x = var_2547_cast_fp16, y = var_2551_cast_fp16)[name = string("op_2552_cast_fp16")];
            fp32 var_2553_epsilon_0 = const()[name = string("op_2553_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2553_cast_fp16 = rsqrt(epsilon = var_2553_epsilon_0, x = var_2552_cast_fp16)[name = string("op_2553_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2554_cast_fp16 = mul(x = y_109_cast_fp16, y = var_2553_cast_fp16)[name = string("op_2554_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_16_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207251520)))];
            tensor<fp16, [1, 12, 768]> input_229_cast_fp16 = mul(x = core_backbone_layers_16_operator_norm_weight_to_fp16, y = var_2554_cast_fp16)[name = string("input_229_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207253120)))];
            tensor<fp16, [1, 12, 768]> linear_92_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_q_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_92_cast_fp16")];
            tensor<int32, [4]> concat_49 = const()[name = string("concat_49"), val = tensor<int32, [4]>([1, 12, 8, 96])];
            tensor<fp16, [1, 12, 8, 96]> x_189_cast_fp16 = reshape(shape = concat_49, x = linear_92_cast_fp16)[name = string("x_189_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2561_cast_fp16 = abs(x = x_189_cast_fp16)[name = string("op_2561_cast_fp16")];
            tensor<int32, [1]> var_2563_axes_0 = const()[name = string("op_2563_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 8, 1]> var_2563_cast_fp16 = reduce_max(axes = var_2563_axes_0, keep_dims = var_128, x = var_2561_cast_fp16)[name = string("op_2563_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> s_103_cast_fp16 = maximum(x = var_2563_cast_fp16, y = var_129_to_fp16)[name = string("s_103_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> y_111_cast_fp16 = real_div(x = x_189_cast_fp16, y = s_103_cast_fp16)[name = string("y_111_cast_fp16")];
            fp16 var_130_promoted_45_to_fp16 = const()[name = string("op_130_promoted_45_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 8, 96]> var_2566_cast_fp16 = pow(x = y_111_cast_fp16, y = var_130_promoted_45_to_fp16)[name = string("op_2566_cast_fp16")];
            tensor<int32, [1]> var_2568_axes_0 = const()[name = string("op_2568_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2568_keep_dims_0 = const()[name = string("op_2568_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 8, 1]> var_2568_cast_fp16 = reduce_mean(axes = var_2568_axes_0, keep_dims = var_2568_keep_dims_0, x = var_2566_cast_fp16)[name = string("op_2568_cast_fp16")];
            fp32 var_2569_epsilon_0 = const()[name = string("op_2569_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 8, 1]> var_2569_cast_fp16 = inverse(epsilon = var_2569_epsilon_0, x = s_103_cast_fp16)[name = string("op_2569_cast_fp16")];
            fp16 var_2570_to_fp16 = const()[name = string("op_2570_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 8, 1]> var_2571_cast_fp16 = mul(x = var_2569_cast_fp16, y = var_2570_to_fp16)[name = string("op_2571_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_2572_cast_fp16 = real_div(x = var_2571_cast_fp16, y = s_103_cast_fp16)[name = string("op_2572_cast_fp16")];
            tensor<fp16, [1, 12, 8, 1]> var_2573_cast_fp16 = add(x = var_2568_cast_fp16, y = var_2572_cast_fp16)[name = string("op_2573_cast_fp16")];
            fp32 var_2574_epsilon_0 = const()[name = string("op_2574_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 8, 1]> var_2574_cast_fp16 = rsqrt(epsilon = var_2574_epsilon_0, x = var_2573_cast_fp16)[name = string("op_2574_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2575_cast_fp16 = mul(x = y_111_cast_fp16, y = var_2574_cast_fp16)[name = string("op_2575_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_16_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208432832)))];
            tensor<fp16, [1, 12, 8, 96]> var_2576_cast_fp16 = mul(x = core_backbone_layers_16_self_attn_q_norm_weight_to_fp16, y = var_2575_cast_fp16)[name = string("op_2576_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208433088)))];
            tensor<fp16, [1, 12, 384]> linear_93_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_k_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_93_cast_fp16")];
            tensor<int32, [4]> concat_50 = const()[name = string("concat_50"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> x_191_cast_fp16 = reshape(shape = concat_50, x = linear_93_cast_fp16)[name = string("x_191_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_2583_cast_fp16 = abs(x = x_191_cast_fp16)[name = string("op_2583_cast_fp16")];
            tensor<int32, [1]> var_2585_axes_0 = const()[name = string("op_2585_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 4, 1]> var_2585_cast_fp16 = reduce_max(axes = var_2585_axes_0, keep_dims = var_128, x = var_2583_cast_fp16)[name = string("op_2585_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> s_105_cast_fp16 = maximum(x = var_2585_cast_fp16, y = var_129_to_fp16)[name = string("s_105_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> y_113_cast_fp16 = real_div(x = x_191_cast_fp16, y = s_105_cast_fp16)[name = string("y_113_cast_fp16")];
            fp16 var_130_promoted_46_to_fp16 = const()[name = string("op_130_promoted_46_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 4, 96]> var_2588_cast_fp16 = pow(x = y_113_cast_fp16, y = var_130_promoted_46_to_fp16)[name = string("op_2588_cast_fp16")];
            tensor<int32, [1]> var_2590_axes_0 = const()[name = string("op_2590_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2590_keep_dims_0 = const()[name = string("op_2590_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 4, 1]> var_2590_cast_fp16 = reduce_mean(axes = var_2590_axes_0, keep_dims = var_2590_keep_dims_0, x = var_2588_cast_fp16)[name = string("op_2590_cast_fp16")];
            fp32 var_2591_epsilon_0 = const()[name = string("op_2591_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 4, 1]> var_2591_cast_fp16 = inverse(epsilon = var_2591_epsilon_0, x = s_105_cast_fp16)[name = string("op_2591_cast_fp16")];
            fp16 var_2592_to_fp16 = const()[name = string("op_2592_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 4, 1]> var_2593_cast_fp16 = mul(x = var_2591_cast_fp16, y = var_2592_to_fp16)[name = string("op_2593_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_2594_cast_fp16 = real_div(x = var_2593_cast_fp16, y = s_105_cast_fp16)[name = string("op_2594_cast_fp16")];
            tensor<fp16, [1, 12, 4, 1]> var_2595_cast_fp16 = add(x = var_2590_cast_fp16, y = var_2594_cast_fp16)[name = string("op_2595_cast_fp16")];
            fp32 var_2596_epsilon_0 = const()[name = string("op_2596_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 4, 1]> var_2596_cast_fp16 = rsqrt(epsilon = var_2596_epsilon_0, x = var_2595_cast_fp16)[name = string("op_2596_cast_fp16")];
            tensor<fp16, [1, 12, 4, 96]> var_2597_cast_fp16 = mul(x = y_113_cast_fp16, y = var_2596_cast_fp16)[name = string("op_2597_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_16_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209022976)))];
            tensor<fp16, [1, 12, 4, 96]> var_2598_cast_fp16 = mul(x = core_backbone_layers_16_self_attn_k_norm_weight_to_fp16, y = var_2597_cast_fp16)[name = string("op_2598_cast_fp16")];
            tensor<int32, [4]> new_45_perm_0 = const()[name = string("new_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209023232)))];
            tensor<fp16, [1, 12, 384]> linear_94_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_v_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_94_cast_fp16")];
            tensor<int32, [4]> concat_51 = const()[name = string("concat_51"), val = tensor<int32, [4]>([1, 12, 4, 96])];
            tensor<fp16, [1, 12, 4, 96]> var_2603_cast_fp16 = reshape(shape = concat_51, x = linear_94_cast_fp16)[name = string("op_2603_cast_fp16")];
            tensor<int32, [4]> new_perm_0 = const()[name = string("new_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_233_begin_0 = const()[name = string("input_233_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])];
            tensor<int32, [5]> input_233_end_0 = const()[name = string("input_233_end_0"), val = tensor<int32, [5]>([7, 1, 4, 31, 96])];
            tensor<bool, [5]> input_233_end_mask_0 = const()[name = string("input_233_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_233_squeeze_mask_0 = const()[name = string("input_233_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_233_cast_fp16 = slice_by_index(begin = input_233_begin_0, end = input_233_end_0, end_mask = input_233_end_mask_0, squeeze_mask = input_233_squeeze_mask_0, x = k)[name = string("input_233_cast_fp16")];
            tensor<int32, [8]> var_2608_pad_0 = const()[name = string("op_2608_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2608_mode_0 = const()[name = string("op_2608_mode_0"), val = string("constant")];
            fp16 const_60_to_fp16 = const()[name = string("const_60_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_45_cast_fp16 = transpose(perm = new_45_perm_0, x = var_2598_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 4, 43, 96]> var_2608_cast_fp16 = pad(constant_val = const_60_to_fp16, mode = var_2608_mode_0, pad = var_2608_pad_0, x = new_45_cast_fp16)[name = string("op_2608_cast_fp16")];
            tensor<int32, [8]> var_2610_pad_0 = const()[name = string("op_2610_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_2610_mode_0 = const()[name = string("op_2610_mode_0"), val = string("constant")];
            fp16 const_61_to_fp16 = const()[name = string("const_61_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_2610_cast_fp16 = pad(constant_val = const_61_to_fp16, mode = var_2610_mode_0, pad = var_2610_pad_0, x = input_233_cast_fp16)[name = string("op_2610_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> x_195_cast_fp16 = add(x = var_2608_cast_fp16, y = var_2610_cast_fp16)[name = string("x_195_cast_fp16")];
            tensor<int32, [5]> input_237_begin_0 = const()[name = string("input_237_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])];
            tensor<int32, [5]> input_237_end_0 = const()[name = string("input_237_end_0"), val = tensor<int32, [5]>([7, 1, 4, 31, 96])];
            tensor<bool, [5]> input_237_end_mask_0 = const()[name = string("input_237_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_237_squeeze_mask_0 = const()[name = string("input_237_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_237_cast_fp16 = slice_by_index(begin = input_237_begin_0, end = input_237_end_0, end_mask = input_237_end_mask_0, squeeze_mask = input_237_squeeze_mask_0, x = v)[name = string("input_237_cast_fp16")];
            tensor<int32, [8]> var_2615_pad_0 = const()[name = string("op_2615_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2615_mode_0 = const()[name = string("op_2615_mode_0"), val = string("constant")];
            fp16 const_62_to_fp16 = const()[name = string("const_62_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 12, 96]> new_cast_fp16 = transpose(perm = new_perm_0, x = var_2603_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 4, 43, 96]> var_2615_cast_fp16 = pad(constant_val = const_62_to_fp16, mode = var_2615_mode_0, pad = var_2615_pad_0, x = new_cast_fp16)[name = string("op_2615_cast_fp16")];
            tensor<int32, [8]> var_2617_pad_0 = const()[name = string("op_2617_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 12, 0, 0])];
            string var_2617_mode_0 = const()[name = string("op_2617_mode_0"), val = string("constant")];
            fp16 const_63_to_fp16 = const()[name = string("const_63_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 43, 96]> var_2617_cast_fp16 = pad(constant_val = const_63_to_fp16, mode = var_2617_mode_0, pad = var_2617_pad_0, x = input_237_cast_fp16)[name = string("op_2617_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> v_27_cast_fp16 = add(x = var_2615_cast_fp16, y = var_2617_cast_fp16)[name = string("v_27_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2619_cast_fp16 = mul(x = var_2576_cast_fp16, y = const_66_to_fp16)[name = string("op_2619_cast_fp16")];
            tensor<int32, [2]> var_2620_split_sizes_0 = const()[name = string("op_2620_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_91 = const()[name = string("const_91"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 48]> var_2620_cast_fp16_0, tensor<fp16, [1, 12, 8, 48]> var_2620_cast_fp16_1 = split(axis = const_91, split_sizes = var_2620_split_sizes_0, x = var_2576_cast_fp16)[name = string("op_2620_cast_fp16")];
            fp16 const_64_promoted_to_fp16 = const()[name = string("const_64_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 12, 8, 48]> var_2622_cast_fp16 = mul(x = var_2620_cast_fp16_1, y = const_64_promoted_to_fp16)[name = string("op_2622_cast_fp16")];
            bool var_2624_interleave_0 = const()[name = string("op_2624_interleave_0"), val = bool(false)];
            int32 const_92 = const()[name = string("const_92"), val = int32(3)];
            tensor<fp16, [1, 12, 8, 96]> var_2624_cast_fp16 = concat(axis = const_92, interleave = var_2624_interleave_0, values = (var_2622_cast_fp16, var_2620_cast_fp16_0))[name = string("op_2624_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> var_2625_cast_fp16 = mul(x = var_2624_cast_fp16, y = const_69_to_fp16)[name = string("op_2625_cast_fp16")];
            tensor<fp16, [1, 12, 8, 96]> q_cast_fp16 = add(x = var_2619_cast_fp16, y = var_2625_cast_fp16)[name = string("q_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_2627_cast_fp16 = mul(x = x_195_cast_fp16, y = const_0_to_fp16)[name = string("op_2627_cast_fp16")];
            tensor<int32, [2]> var_2628_split_sizes_0 = const()[name = string("op_2628_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2628_axis_0 = const()[name = string("op_2628_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 43, 48]> var_2628_cast_fp16_0, tensor<fp16, [1, 4, 43, 48]> var_2628_cast_fp16_1 = split(axis = var_2628_axis_0, split_sizes = var_2628_split_sizes_0, x = x_195_cast_fp16)[name = string("op_2628_cast_fp16")];
            fp16 const_65_promoted_to_fp16 = const()[name = string("const_65_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 43, 48]> var_2630_cast_fp16 = mul(x = var_2628_cast_fp16_1, y = const_65_promoted_to_fp16)[name = string("op_2630_cast_fp16")];
            bool var_2632_interleave_0 = const()[name = string("op_2632_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 43, 96]> var_2632_cast_fp16 = concat(axis = var_127, interleave = var_2632_interleave_0, values = (var_2630_cast_fp16, var_2628_cast_fp16_0))[name = string("op_2632_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> var_2633_cast_fp16 = mul(x = var_2632_cast_fp16, y = const_1_to_fp16)[name = string("op_2633_cast_fp16")];
            tensor<fp16, [1, 4, 43, 96]> k_27_cast_fp16 = add(x = var_2627_cast_fp16, y = var_2633_cast_fp16)[name = string("k_27_cast_fp16")];
            tensor<int32, [1]> var_2635_axes_0 = const()[name = string("op_2635_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_2635_cast_fp16 = expand_dims(axes = var_2635_axes_0, x = k_27_cast_fp16)[name = string("op_2635_cast_fp16")];
            tensor<int32, [5]> var_2637_reps_0 = const()[name = string("op_2637_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_2637_cast_fp16 = tile(reps = var_2637_reps_0, x = var_2635_cast_fp16)[name = string("op_2637_cast_fp16")];
            tensor<int32, [4]> concat_53 = const()[name = string("concat_53"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> k_cast_fp16 = reshape(shape = concat_53, x = var_2637_cast_fp16)[name = string("k_cast_fp16")];
            tensor<int32, [1]> var_2639_axes_0 = const()[name = string("op_2639_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 43, 96]> var_2639_cast_fp16 = expand_dims(axes = var_2639_axes_0, x = v_27_cast_fp16)[name = string("op_2639_cast_fp16")];
            tensor<int32, [5]> var_2641_reps_0 = const()[name = string("op_2641_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 43, 96]> var_2641_cast_fp16 = tile(reps = var_2641_reps_0, x = var_2639_cast_fp16)[name = string("op_2641_cast_fp16")];
            tensor<int32, [4]> concat_55 = const()[name = string("concat_55"), val = tensor<int32, [4]>([1, 8, 43, 96])];
            tensor<fp16, [1, 8, 43, 96]> v_cast_fp16 = reshape(shape = concat_55, x = var_2641_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [4]> transpose_12_perm_0 = const()[name = string("transpose_12_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2644_transpose_x_1 = const()[name = string("op_2644_transpose_x_1"), val = bool(false)];
            bool var_2644_transpose_y_1 = const()[name = string("op_2644_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 96]> transpose_12_cast_fp16 = transpose(perm = transpose_12_perm_0, x = q_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 8, 12, 43]> var_2644_cast_fp16 = matmul(transpose_x = var_2644_transpose_x_1, transpose_y = var_2644_transpose_y_1, x = transpose_12_cast_fp16, y = k_cast_fp16)[name = string("op_2644_cast_fp16")];
            fp16 var_2645_to_fp16 = const()[name = string("op_2645_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 12, 43]> s_107_cast_fp16 = mul(x = var_2644_cast_fp16, y = var_2645_to_fp16)[name = string("s_107_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_197_cast_fp16 = add(x = s_107_cast_fp16, y = mask)[name = string("x_197_cast_fp16")];
            tensor<int32, [1]> var_2649_axes_0 = const()[name = string("op_2649_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 12, 1]> var_2649_cast_fp16 = reduce_max(axes = var_2649_axes_0, keep_dims = var_128, x = x_197_cast_fp16)[name = string("op_2649_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_199_cast_fp16 = sub(x = x_197_cast_fp16, y = var_2649_cast_fp16)[name = string("x_199_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> x_201_cast_fp16 = exp(x = x_199_cast_fp16)[name = string("x_201_cast_fp16")];
            tensor<int32, [1]> var_2653_axes_0 = const()[name = string("op_2653_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2653_keep_dims_0 = const()[name = string("op_2653_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 12, 1]> var_2653_cast_fp16 = reduce_sum(axes = var_2653_axes_0, keep_dims = var_2653_keep_dims_0, x = x_201_cast_fp16)[name = string("op_2653_cast_fp16")];
            tensor<fp16, [1, 8, 12, 43]> var_2654_cast_fp16 = real_div(x = x_201_cast_fp16, y = var_2653_cast_fp16)[name = string("op_2654_cast_fp16")];
            bool o_25_transpose_x_0 = const()[name = string("o_25_transpose_x_0"), val = bool(false)];
            bool o_25_transpose_y_0 = const()[name = string("o_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 12, 96]> o_25_cast_fp16 = matmul(transpose_x = o_25_transpose_x_0, transpose_y = o_25_transpose_y_0, x = var_2654_cast_fp16, y = v_cast_fp16)[name = string("o_25_cast_fp16")];
            tensor<int32, [4]> o_perm_0 = const()[name = string("o_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_56 = const()[name = string("concat_56"), val = tensor<int32, [3]>([1, 12, 768])];
            tensor<fp16, [1, 12, 8, 96]> o_cast_fp16 = transpose(perm = o_perm_0, x = o_25_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 12, 768]> input_239_cast_fp16 = reshape(shape = concat_56, x = o_cast_fp16)[name = string("input_239_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209613120)))];
            tensor<fp16, [1, 12, 768]> linear_95_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_o_proj_weight_to_fp16, x = input_239_cast_fp16)[name = string("linear_95_cast_fp16")];
            tensor<int32, [4]> var_2663_begin_0 = const()[name = string("op_2663_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_2663_end_0 = const()[name = string("op_2663_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_2663_end_mask_0 = const()[name = string("op_2663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2663_cast_fp16 = slice_by_index(begin = var_2663_begin_0, end = var_2663_end_0, end_mask = var_2663_end_mask_0, x = x_195_cast_fp16)[name = string("op_2663_cast_fp16")];
            tensor<int32, [4]> var_2666_begin_0 = const()[name = string("op_2666_begin_0"), val = tensor<int32, [4]>([0, 0, 12, 0])];
            tensor<int32, [4]> var_2666_end_0 = const()[name = string("op_2666_end_0"), val = tensor<int32, [4]>([1, 4, 43, 96])];
            tensor<bool, [4]> var_2666_end_mask_0 = const()[name = string("op_2666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2666_cast_fp16 = slice_by_index(begin = var_2666_begin_0, end = var_2666_end_0, end_mask = var_2666_end_mask_0, x = v_27_cast_fp16)[name = string("op_2666_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_203_cast_fp16 = add(x = x_187_cast_fp16, y = linear_95_cast_fp16)[name = string("x_203_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2669_cast_fp16 = abs(x = x_203_cast_fp16)[name = string("op_2669_cast_fp16")];
            tensor<int32, [1]> var_2671_axes_0 = const()[name = string("op_2671_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2671_cast_fp16 = reduce_max(axes = var_2671_axes_0, keep_dims = var_128, x = var_2669_cast_fp16)[name = string("op_2671_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_109_cast_fp16 = maximum(x = var_2671_cast_fp16, y = var_129_to_fp16)[name = string("s_109_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_115_cast_fp16 = real_div(x = x_203_cast_fp16, y = s_109_cast_fp16)[name = string("y_115_cast_fp16")];
            fp16 var_130_promoted_47_to_fp16 = const()[name = string("op_130_promoted_47_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2674_cast_fp16 = pow(x = y_115_cast_fp16, y = var_130_promoted_47_to_fp16)[name = string("op_2674_cast_fp16")];
            tensor<int32, [1]> var_2676_axes_0 = const()[name = string("op_2676_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2676_keep_dims_0 = const()[name = string("op_2676_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2676_cast_fp16 = reduce_mean(axes = var_2676_axes_0, keep_dims = var_2676_keep_dims_0, x = var_2674_cast_fp16)[name = string("op_2676_cast_fp16")];
            fp32 var_2677_epsilon_0 = const()[name = string("op_2677_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2677_cast_fp16 = inverse(epsilon = var_2677_epsilon_0, x = s_109_cast_fp16)[name = string("op_2677_cast_fp16")];
            fp16 var_2678_to_fp16 = const()[name = string("op_2678_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2679_cast_fp16 = mul(x = var_2677_cast_fp16, y = var_2678_to_fp16)[name = string("op_2679_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2680_cast_fp16 = real_div(x = var_2679_cast_fp16, y = s_109_cast_fp16)[name = string("op_2680_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2681_cast_fp16 = add(x = var_2676_cast_fp16, y = var_2680_cast_fp16)[name = string("op_2681_cast_fp16")];
            fp32 var_2682_epsilon_0 = const()[name = string("op_2682_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2682_cast_fp16 = rsqrt(epsilon = var_2682_epsilon_0, x = var_2681_cast_fp16)[name = string("op_2682_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2683_cast_fp16 = mul(x = y_115_cast_fp16, y = var_2682_cast_fp16)[name = string("op_2683_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_16_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210792832)))];
            tensor<fp16, [1, 12, 768]> input_241_cast_fp16 = mul(x = core_backbone_layers_16_ffn_norm_weight_to_fp16, y = var_2683_cast_fp16)[name = string("input_241_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_16_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210794432)))];
            tensor<fp16, [1, 12, 1792]> linear_96_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w1_weight_to_fp16, x = input_241_cast_fp16)[name = string("linear_96_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_16_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213547008)))];
            tensor<fp16, [1, 12, 1792]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w3_weight_to_fp16, x = input_241_cast_fp16)[name = string("linear_97_cast_fp16")];
            fp16 var_2692_to_fp16 = const()[name = string("op_2692_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 12, 1792]> var_2693_cast_fp16 = mul(x = linear_96_cast_fp16, y = var_2692_to_fp16)[name = string("op_2693_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2696_cast_fp16 = tanh(x = var_2693_cast_fp16)[name = string("op_2696_cast_fp16")];
            fp16 var_2697_to_fp16 = const()[name = string("op_2697_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 12, 1792]> var_2698_cast_fp16 = add(x = var_2696_cast_fp16, y = var_2697_to_fp16)[name = string("op_2698_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> var_2699_cast_fp16 = mul(x = var_2693_cast_fp16, y = var_2698_cast_fp16)[name = string("op_2699_cast_fp16")];
            tensor<fp16, [1, 12, 1792]> input_243_cast_fp16 = mul(x = var_2699_cast_fp16, y = linear_97_cast_fp16)[name = string("input_243_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_16_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216299584)))];
            tensor<fp16, [1, 12, 768]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w2_weight_to_fp16, x = input_243_cast_fp16)[name = string("linear_98_cast_fp16")];
            tensor<fp16, [1, 12, 768]> x_cast_fp16 = add(x = x_203_cast_fp16, y = linear_98_cast_fp16)[name = string("x_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2705_cast_fp16 = abs(x = x_cast_fp16)[name = string("op_2705_cast_fp16")];
            tensor<int32, [1]> var_2707_axes_0 = const()[name = string("op_2707_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 12, 1]> var_2707_cast_fp16 = reduce_max(axes = var_2707_axes_0, keep_dims = var_128, x = var_2705_cast_fp16)[name = string("op_2707_cast_fp16")];
            tensor<fp16, [1, 12, 1]> s_cast_fp16 = maximum(x = var_2707_cast_fp16, y = var_129_to_fp16)[name = string("s_cast_fp16")];
            tensor<fp16, [1, 12, 768]> y_cast_fp16 = real_div(x = x_cast_fp16, y = s_cast_fp16)[name = string("y_cast_fp16")];
            fp16 var_130_promoted_48_to_fp16 = const()[name = string("op_130_promoted_48_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 12, 768]> var_2710_cast_fp16 = pow(x = y_cast_fp16, y = var_130_promoted_48_to_fp16)[name = string("op_2710_cast_fp16")];
            tensor<int32, [1]> var_2712_axes_0 = const()[name = string("op_2712_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2712_keep_dims_0 = const()[name = string("op_2712_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 12, 1]> var_2712_cast_fp16 = reduce_mean(axes = var_2712_axes_0, keep_dims = var_2712_keep_dims_0, x = var_2710_cast_fp16)[name = string("op_2712_cast_fp16")];
            fp32 var_2713_epsilon_0 = const()[name = string("op_2713_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 12, 1]> var_2713_cast_fp16 = inverse(epsilon = var_2713_epsilon_0, x = s_cast_fp16)[name = string("op_2713_cast_fp16")];
            fp16 var_2714_to_fp16 = const()[name = string("op_2714_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 1]> var_2715_cast_fp16 = mul(x = var_2713_cast_fp16, y = var_2714_to_fp16)[name = string("op_2715_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2716_cast_fp16 = real_div(x = var_2715_cast_fp16, y = s_cast_fp16)[name = string("op_2716_cast_fp16")];
            tensor<fp16, [1, 12, 1]> var_2717_cast_fp16 = add(x = var_2712_cast_fp16, y = var_2716_cast_fp16)[name = string("op_2717_cast_fp16")];
            fp32 var_2718_epsilon_0 = const()[name = string("op_2718_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 12, 1]> var_2718_cast_fp16 = rsqrt(epsilon = var_2718_epsilon_0, x = var_2717_cast_fp16)[name = string("op_2718_cast_fp16")];
            tensor<fp16, [1, 12, 768]> var_2719_cast_fp16 = mul(x = y_cast_fp16, y = var_2718_cast_fp16)[name = string("op_2719_cast_fp16")];
            tensor<fp16, [768]> core_backbone_embedding_norm_weight_to_fp16 = const()[name = string("core_backbone_embedding_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219052160)))];
            tensor<fp16, [1, 12, 768]> input_245_cast_fp16 = mul(x = core_backbone_embedding_norm_weight_to_fp16, y = var_2719_cast_fp16)[name = string("input_245_cast_fp16")];
            int32 var_2722_axis_0 = const()[name = string("op_2722_axis_0"), val = int32(0)];
            tensor<fp16, [10, 1, 768, 2]> conv_out = stack(axis = var_2722_axis_0, values = (var_753_cast_fp16, var_836_cast_fp16, var_1084_cast_fp16, var_1167_cast_fp16, var_1415_cast_fp16, var_1498_cast_fp16, var_1746_cast_fp16, var_1829_cast_fp16, var_2077_cast_fp16, var_2160_cast_fp16))[name = string("op_2722_cast_fp16")];
            int32 var_2724_axis_0 = const()[name = string("op_2724_axis_0"), val = int32(0)];
            tensor<fp16, [7, 1, 4, 31, 96]> k_out = stack(axis = var_2724_axis_0, values = (var_1009_cast_fp16, var_1340_cast_fp16, var_1671_cast_fp16, var_2002_cast_fp16, var_2333_cast_fp16, var_2498_cast_fp16, var_2663_cast_fp16))[name = string("op_2724_cast_fp16")];
            int32 var_2726_axis_0 = const()[name = string("op_2726_axis_0"), val = int32(0)];
            tensor<fp16, [7, 1, 4, 31, 96]> v_out = stack(axis = var_2726_axis_0, values = (var_1012_cast_fp16, var_1343_cast_fp16, var_1674_cast_fp16, var_2005_cast_fp16, var_2336_cast_fp16, var_2501_cast_fp16, var_2666_cast_fp16))[name = string("op_2726_cast_fp16")];
            tensor<int32, [1]> input_247_axes_0 = const()[name = string("input_247_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> projector_layer_norm_weight_to_fp16 = const()[name = string("projector_layer_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219053760)))];
            tensor<fp16, [768]> projector_layer_norm_bias_to_fp16 = const()[name = string("projector_layer_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219055360)))];
            fp16 var_2734_to_fp16 = const()[name = string("op_2734_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 12, 768]> input_247_cast_fp16 = layer_norm(axes = input_247_axes_0, beta = projector_layer_norm_bias_to_fp16, epsilon = var_2734_to_fp16, gamma = projector_layer_norm_weight_to_fp16, x = input_245_cast_fp16)[name = string("input_247_cast_fp16")];
            tensor<fp16, [2048, 768]> projector_linear_1_weight_to_fp16 = const()[name = string("projector_linear_1_weight_to_fp16"), val = tensor<fp16, [2048, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219056960)))];
            tensor<fp16, [2048]> projector_linear_1_bias_to_fp16 = const()[name = string("projector_linear_1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222202752)))];
            tensor<fp16, [1, 12, 2048]> linear_99_cast_fp16 = linear(bias = projector_linear_1_bias_to_fp16, weight = projector_linear_1_weight_to_fp16, x = input_247_cast_fp16)[name = string("linear_99_cast_fp16")];
            string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 12, 2048]> input_cast_fp16 = gelu(mode = input_mode_0, x = linear_99_cast_fp16)[name = string("input_cast_fp16")];
            tensor<fp16, [2048, 2048]> projector_linear_2_weight_to_fp16 = const()[name = string("projector_linear_2_weight_to_fp16"), val = tensor<fp16, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222206912)))];
            tensor<fp16, [2048]> projector_linear_2_bias_to_fp16 = const()[name = string("projector_linear_2_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230595584)))];
            tensor<fp16, [1, 12, 2048]> emb = linear(bias = projector_linear_2_bias_to_fp16, weight = projector_linear_2_weight_to_fp16, x = input_cast_fp16)[name = string("linear_100_cast_fp16")];
        } -> (emb, conv_out, k_out, v_out);
    func n375<ios18>(tensor<fp16, [10, 1, 768, 2]> conv, tensor<fp16, [7, 1, 4, 31, 96]> k, tensor<fp16, [1, 1, 1, 1504]> m1, tensor<fp16, [1, 1, 1, 752]> m2, tensor<fp16, [1, 1, 375, 406]> mask, tensor<fp16, [1, 128, 3008]> mel, tensor<fp16, [7, 1, 4, 31, 96]> v) {
            tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 128, 3008]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = mel)[name = string("input_1_cast_fp16")];
            string x_1_pad_type_0 = const()[name = string("x_1_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> x_1_pad_0 = const()[name = string("x_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> x_1_strides_0 = const()[name = string("x_1_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> x_1_dilations_0 = const()[name = string("x_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_1_groups_0 = const()[name = string("x_1_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_in_weight_to_fp16 = const()[name = string("downsample_conv_in_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [256]> downsample_conv_in_bias_to_fp16 = const()[name = string("downsample_conv_in_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4736)))];
            tensor<fp16, [1, 256, 64, 1504]> x_1_cast_fp16 = conv(bias = downsample_conv_in_bias_to_fp16, dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = downsample_conv_in_weight_to_fp16, x = input_1_cast_fp16)[name = string("x_1_cast_fp16")];
            fp16 var_31_to_fp16 = const()[name = string("op_31_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 64, 1504]> var_32_cast_fp16 = mul(x = x_1_cast_fp16, y = var_31_to_fp16)[name = string("op_32_cast_fp16")];
            tensor<fp16, [1, 256, 64, 1504]> var_35_cast_fp16 = tanh(x = var_32_cast_fp16)[name = string("op_35_cast_fp16")];
            fp16 var_36_to_fp16 = const()[name = string("op_36_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 64, 1504]> var_37_cast_fp16 = add(x = var_35_cast_fp16, y = var_36_to_fp16)[name = string("op_37_cast_fp16")];
            tensor<fp16, [1, 256, 64, 1504]> h_1_cast_fp16 = mul(x = var_32_cast_fp16, y = var_37_cast_fp16)[name = string("h_1_cast_fp16")];
            tensor<fp16, [1, 256, 64, 1504]> input_3_cast_fp16 = mul(x = h_1_cast_fp16, y = m1)[name = string("input_3_cast_fp16")];
            string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(256)];
            tensor<int32, [2]> input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_mid_weight_to_fp16 = const()[name = string("downsample_conv_mid_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5312)))];
            tensor<fp16, [256]> downsample_conv_mid_bias_to_fp16 = const()[name = string("downsample_conv_mid_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))];
            tensor<fp16, [1, 256, 32, 752]> input_5_cast_fp16 = conv(bias = downsample_conv_mid_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = downsample_conv_mid_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string x_3_pad_type_0 = const()[name = string("x_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_3_strides_0 = const()[name = string("x_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_3_pad_0 = const()[name = string("x_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_3_dilations_0 = const()[name = string("x_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_3_groups_0 = const()[name = string("x_3_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> downsample_lin_mid_weight_to_fp16 = const()[name = string("downsample_lin_mid_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10560)))];
            tensor<fp16, [256]> downsample_lin_mid_bias_to_fp16 = const()[name = string("downsample_lin_mid_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141696)))];
            tensor<fp16, [1, 256, 32, 752]> x_3_cast_fp16 = conv(bias = downsample_lin_mid_bias_to_fp16, dilations = x_3_dilations_0, groups = x_3_groups_0, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = x_3_strides_0, weight = downsample_lin_mid_weight_to_fp16, x = input_5_cast_fp16)[name = string("x_3_cast_fp16")];
            fp16 var_54_to_fp16 = const()[name = string("op_54_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 32, 752]> var_55_cast_fp16 = mul(x = x_3_cast_fp16, y = var_54_to_fp16)[name = string("op_55_cast_fp16")];
            tensor<fp16, [1, 256, 32, 752]> var_58_cast_fp16 = tanh(x = var_55_cast_fp16)[name = string("op_58_cast_fp16")];
            fp16 var_59_to_fp16 = const()[name = string("op_59_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 32, 752]> var_60_cast_fp16 = add(x = var_58_cast_fp16, y = var_59_to_fp16)[name = string("op_60_cast_fp16")];
            tensor<fp16, [1, 256, 32, 752]> h_3_cast_fp16 = mul(x = var_55_cast_fp16, y = var_60_cast_fp16)[name = string("h_3_cast_fp16")];
            tensor<fp16, [1, 256, 32, 752]> input_7_cast_fp16 = mul(x = h_3_cast_fp16, y = m2)[name = string("input_7_cast_fp16")];
            string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(256)];
            tensor<int32, [2]> input_9_dilations_0 = const()[name = string("input_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_end_weight_to_fp16 = const()[name = string("downsample_conv_end_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142272)))];
            tensor<fp16, [256]> downsample_conv_end_bias_to_fp16 = const()[name = string("downsample_conv_end_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146944)))];
            tensor<fp16, [1, 256, 16, 376]> input_9_cast_fp16 = conv(bias = downsample_conv_end_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = downsample_conv_end_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
            string x_5_pad_type_0 = const()[name = string("x_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_5_strides_0 = const()[name = string("x_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_5_pad_0 = const()[name = string("x_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_5_dilations_0 = const()[name = string("x_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_5_groups_0 = const()[name = string("x_5_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> downsample_lin_end_weight_to_fp16 = const()[name = string("downsample_lin_end_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147520)))];
            tensor<fp16, [256]> downsample_lin_end_bias_to_fp16 = const()[name = string("downsample_lin_end_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278656)))];
            tensor<fp16, [1, 256, 16, 376]> x_5_cast_fp16 = conv(bias = downsample_lin_end_bias_to_fp16, dilations = x_5_dilations_0, groups = x_5_groups_0, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = x_5_strides_0, weight = downsample_lin_end_weight_to_fp16, x = input_9_cast_fp16)[name = string("x_5_cast_fp16")];
            fp16 var_77_to_fp16 = const()[name = string("op_77_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 16, 376]> var_78_cast_fp16 = mul(x = x_5_cast_fp16, y = var_77_to_fp16)[name = string("op_78_cast_fp16")];
            tensor<fp16, [1, 256, 16, 376]> var_81_cast_fp16 = tanh(x = var_78_cast_fp16)[name = string("op_81_cast_fp16")];
            fp16 var_82_to_fp16 = const()[name = string("op_82_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 16, 376]> var_83_cast_fp16 = add(x = var_81_cast_fp16, y = var_82_to_fp16)[name = string("op_83_cast_fp16")];
            tensor<fp16, [1, 256, 16, 376]> h_cast_fp16 = mul(x = var_78_cast_fp16, y = var_83_cast_fp16)[name = string("h_cast_fp16")];
            tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 4096, 376])];
            tensor<fp16, [1, 4096, 376]> input_11_cast_fp16 = reshape(shape = concat_0, x = h_cast_fp16)[name = string("input_11_cast_fp16")];
            string d_pad_type_0 = const()[name = string("d_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> d_strides_0 = const()[name = string("d_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> d_pad_0 = const()[name = string("d_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> d_dilations_0 = const()[name = string("d_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 d_groups_0 = const()[name = string("d_groups_0"), val = int32(1)];
            tensor<fp16, [768, 4096, 1]> downsample_lin_out_weight_to_fp16 = const()[name = string("downsample_lin_out_weight_to_fp16"), val = tensor<fp16, [768, 4096, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279232)))];
            tensor<fp16, [768]> downsample_lin_out_bias_to_fp16 = const()[name = string("downsample_lin_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6570752)))];
            tensor<fp16, [1, 768, 376]> d_cast_fp16 = conv(bias = downsample_lin_out_bias_to_fp16, dilations = d_dilations_0, groups = d_groups_0, pad = d_pad_0, pad_type = d_pad_type_0, strides = d_strides_0, weight = downsample_lin_out_weight_to_fp16, x = input_11_cast_fp16)[name = string("d_cast_fp16")];
            tensor<int32, [3]> var_107_begin_0 = const()[name = string("op_107_begin_0"), val = tensor<int32, [3]>([0, 0, 1])];
            tensor<int32, [3]> var_107_end_0 = const()[name = string("op_107_end_0"), val = tensor<int32, [3]>([1, 768, 376])];
            tensor<bool, [3]> var_107_end_mask_0 = const()[name = string("op_107_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> var_107_cast_fp16 = slice_by_index(begin = var_107_begin_0, end = var_107_end_0, end_mask = var_107_end_mask_0, x = d_cast_fp16)[name = string("op_107_cast_fp16")];
            tensor<int32, [3]> x_7_perm_0 = const()[name = string("x_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_127 = const()[name = string("op_127"), val = int32(-1)];
            bool var_128 = const()[name = string("op_128"), val = bool(true)];
            tensor<fp16, [1, 375, 768]> x_7_cast_fp16 = transpose(perm = x_7_perm_0, x = var_107_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [1, 375, 768]> var_720_cast_fp16 = abs(x = x_7_cast_fp16)[name = string("op_720_cast_fp16")];
            tensor<int32, [1]> var_722_axes_0 = const()[name = string("op_722_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_722_cast_fp16 = reduce_max(axes = var_722_axes_0, keep_dims = var_128, x = var_720_cast_fp16)[name = string("op_722_cast_fp16")];
            fp16 var_129_to_fp16 = const()[name = string("op_129_to_fp16"), val = fp16(0x1.a38p-14)];
            tensor<fp16, [1, 375, 1]> s_1_cast_fp16 = maximum(x = var_722_cast_fp16, y = var_129_to_fp16)[name = string("s_1_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_1_cast_fp16 = real_div(x = x_7_cast_fp16, y = s_1_cast_fp16)[name = string("y_1_cast_fp16")];
            fp16 var_130_promoted_to_fp16 = const()[name = string("op_130_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_725_cast_fp16 = pow(x = y_1_cast_fp16, y = var_130_promoted_to_fp16)[name = string("op_725_cast_fp16")];
            tensor<int32, [1]> var_727_axes_0 = const()[name = string("op_727_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_727_keep_dims_0 = const()[name = string("op_727_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_727_cast_fp16 = reduce_mean(axes = var_727_axes_0, keep_dims = var_727_keep_dims_0, x = var_725_cast_fp16)[name = string("op_727_cast_fp16")];
            fp32 var_728_epsilon_0 = const()[name = string("op_728_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_728_cast_fp16 = inverse(epsilon = var_728_epsilon_0, x = s_1_cast_fp16)[name = string("op_728_cast_fp16")];
            fp16 var_729_to_fp16 = const()[name = string("op_729_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_730_cast_fp16 = mul(x = var_728_cast_fp16, y = var_729_to_fp16)[name = string("op_730_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_731_cast_fp16 = real_div(x = var_730_cast_fp16, y = s_1_cast_fp16)[name = string("op_731_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_732_cast_fp16 = add(x = var_727_cast_fp16, y = var_731_cast_fp16)[name = string("op_732_cast_fp16")];
            fp32 var_733_epsilon_0 = const()[name = string("op_733_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_733_cast_fp16 = rsqrt(epsilon = var_733_epsilon_0, x = var_732_cast_fp16)[name = string("op_733_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_734_cast_fp16 = mul(x = y_1_cast_fp16, y = var_733_cast_fp16)[name = string("op_734_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_0_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6572352)))];
            tensor<fp16, [1, 375, 768]> input_13_cast_fp16 = mul(x = core_backbone_layers_0_operator_norm_weight_to_fp16, y = var_734_cast_fp16)[name = string("input_13_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_0_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6573952)))];
            tensor<fp16, [2304]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10112960)))];
            tensor<fp16, [1, 375, 2304]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_0_conv_in_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<int32, [3]> bcx_1_perm_0 = const()[name = string("bcx_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_1_begin_0 = const()[name = string("b_1_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_1_end_0 = const()[name = string("b_1_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_1_end_mask_0 = const()[name = string("b_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_1_cast_fp16 = transpose(perm = bcx_1_perm_0, x = linear_0_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [1, 768, 375]> b_1_cast_fp16 = slice_by_index(begin = b_1_begin_0, end = b_1_end_0, end_mask = b_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("b_1_cast_fp16")];
            tensor<int32, [3]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("c_1_cast_fp16")];
            tensor<int32, [3]> xx_1_begin_0 = const()[name = string("xx_1_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_1_end_0 = const()[name = string("xx_1_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_1_end_mask_0 = const()[name = string("xx_1_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_1_cast_fp16 = slice_by_index(begin = xx_1_begin_0, end = xx_1_end_0, end_mask = xx_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("xx_1_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_1_cast_fp16 = mul(x = b_1_cast_fp16, y = xx_1_cast_fp16)[name = string("new_1_cast_fp16")];
            tensor<int32, [4]> input_17_begin_0 = const()[name = string("input_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_17_end_0 = const()[name = string("input_17_end_0"), val = tensor<int32, [4]>([1, 1, 768, 2])];
            tensor<bool, [4]> input_17_end_mask_0 = const()[name = string("input_17_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_17_squeeze_mask_0 = const()[name = string("input_17_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_17_cast_fp16 = slice_by_index(begin = input_17_begin_0, end = input_17_end_0, end_mask = input_17_end_mask_0, squeeze_mask = input_17_squeeze_mask_0, x = conv)[name = string("input_17_cast_fp16")];
            tensor<int32, [6]> var_749_pad_0 = const()[name = string("op_749_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_749_mode_0 = const()[name = string("op_749_mode_0"), val = string("constant")];
            fp16 const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_749_cast_fp16 = pad(constant_val = const_4_to_fp16, mode = var_749_mode_0, pad = var_749_pad_0, x = new_1_cast_fp16)[name = string("op_749_cast_fp16")];
            tensor<int32, [6]> var_751_pad_0 = const()[name = string("op_751_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_751_mode_0 = const()[name = string("op_751_mode_0"), val = string("constant")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_751_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = var_751_mode_0, pad = var_751_pad_0, x = input_17_cast_fp16)[name = string("op_751_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_1_cast_fp16 = add(x = var_749_cast_fp16, y = var_751_cast_fp16)[name = string("win_1_cast_fp16")];
            tensor<int32, [3]> var_753_begin_0 = const()[name = string("op_753_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_753_end_0 = const()[name = string("op_753_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_753_end_mask_0 = const()[name = string("op_753_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_753_cast_fp16 = slice_by_index(begin = var_753_begin_0, end = var_753_end_0, end_mask = var_753_end_mask_0, x = win_1_cast_fp16)[name = string("op_753_cast_fp16")];
            string var_759_pad_type_0 = const()[name = string("op_759_pad_type_0"), val = string("valid")];
            int32 var_759_groups_0 = const()[name = string("op_759_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_759_strides_0 = const()[name = string("op_759_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_759_pad_0 = const()[name = string("op_759_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_759_dilations_0 = const()[name = string("op_759_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_0_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10117632)))];
            tensor<fp16, [1, 768, 375]> var_759_cast_fp16 = conv(dilations = var_759_dilations_0, groups = var_759_groups_0, pad = var_759_pad_0, pad_type = var_759_pad_type_0, strides = var_759_strides_0, weight = core_backbone_layers_0_conv_conv_weight_to_fp16, x = win_1_cast_fp16)[name = string("op_759_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_761_cast_fp16 = mul(x = c_1_cast_fp16, y = var_759_cast_fp16)[name = string("op_761_cast_fp16")];
            tensor<int32, [3]> input_19_perm_0 = const()[name = string("input_19_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_0_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10122304)))];
            tensor<fp16, [768]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11302016)))];
            tensor<fp16, [1, 375, 768]> input_19_cast_fp16 = transpose(perm = input_19_perm_0, x = var_761_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [1, 375, 768]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_conv_out_proj_weight_to_fp16, x = input_19_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_9_cast_fp16 = add(x = x_7_cast_fp16, y = linear_1_cast_fp16)[name = string("x_9_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_767_cast_fp16 = abs(x = x_9_cast_fp16)[name = string("op_767_cast_fp16")];
            tensor<int32, [1]> var_769_axes_0 = const()[name = string("op_769_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_769_cast_fp16 = reduce_max(axes = var_769_axes_0, keep_dims = var_128, x = var_767_cast_fp16)[name = string("op_769_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_3_cast_fp16 = maximum(x = var_769_cast_fp16, y = var_129_to_fp16)[name = string("s_3_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_5_cast_fp16 = real_div(x = x_9_cast_fp16, y = s_3_cast_fp16)[name = string("y_5_cast_fp16")];
            fp16 var_130_promoted_1_to_fp16 = const()[name = string("op_130_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_772_cast_fp16 = pow(x = y_5_cast_fp16, y = var_130_promoted_1_to_fp16)[name = string("op_772_cast_fp16")];
            tensor<int32, [1]> var_774_axes_0 = const()[name = string("op_774_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_774_keep_dims_0 = const()[name = string("op_774_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_774_cast_fp16 = reduce_mean(axes = var_774_axes_0, keep_dims = var_774_keep_dims_0, x = var_772_cast_fp16)[name = string("op_774_cast_fp16")];
            fp32 var_775_epsilon_0 = const()[name = string("op_775_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_775_cast_fp16 = inverse(epsilon = var_775_epsilon_0, x = s_3_cast_fp16)[name = string("op_775_cast_fp16")];
            fp16 var_776_to_fp16 = const()[name = string("op_776_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_777_cast_fp16 = mul(x = var_775_cast_fp16, y = var_776_to_fp16)[name = string("op_777_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_778_cast_fp16 = real_div(x = var_777_cast_fp16, y = s_3_cast_fp16)[name = string("op_778_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_779_cast_fp16 = add(x = var_774_cast_fp16, y = var_778_cast_fp16)[name = string("op_779_cast_fp16")];
            fp32 var_780_epsilon_0 = const()[name = string("op_780_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_780_cast_fp16 = rsqrt(epsilon = var_780_epsilon_0, x = var_779_cast_fp16)[name = string("op_780_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_781_cast_fp16 = mul(x = y_5_cast_fp16, y = var_780_cast_fp16)[name = string("op_781_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_0_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11303616)))];
            tensor<fp16, [1, 375, 768]> input_21_cast_fp16 = mul(x = core_backbone_layers_0_ffn_norm_weight_to_fp16, y = var_781_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_0_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11305216)))];
            tensor<fp16, [1792]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14057792)))];
            tensor<fp16, [1, 375, 1792]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w1_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_0_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14061440)))];
            tensor<fp16, [1, 375, 1792]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w3_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_3_cast_fp16")];
            fp16 var_790_to_fp16 = const()[name = string("op_790_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_791_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_790_to_fp16)[name = string("op_791_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_794_cast_fp16 = tanh(x = var_791_cast_fp16)[name = string("op_794_cast_fp16")];
            fp16 var_795_to_fp16 = const()[name = string("op_795_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_796_cast_fp16 = add(x = var_794_cast_fp16, y = var_795_to_fp16)[name = string("op_796_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_797_cast_fp16 = mul(x = var_791_cast_fp16, y = var_796_cast_fp16)[name = string("op_797_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_23_cast_fp16 = mul(x = var_797_cast_fp16, y = linear_3_cast_fp16)[name = string("input_23_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_0_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16814016)))];
            tensor<fp16, [1, 375, 768]> linear_4_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w2_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_13_cast_fp16 = add(x = x_9_cast_fp16, y = linear_4_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_803_cast_fp16 = abs(x = x_13_cast_fp16)[name = string("op_803_cast_fp16")];
            tensor<int32, [1]> var_805_axes_0 = const()[name = string("op_805_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_805_cast_fp16 = reduce_max(axes = var_805_axes_0, keep_dims = var_128, x = var_803_cast_fp16)[name = string("op_805_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_5_cast_fp16 = maximum(x = var_805_cast_fp16, y = var_129_to_fp16)[name = string("s_5_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_7_cast_fp16 = real_div(x = x_13_cast_fp16, y = s_5_cast_fp16)[name = string("y_7_cast_fp16")];
            fp16 var_130_promoted_2_to_fp16 = const()[name = string("op_130_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_808_cast_fp16 = pow(x = y_7_cast_fp16, y = var_130_promoted_2_to_fp16)[name = string("op_808_cast_fp16")];
            tensor<int32, [1]> var_810_axes_0 = const()[name = string("op_810_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_810_keep_dims_0 = const()[name = string("op_810_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_810_cast_fp16 = reduce_mean(axes = var_810_axes_0, keep_dims = var_810_keep_dims_0, x = var_808_cast_fp16)[name = string("op_810_cast_fp16")];
            fp32 var_811_epsilon_0 = const()[name = string("op_811_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_811_cast_fp16 = inverse(epsilon = var_811_epsilon_0, x = s_5_cast_fp16)[name = string("op_811_cast_fp16")];
            fp16 var_812_to_fp16 = const()[name = string("op_812_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_813_cast_fp16 = mul(x = var_811_cast_fp16, y = var_812_to_fp16)[name = string("op_813_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_814_cast_fp16 = real_div(x = var_813_cast_fp16, y = s_5_cast_fp16)[name = string("op_814_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_815_cast_fp16 = add(x = var_810_cast_fp16, y = var_814_cast_fp16)[name = string("op_815_cast_fp16")];
            fp32 var_816_epsilon_0 = const()[name = string("op_816_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_816_cast_fp16 = rsqrt(epsilon = var_816_epsilon_0, x = var_815_cast_fp16)[name = string("op_816_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_817_cast_fp16 = mul(x = y_7_cast_fp16, y = var_816_cast_fp16)[name = string("op_817_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_1_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19566592)))];
            tensor<fp16, [1, 375, 768]> input_25_cast_fp16 = mul(x = core_backbone_layers_1_operator_norm_weight_to_fp16, y = var_817_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_1_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19568192)))];
            tensor<fp16, [1, 375, 2304]> linear_5_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_1_conv_in_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [3]> bcx_3_perm_0 = const()[name = string("bcx_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_5_begin_0 = const()[name = string("b_5_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_5_end_0 = const()[name = string("b_5_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_5_end_mask_0 = const()[name = string("b_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_3_cast_fp16 = transpose(perm = bcx_3_perm_0, x = linear_5_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [1, 768, 375]> b_5_cast_fp16 = slice_by_index(begin = b_5_begin_0, end = b_5_end_0, end_mask = b_5_end_mask_0, x = bcx_3_cast_fp16)[name = string("b_5_cast_fp16")];
            tensor<int32, [3]> c_3_begin_0 = const()[name = string("c_3_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_3_end_0 = const()[name = string("c_3_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_3_end_mask_0 = const()[name = string("c_3_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_3_cast_fp16 = slice_by_index(begin = c_3_begin_0, end = c_3_end_0, end_mask = c_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("c_3_cast_fp16")];
            tensor<int32, [3]> xx_3_begin_0 = const()[name = string("xx_3_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_3_end_0 = const()[name = string("xx_3_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_3_end_mask_0 = const()[name = string("xx_3_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_3_cast_fp16 = slice_by_index(begin = xx_3_begin_0, end = xx_3_end_0, end_mask = xx_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("xx_3_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_3_cast_fp16 = mul(x = b_5_cast_fp16, y = xx_3_cast_fp16)[name = string("new_3_cast_fp16")];
            tensor<int32, [4]> input_29_begin_0 = const()[name = string("input_29_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])];
            tensor<int32, [4]> input_29_end_0 = const()[name = string("input_29_end_0"), val = tensor<int32, [4]>([2, 1, 768, 2])];
            tensor<bool, [4]> input_29_end_mask_0 = const()[name = string("input_29_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_29_squeeze_mask_0 = const()[name = string("input_29_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_29_cast_fp16 = slice_by_index(begin = input_29_begin_0, end = input_29_end_0, end_mask = input_29_end_mask_0, squeeze_mask = input_29_squeeze_mask_0, x = conv)[name = string("input_29_cast_fp16")];
            tensor<int32, [6]> var_832_pad_0 = const()[name = string("op_832_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_832_mode_0 = const()[name = string("op_832_mode_0"), val = string("constant")];
            fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_832_cast_fp16 = pad(constant_val = const_6_to_fp16, mode = var_832_mode_0, pad = var_832_pad_0, x = new_3_cast_fp16)[name = string("op_832_cast_fp16")];
            tensor<int32, [6]> var_834_pad_0 = const()[name = string("op_834_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_834_mode_0 = const()[name = string("op_834_mode_0"), val = string("constant")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_834_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = var_834_mode_0, pad = var_834_pad_0, x = input_29_cast_fp16)[name = string("op_834_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_5_cast_fp16 = add(x = var_832_cast_fp16, y = var_834_cast_fp16)[name = string("win_5_cast_fp16")];
            tensor<int32, [3]> var_836_begin_0 = const()[name = string("op_836_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_836_end_0 = const()[name = string("op_836_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_836_end_mask_0 = const()[name = string("op_836_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_836_cast_fp16 = slice_by_index(begin = var_836_begin_0, end = var_836_end_0, end_mask = var_836_end_mask_0, x = win_5_cast_fp16)[name = string("op_836_cast_fp16")];
            string var_842_pad_type_0 = const()[name = string("op_842_pad_type_0"), val = string("valid")];
            int32 var_842_groups_0 = const()[name = string("op_842_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_842_strides_0 = const()[name = string("op_842_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_842_pad_0 = const()[name = string("op_842_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_842_dilations_0 = const()[name = string("op_842_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_1_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23107200)))];
            tensor<fp16, [1, 768, 375]> var_842_cast_fp16 = conv(dilations = var_842_dilations_0, groups = var_842_groups_0, pad = var_842_pad_0, pad_type = var_842_pad_type_0, strides = var_842_strides_0, weight = core_backbone_layers_1_conv_conv_weight_to_fp16, x = win_5_cast_fp16)[name = string("op_842_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_844_cast_fp16 = mul(x = c_3_cast_fp16, y = var_842_cast_fp16)[name = string("op_844_cast_fp16")];
            tensor<int32, [3]> input_31_perm_0 = const()[name = string("input_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_1_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23111872)))];
            tensor<fp16, [1, 375, 768]> input_31_cast_fp16 = transpose(perm = input_31_perm_0, x = var_844_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [1, 375, 768]> linear_6_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_conv_out_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_15_cast_fp16 = add(x = x_13_cast_fp16, y = linear_6_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_850_cast_fp16 = abs(x = x_15_cast_fp16)[name = string("op_850_cast_fp16")];
            tensor<int32, [1]> var_852_axes_0 = const()[name = string("op_852_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_852_cast_fp16 = reduce_max(axes = var_852_axes_0, keep_dims = var_128, x = var_850_cast_fp16)[name = string("op_852_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_7_cast_fp16 = maximum(x = var_852_cast_fp16, y = var_129_to_fp16)[name = string("s_7_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_11_cast_fp16 = real_div(x = x_15_cast_fp16, y = s_7_cast_fp16)[name = string("y_11_cast_fp16")];
            fp16 var_130_promoted_3_to_fp16 = const()[name = string("op_130_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_855_cast_fp16 = pow(x = y_11_cast_fp16, y = var_130_promoted_3_to_fp16)[name = string("op_855_cast_fp16")];
            tensor<int32, [1]> var_857_axes_0 = const()[name = string("op_857_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_857_keep_dims_0 = const()[name = string("op_857_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_857_cast_fp16 = reduce_mean(axes = var_857_axes_0, keep_dims = var_857_keep_dims_0, x = var_855_cast_fp16)[name = string("op_857_cast_fp16")];
            fp32 var_858_epsilon_0 = const()[name = string("op_858_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_858_cast_fp16 = inverse(epsilon = var_858_epsilon_0, x = s_7_cast_fp16)[name = string("op_858_cast_fp16")];
            fp16 var_859_to_fp16 = const()[name = string("op_859_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_860_cast_fp16 = mul(x = var_858_cast_fp16, y = var_859_to_fp16)[name = string("op_860_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_861_cast_fp16 = real_div(x = var_860_cast_fp16, y = s_7_cast_fp16)[name = string("op_861_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_862_cast_fp16 = add(x = var_857_cast_fp16, y = var_861_cast_fp16)[name = string("op_862_cast_fp16")];
            fp32 var_863_epsilon_0 = const()[name = string("op_863_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_863_cast_fp16 = rsqrt(epsilon = var_863_epsilon_0, x = var_862_cast_fp16)[name = string("op_863_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_864_cast_fp16 = mul(x = y_11_cast_fp16, y = var_863_cast_fp16)[name = string("op_864_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_1_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24291584)))];
            tensor<fp16, [1, 375, 768]> input_33_cast_fp16 = mul(x = core_backbone_layers_1_ffn_norm_weight_to_fp16, y = var_864_cast_fp16)[name = string("input_33_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_1_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24293184)))];
            tensor<fp16, [1, 375, 1792]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w1_weight_to_fp16, x = input_33_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_1_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27045760)))];
            tensor<fp16, [1, 375, 1792]> linear_8_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w3_weight_to_fp16, x = input_33_cast_fp16)[name = string("linear_8_cast_fp16")];
            fp16 var_873_to_fp16 = const()[name = string("op_873_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_874_cast_fp16 = mul(x = linear_7_cast_fp16, y = var_873_to_fp16)[name = string("op_874_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_877_cast_fp16 = tanh(x = var_874_cast_fp16)[name = string("op_877_cast_fp16")];
            fp16 var_878_to_fp16 = const()[name = string("op_878_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_879_cast_fp16 = add(x = var_877_cast_fp16, y = var_878_to_fp16)[name = string("op_879_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_880_cast_fp16 = mul(x = var_874_cast_fp16, y = var_879_cast_fp16)[name = string("op_880_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_35_cast_fp16 = mul(x = var_880_cast_fp16, y = linear_8_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_1_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29798336)))];
            tensor<fp16, [1, 375, 768]> linear_9_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w2_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_9_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_19_cast_fp16 = add(x = x_15_cast_fp16, y = linear_9_cast_fp16)[name = string("x_19_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_886_cast_fp16 = abs(x = x_19_cast_fp16)[name = string("op_886_cast_fp16")];
            tensor<int32, [1]> var_888_axes_0 = const()[name = string("op_888_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_888_cast_fp16 = reduce_max(axes = var_888_axes_0, keep_dims = var_128, x = var_886_cast_fp16)[name = string("op_888_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_9_cast_fp16 = maximum(x = var_888_cast_fp16, y = var_129_to_fp16)[name = string("s_9_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_13_cast_fp16 = real_div(x = x_19_cast_fp16, y = s_9_cast_fp16)[name = string("y_13_cast_fp16")];
            fp16 var_130_promoted_4_to_fp16 = const()[name = string("op_130_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_891_cast_fp16 = pow(x = y_13_cast_fp16, y = var_130_promoted_4_to_fp16)[name = string("op_891_cast_fp16")];
            tensor<int32, [1]> var_893_axes_0 = const()[name = string("op_893_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_893_keep_dims_0 = const()[name = string("op_893_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_893_cast_fp16 = reduce_mean(axes = var_893_axes_0, keep_dims = var_893_keep_dims_0, x = var_891_cast_fp16)[name = string("op_893_cast_fp16")];
            fp32 var_894_epsilon_0 = const()[name = string("op_894_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_894_cast_fp16 = inverse(epsilon = var_894_epsilon_0, x = s_9_cast_fp16)[name = string("op_894_cast_fp16")];
            fp16 var_895_to_fp16 = const()[name = string("op_895_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_896_cast_fp16 = mul(x = var_894_cast_fp16, y = var_895_to_fp16)[name = string("op_896_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_897_cast_fp16 = real_div(x = var_896_cast_fp16, y = s_9_cast_fp16)[name = string("op_897_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_898_cast_fp16 = add(x = var_893_cast_fp16, y = var_897_cast_fp16)[name = string("op_898_cast_fp16")];
            fp32 var_899_epsilon_0 = const()[name = string("op_899_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_899_cast_fp16 = rsqrt(epsilon = var_899_epsilon_0, x = var_898_cast_fp16)[name = string("op_899_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_900_cast_fp16 = mul(x = y_13_cast_fp16, y = var_899_cast_fp16)[name = string("op_900_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_2_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32550912)))];
            tensor<fp16, [1, 375, 768]> input_37_cast_fp16 = mul(x = core_backbone_layers_2_operator_norm_weight_to_fp16, y = var_900_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32552512)))];
            tensor<fp16, [1, 375, 768]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_q_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<int32, [4]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [4]>([1, 375, 8, 96])];
            tensor<fp16, [1, 375, 8, 96]> x_21_cast_fp16 = reshape(shape = concat_1, x = linear_10_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_907_cast_fp16 = abs(x = x_21_cast_fp16)[name = string("op_907_cast_fp16")];
            tensor<int32, [1]> var_909_axes_0 = const()[name = string("op_909_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 8, 1]> var_909_cast_fp16 = reduce_max(axes = var_909_axes_0, keep_dims = var_128, x = var_907_cast_fp16)[name = string("op_909_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> s_11_cast_fp16 = maximum(x = var_909_cast_fp16, y = var_129_to_fp16)[name = string("s_11_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> y_15_cast_fp16 = real_div(x = x_21_cast_fp16, y = s_11_cast_fp16)[name = string("y_15_cast_fp16")];
            fp16 var_130_promoted_5_to_fp16 = const()[name = string("op_130_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 8, 96]> var_912_cast_fp16 = pow(x = y_15_cast_fp16, y = var_130_promoted_5_to_fp16)[name = string("op_912_cast_fp16")];
            tensor<int32, [1]> var_914_axes_0 = const()[name = string("op_914_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_914_keep_dims_0 = const()[name = string("op_914_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 8, 1]> var_914_cast_fp16 = reduce_mean(axes = var_914_axes_0, keep_dims = var_914_keep_dims_0, x = var_912_cast_fp16)[name = string("op_914_cast_fp16")];
            fp32 var_915_epsilon_0 = const()[name = string("op_915_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 8, 1]> var_915_cast_fp16 = inverse(epsilon = var_915_epsilon_0, x = s_11_cast_fp16)[name = string("op_915_cast_fp16")];
            fp16 var_916_to_fp16 = const()[name = string("op_916_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 8, 1]> var_917_cast_fp16 = mul(x = var_915_cast_fp16, y = var_916_to_fp16)[name = string("op_917_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_918_cast_fp16 = real_div(x = var_917_cast_fp16, y = s_11_cast_fp16)[name = string("op_918_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_919_cast_fp16 = add(x = var_914_cast_fp16, y = var_918_cast_fp16)[name = string("op_919_cast_fp16")];
            fp32 var_920_epsilon_0 = const()[name = string("op_920_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 8, 1]> var_920_cast_fp16 = rsqrt(epsilon = var_920_epsilon_0, x = var_919_cast_fp16)[name = string("op_920_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_921_cast_fp16 = mul(x = y_15_cast_fp16, y = var_920_cast_fp16)[name = string("op_921_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_2_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33732224)))];
            tensor<fp16, [1, 375, 8, 96]> var_922_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_q_norm_weight_to_fp16, y = var_921_cast_fp16)[name = string("op_922_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33732480)))];
            tensor<fp16, [384]> linear_11_bias_0_to_fp16 = const()[name = string("linear_11_bias_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34322368)))];
            tensor<fp16, [1, 375, 384]> linear_11_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_k_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<int32, [4]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> x_23_cast_fp16 = reshape(shape = concat_2, x = linear_11_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_929_cast_fp16 = abs(x = x_23_cast_fp16)[name = string("op_929_cast_fp16")];
            tensor<int32, [1]> var_931_axes_0 = const()[name = string("op_931_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 4, 1]> var_931_cast_fp16 = reduce_max(axes = var_931_axes_0, keep_dims = var_128, x = var_929_cast_fp16)[name = string("op_931_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> s_13_cast_fp16 = maximum(x = var_931_cast_fp16, y = var_129_to_fp16)[name = string("s_13_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> y_17_cast_fp16 = real_div(x = x_23_cast_fp16, y = s_13_cast_fp16)[name = string("y_17_cast_fp16")];
            fp16 var_130_promoted_6_to_fp16 = const()[name = string("op_130_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 4, 96]> var_934_cast_fp16 = pow(x = y_17_cast_fp16, y = var_130_promoted_6_to_fp16)[name = string("op_934_cast_fp16")];
            tensor<int32, [1]> var_936_axes_0 = const()[name = string("op_936_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_936_keep_dims_0 = const()[name = string("op_936_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 4, 1]> var_936_cast_fp16 = reduce_mean(axes = var_936_axes_0, keep_dims = var_936_keep_dims_0, x = var_934_cast_fp16)[name = string("op_936_cast_fp16")];
            fp32 var_937_epsilon_0 = const()[name = string("op_937_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 4, 1]> var_937_cast_fp16 = inverse(epsilon = var_937_epsilon_0, x = s_13_cast_fp16)[name = string("op_937_cast_fp16")];
            fp16 var_938_to_fp16 = const()[name = string("op_938_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 4, 1]> var_939_cast_fp16 = mul(x = var_937_cast_fp16, y = var_938_to_fp16)[name = string("op_939_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_940_cast_fp16 = real_div(x = var_939_cast_fp16, y = s_13_cast_fp16)[name = string("op_940_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_941_cast_fp16 = add(x = var_936_cast_fp16, y = var_940_cast_fp16)[name = string("op_941_cast_fp16")];
            fp32 var_942_epsilon_0 = const()[name = string("op_942_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 4, 1]> var_942_cast_fp16 = rsqrt(epsilon = var_942_epsilon_0, x = var_941_cast_fp16)[name = string("op_942_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_943_cast_fp16 = mul(x = y_17_cast_fp16, y = var_942_cast_fp16)[name = string("op_943_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_2_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34323200)))];
            tensor<fp16, [1, 375, 4, 96]> var_944_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_k_norm_weight_to_fp16, y = var_943_cast_fp16)[name = string("op_944_cast_fp16")];
            tensor<int32, [4]> new_5_perm_0 = const()[name = string("new_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34323456)))];
            tensor<fp16, [1, 375, 384]> linear_12_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_v_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> var_949_cast_fp16 = reshape(shape = concat_3, x = linear_12_cast_fp16)[name = string("op_949_cast_fp16")];
            tensor<int32, [4]> new_7_perm_0 = const()[name = string("new_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_41_begin_0 = const()[name = string("input_41_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_41_end_0 = const()[name = string("input_41_end_0"), val = tensor<int32, [5]>([1, 1, 4, 31, 96])];
            tensor<bool, [5]> input_41_end_mask_0 = const()[name = string("input_41_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_41_squeeze_mask_0 = const()[name = string("input_41_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_41_cast_fp16 = slice_by_index(begin = input_41_begin_0, end = input_41_end_0, end_mask = input_41_end_mask_0, squeeze_mask = input_41_squeeze_mask_0, x = k)[name = string("input_41_cast_fp16")];
            tensor<int32, [8]> var_954_pad_0 = const()[name = string("op_954_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_954_mode_0 = const()[name = string("op_954_mode_0"), val = string("constant")];
            fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_5_cast_fp16 = transpose(perm = new_5_perm_0, x = var_944_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [1, 4, 406, 96]> var_954_cast_fp16 = pad(constant_val = const_8_to_fp16, mode = var_954_mode_0, pad = var_954_pad_0, x = new_5_cast_fp16)[name = string("op_954_cast_fp16")];
            tensor<int32, [8]> var_956_pad_0 = const()[name = string("op_956_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_956_mode_0 = const()[name = string("op_956_mode_0"), val = string("constant")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_956_cast_fp16 = pad(constant_val = const_9_to_fp16, mode = var_956_mode_0, pad = var_956_pad_0, x = input_41_cast_fp16)[name = string("op_956_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> x_27_cast_fp16 = add(x = var_954_cast_fp16, y = var_956_cast_fp16)[name = string("x_27_cast_fp16")];
            tensor<int32, [5]> input_45_begin_0 = const()[name = string("input_45_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_45_end_0 = const()[name = string("input_45_end_0"), val = tensor<int32, [5]>([1, 1, 4, 31, 96])];
            tensor<bool, [5]> input_45_end_mask_0 = const()[name = string("input_45_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_45_squeeze_mask_0 = const()[name = string("input_45_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_45_cast_fp16 = slice_by_index(begin = input_45_begin_0, end = input_45_end_0, end_mask = input_45_end_mask_0, squeeze_mask = input_45_squeeze_mask_0, x = v)[name = string("input_45_cast_fp16")];
            tensor<int32, [8]> var_961_pad_0 = const()[name = string("op_961_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_961_mode_0 = const()[name = string("op_961_mode_0"), val = string("constant")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_7_cast_fp16 = transpose(perm = new_7_perm_0, x = var_949_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [1, 4, 406, 96]> var_961_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = var_961_mode_0, pad = var_961_pad_0, x = new_7_cast_fp16)[name = string("op_961_cast_fp16")];
            tensor<int32, [8]> var_963_pad_0 = const()[name = string("op_963_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_963_mode_0 = const()[name = string("op_963_mode_0"), val = string("constant")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_963_cast_fp16 = pad(constant_val = const_11_to_fp16, mode = var_963_mode_0, pad = var_963_pad_0, x = input_45_cast_fp16)[name = string("op_963_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> v_3_cast_fp16 = add(x = var_961_cast_fp16, y = var_963_cast_fp16)[name = string("v_3_cast_fp16")];
            tensor<fp16, [1, 375, 1, 96]> const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = tensor<fp16, [1, 375, 1, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230685632)))];
            tensor<fp16, [1, 375, 8, 96]> var_965_cast_fp16 = mul(x = var_922_cast_fp16, y = const_66_to_fp16)[name = string("op_965_cast_fp16")];
            tensor<int32, [2]> var_966_split_sizes_0 = const()[name = string("op_966_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_67 = const()[name = string("const_67"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 48]> var_966_cast_fp16_0, tensor<fp16, [1, 375, 8, 48]> var_966_cast_fp16_1 = split(axis = const_67, split_sizes = var_966_split_sizes_0, x = var_922_cast_fp16)[name = string("op_966_cast_fp16")];
            fp16 const_12_promoted_to_fp16 = const()[name = string("const_12_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 375, 8, 48]> var_968_cast_fp16 = mul(x = var_966_cast_fp16_1, y = const_12_promoted_to_fp16)[name = string("op_968_cast_fp16")];
            bool var_970_interleave_0 = const()[name = string("op_970_interleave_0"), val = bool(false)];
            int32 const_68 = const()[name = string("const_68"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 96]> var_970_cast_fp16 = concat(axis = const_68, interleave = var_970_interleave_0, values = (var_968_cast_fp16, var_966_cast_fp16_0))[name = string("op_970_cast_fp16")];
            tensor<fp16, [1, 375, 1, 96]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [1, 375, 1, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230757696)))];
            tensor<fp16, [1, 375, 8, 96]> var_971_cast_fp16 = mul(x = var_970_cast_fp16, y = const_69_to_fp16)[name = string("op_971_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> q_1_cast_fp16 = add(x = var_965_cast_fp16, y = var_971_cast_fp16)[name = string("q_1_cast_fp16")];
            tensor<fp16, [1, 1, 406, 96]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [1, 1, 406, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230829760)))];
            tensor<fp16, [1, 4, 406, 96]> var_973_cast_fp16 = mul(x = x_27_cast_fp16, y = const_0_to_fp16)[name = string("op_973_cast_fp16")];
            tensor<int32, [2]> var_974_split_sizes_0 = const()[name = string("op_974_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_974_axis_0 = const()[name = string("op_974_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 406, 48]> var_974_cast_fp16_0, tensor<fp16, [1, 4, 406, 48]> var_974_cast_fp16_1 = split(axis = var_974_axis_0, split_sizes = var_974_split_sizes_0, x = x_27_cast_fp16)[name = string("op_974_cast_fp16")];
            fp16 const_13_promoted_to_fp16 = const()[name = string("const_13_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 406, 48]> var_976_cast_fp16 = mul(x = var_974_cast_fp16_1, y = const_13_promoted_to_fp16)[name = string("op_976_cast_fp16")];
            bool var_978_interleave_0 = const()[name = string("op_978_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 406, 96]> var_978_cast_fp16 = concat(axis = var_127, interleave = var_978_interleave_0, values = (var_976_cast_fp16, var_974_cast_fp16_0))[name = string("op_978_cast_fp16")];
            tensor<fp16, [1, 1, 406, 96]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [1, 1, 406, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230907776)))];
            tensor<fp16, [1, 4, 406, 96]> var_979_cast_fp16 = mul(x = var_978_cast_fp16, y = const_1_to_fp16)[name = string("op_979_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> k_3_cast_fp16 = add(x = var_973_cast_fp16, y = var_979_cast_fp16)[name = string("k_3_cast_fp16")];
            tensor<int32, [1]> var_981_axes_0 = const()[name = string("op_981_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_981_cast_fp16 = expand_dims(axes = var_981_axes_0, x = k_3_cast_fp16)[name = string("op_981_cast_fp16")];
            tensor<int32, [5]> var_983_reps_0 = const()[name = string("op_983_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_983_cast_fp16 = tile(reps = var_983_reps_0, x = var_981_cast_fp16)[name = string("op_983_cast_fp16")];
            tensor<int32, [4]> concat_5 = const()[name = string("concat_5"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> k_5_cast_fp16 = reshape(shape = concat_5, x = var_983_cast_fp16)[name = string("k_5_cast_fp16")];
            tensor<int32, [1]> var_985_axes_0 = const()[name = string("op_985_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_985_cast_fp16 = expand_dims(axes = var_985_axes_0, x = v_3_cast_fp16)[name = string("op_985_cast_fp16")];
            tensor<int32, [5]> var_987_reps_0 = const()[name = string("op_987_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_987_cast_fp16 = tile(reps = var_987_reps_0, x = var_985_cast_fp16)[name = string("op_987_cast_fp16")];
            tensor<int32, [4]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> v_5_cast_fp16 = reshape(shape = concat_7, x = var_987_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> transpose_0_perm_0 = const()[name = string("transpose_0_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_990_transpose_x_1 = const()[name = string("op_990_transpose_x_1"), val = bool(false)];
            bool var_990_transpose_y_1 = const()[name = string("op_990_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 96]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = q_1_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [1, 8, 375, 406]> var_990_cast_fp16 = matmul(transpose_x = var_990_transpose_x_1, transpose_y = var_990_transpose_y_1, x = transpose_0_cast_fp16, y = k_5_cast_fp16)[name = string("op_990_cast_fp16")];
            fp16 var_991_to_fp16 = const()[name = string("op_991_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 375, 406]> s_15_cast_fp16 = mul(x = var_990_cast_fp16, y = var_991_to_fp16)[name = string("s_15_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_29_cast_fp16 = add(x = s_15_cast_fp16, y = mask)[name = string("x_29_cast_fp16")];
            tensor<int32, [1]> var_995_axes_0 = const()[name = string("op_995_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 375, 1]> var_995_cast_fp16 = reduce_max(axes = var_995_axes_0, keep_dims = var_128, x = x_29_cast_fp16)[name = string("op_995_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_31_cast_fp16 = sub(x = x_29_cast_fp16, y = var_995_cast_fp16)[name = string("x_31_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_33_cast_fp16 = exp(x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<int32, [1]> var_999_axes_0 = const()[name = string("op_999_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_999_keep_dims_0 = const()[name = string("op_999_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 1]> var_999_cast_fp16 = reduce_sum(axes = var_999_axes_0, keep_dims = var_999_keep_dims_0, x = x_33_cast_fp16)[name = string("op_999_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> var_1000_cast_fp16 = real_div(x = x_33_cast_fp16, y = var_999_cast_fp16)[name = string("op_1000_cast_fp16")];
            bool o_1_transpose_x_0 = const()[name = string("o_1_transpose_x_0"), val = bool(false)];
            bool o_1_transpose_y_0 = const()[name = string("o_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 375, 96]> o_1_cast_fp16 = matmul(transpose_x = o_1_transpose_x_0, transpose_y = o_1_transpose_y_0, x = var_1000_cast_fp16, y = v_5_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<int32, [4]> o_3_perm_0 = const()[name = string("o_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [3]>([1, 375, 768])];
            tensor<fp16, [1, 375, 8, 96]> o_3_cast_fp16 = transpose(perm = o_3_perm_0, x = o_1_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [1, 375, 768]> input_47_cast_fp16 = reshape(shape = concat_8, x = o_3_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34934720)))];
            tensor<fp16, [1, 375, 768]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_o_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> var_1009_begin_0 = const()[name = string("op_1009_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_1009_end_0 = const()[name = string("op_1009_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_1009_end_mask_0 = const()[name = string("op_1009_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1009_cast_fp16 = slice_by_index(begin = var_1009_begin_0, end = var_1009_end_0, end_mask = var_1009_end_mask_0, x = x_27_cast_fp16)[name = string("op_1009_cast_fp16")];
            tensor<int32, [4]> var_1012_begin_0 = const()[name = string("op_1012_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_1012_end_0 = const()[name = string("op_1012_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_1012_end_mask_0 = const()[name = string("op_1012_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = v_3_cast_fp16)[name = string("op_1012_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_35_cast_fp16 = add(x = x_19_cast_fp16, y = linear_13_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1015_cast_fp16 = abs(x = x_35_cast_fp16)[name = string("op_1015_cast_fp16")];
            tensor<int32, [1]> var_1017_axes_0 = const()[name = string("op_1017_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1017_cast_fp16 = reduce_max(axes = var_1017_axes_0, keep_dims = var_128, x = var_1015_cast_fp16)[name = string("op_1017_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_17_cast_fp16 = maximum(x = var_1017_cast_fp16, y = var_129_to_fp16)[name = string("s_17_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_19_cast_fp16 = real_div(x = x_35_cast_fp16, y = s_17_cast_fp16)[name = string("y_19_cast_fp16")];
            fp16 var_130_promoted_7_to_fp16 = const()[name = string("op_130_promoted_7_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1020_cast_fp16 = pow(x = y_19_cast_fp16, y = var_130_promoted_7_to_fp16)[name = string("op_1020_cast_fp16")];
            tensor<int32, [1]> var_1022_axes_0 = const()[name = string("op_1022_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1022_keep_dims_0 = const()[name = string("op_1022_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1022_cast_fp16 = reduce_mean(axes = var_1022_axes_0, keep_dims = var_1022_keep_dims_0, x = var_1020_cast_fp16)[name = string("op_1022_cast_fp16")];
            fp32 var_1023_epsilon_0 = const()[name = string("op_1023_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1023_cast_fp16 = inverse(epsilon = var_1023_epsilon_0, x = s_17_cast_fp16)[name = string("op_1023_cast_fp16")];
            fp16 var_1024_to_fp16 = const()[name = string("op_1024_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1025_cast_fp16 = mul(x = var_1023_cast_fp16, y = var_1024_to_fp16)[name = string("op_1025_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1026_cast_fp16 = real_div(x = var_1025_cast_fp16, y = s_17_cast_fp16)[name = string("op_1026_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1027_cast_fp16 = add(x = var_1022_cast_fp16, y = var_1026_cast_fp16)[name = string("op_1027_cast_fp16")];
            fp32 var_1028_epsilon_0 = const()[name = string("op_1028_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1028_cast_fp16 = rsqrt(epsilon = var_1028_epsilon_0, x = var_1027_cast_fp16)[name = string("op_1028_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1029_cast_fp16 = mul(x = y_19_cast_fp16, y = var_1028_cast_fp16)[name = string("op_1029_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_2_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36114432)))];
            tensor<fp16, [1, 375, 768]> input_49_cast_fp16 = mul(x = core_backbone_layers_2_ffn_norm_weight_to_fp16, y = var_1029_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_2_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36116032)))];
            tensor<fp16, [1, 375, 1792]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w1_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_2_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38868608)))];
            tensor<fp16, [1, 375, 1792]> linear_15_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w3_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_15_cast_fp16")];
            fp16 var_1038_to_fp16 = const()[name = string("op_1038_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1039_cast_fp16 = mul(x = linear_14_cast_fp16, y = var_1038_to_fp16)[name = string("op_1039_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1042_cast_fp16 = tanh(x = var_1039_cast_fp16)[name = string("op_1042_cast_fp16")];
            fp16 var_1043_to_fp16 = const()[name = string("op_1043_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1044_cast_fp16 = add(x = var_1042_cast_fp16, y = var_1043_to_fp16)[name = string("op_1044_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1045_cast_fp16 = mul(x = var_1039_cast_fp16, y = var_1044_cast_fp16)[name = string("op_1045_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_51_cast_fp16 = mul(x = var_1045_cast_fp16, y = linear_15_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_2_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41621184)))];
            tensor<fp16, [1, 375, 768]> linear_16_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w2_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_39_cast_fp16 = add(x = x_35_cast_fp16, y = linear_16_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1051_cast_fp16 = abs(x = x_39_cast_fp16)[name = string("op_1051_cast_fp16")];
            tensor<int32, [1]> var_1053_axes_0 = const()[name = string("op_1053_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1053_cast_fp16 = reduce_max(axes = var_1053_axes_0, keep_dims = var_128, x = var_1051_cast_fp16)[name = string("op_1053_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_19_cast_fp16 = maximum(x = var_1053_cast_fp16, y = var_129_to_fp16)[name = string("s_19_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_21_cast_fp16 = real_div(x = x_39_cast_fp16, y = s_19_cast_fp16)[name = string("y_21_cast_fp16")];
            fp16 var_130_promoted_8_to_fp16 = const()[name = string("op_130_promoted_8_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1056_cast_fp16 = pow(x = y_21_cast_fp16, y = var_130_promoted_8_to_fp16)[name = string("op_1056_cast_fp16")];
            tensor<int32, [1]> var_1058_axes_0 = const()[name = string("op_1058_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1058_keep_dims_0 = const()[name = string("op_1058_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1058_cast_fp16 = reduce_mean(axes = var_1058_axes_0, keep_dims = var_1058_keep_dims_0, x = var_1056_cast_fp16)[name = string("op_1058_cast_fp16")];
            fp32 var_1059_epsilon_0 = const()[name = string("op_1059_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1059_cast_fp16 = inverse(epsilon = var_1059_epsilon_0, x = s_19_cast_fp16)[name = string("op_1059_cast_fp16")];
            fp16 var_1060_to_fp16 = const()[name = string("op_1060_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1061_cast_fp16 = mul(x = var_1059_cast_fp16, y = var_1060_to_fp16)[name = string("op_1061_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1062_cast_fp16 = real_div(x = var_1061_cast_fp16, y = s_19_cast_fp16)[name = string("op_1062_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1063_cast_fp16 = add(x = var_1058_cast_fp16, y = var_1062_cast_fp16)[name = string("op_1063_cast_fp16")];
            fp32 var_1064_epsilon_0 = const()[name = string("op_1064_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1064_cast_fp16 = rsqrt(epsilon = var_1064_epsilon_0, x = var_1063_cast_fp16)[name = string("op_1064_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1065_cast_fp16 = mul(x = y_21_cast_fp16, y = var_1064_cast_fp16)[name = string("op_1065_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_3_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44373760)))];
            tensor<fp16, [1, 375, 768]> input_53_cast_fp16 = mul(x = core_backbone_layers_3_operator_norm_weight_to_fp16, y = var_1065_cast_fp16)[name = string("input_53_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_3_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44375360)))];
            tensor<fp16, [1, 375, 2304]> linear_17_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_3_conv_in_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<int32, [3]> bcx_5_perm_0 = const()[name = string("bcx_5_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_11_begin_0 = const()[name = string("b_11_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_11_end_0 = const()[name = string("b_11_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_11_end_mask_0 = const()[name = string("b_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_5_cast_fp16 = transpose(perm = bcx_5_perm_0, x = linear_17_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [1, 768, 375]> b_11_cast_fp16 = slice_by_index(begin = b_11_begin_0, end = b_11_end_0, end_mask = b_11_end_mask_0, x = bcx_5_cast_fp16)[name = string("b_11_cast_fp16")];
            tensor<int32, [3]> c_5_begin_0 = const()[name = string("c_5_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_5_end_0 = const()[name = string("c_5_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_5_end_mask_0 = const()[name = string("c_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_5_cast_fp16 = slice_by_index(begin = c_5_begin_0, end = c_5_end_0, end_mask = c_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("c_5_cast_fp16")];
            tensor<int32, [3]> xx_5_begin_0 = const()[name = string("xx_5_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_5_end_0 = const()[name = string("xx_5_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_5_end_mask_0 = const()[name = string("xx_5_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_5_cast_fp16 = slice_by_index(begin = xx_5_begin_0, end = xx_5_end_0, end_mask = xx_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("xx_5_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_9_cast_fp16 = mul(x = b_11_cast_fp16, y = xx_5_cast_fp16)[name = string("new_9_cast_fp16")];
            tensor<int32, [4]> input_57_begin_0 = const()[name = string("input_57_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])];
            tensor<int32, [4]> input_57_end_0 = const()[name = string("input_57_end_0"), val = tensor<int32, [4]>([3, 1, 768, 2])];
            tensor<bool, [4]> input_57_end_mask_0 = const()[name = string("input_57_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_57_squeeze_mask_0 = const()[name = string("input_57_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_57_cast_fp16 = slice_by_index(begin = input_57_begin_0, end = input_57_end_0, end_mask = input_57_end_mask_0, squeeze_mask = input_57_squeeze_mask_0, x = conv)[name = string("input_57_cast_fp16")];
            tensor<int32, [6]> var_1080_pad_0 = const()[name = string("op_1080_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1080_mode_0 = const()[name = string("op_1080_mode_0"), val = string("constant")];
            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1080_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = var_1080_mode_0, pad = var_1080_pad_0, x = new_9_cast_fp16)[name = string("op_1080_cast_fp16")];
            tensor<int32, [6]> var_1082_pad_0 = const()[name = string("op_1082_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_1082_mode_0 = const()[name = string("op_1082_mode_0"), val = string("constant")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1082_cast_fp16 = pad(constant_val = const_15_to_fp16, mode = var_1082_mode_0, pad = var_1082_pad_0, x = input_57_cast_fp16)[name = string("op_1082_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_9_cast_fp16 = add(x = var_1080_cast_fp16, y = var_1082_cast_fp16)[name = string("win_9_cast_fp16")];
            tensor<int32, [3]> var_1084_begin_0 = const()[name = string("op_1084_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_1084_end_0 = const()[name = string("op_1084_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_1084_end_mask_0 = const()[name = string("op_1084_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1084_cast_fp16 = slice_by_index(begin = var_1084_begin_0, end = var_1084_end_0, end_mask = var_1084_end_mask_0, x = win_9_cast_fp16)[name = string("op_1084_cast_fp16")];
            string var_1090_pad_type_0 = const()[name = string("op_1090_pad_type_0"), val = string("valid")];
            int32 var_1090_groups_0 = const()[name = string("op_1090_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1090_strides_0 = const()[name = string("op_1090_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1090_pad_0 = const()[name = string("op_1090_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1090_dilations_0 = const()[name = string("op_1090_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_3_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47914368)))];
            tensor<fp16, [1, 768, 375]> var_1090_cast_fp16 = conv(dilations = var_1090_dilations_0, groups = var_1090_groups_0, pad = var_1090_pad_0, pad_type = var_1090_pad_type_0, strides = var_1090_strides_0, weight = core_backbone_layers_3_conv_conv_weight_to_fp16, x = win_9_cast_fp16)[name = string("op_1090_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_1092_cast_fp16 = mul(x = c_5_cast_fp16, y = var_1090_cast_fp16)[name = string("op_1092_cast_fp16")];
            tensor<int32, [3]> input_59_perm_0 = const()[name = string("input_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_3_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47919040)))];
            tensor<fp16, [1, 375, 768]> input_59_cast_fp16 = transpose(perm = input_59_perm_0, x = var_1092_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [1, 375, 768]> linear_18_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_conv_out_proj_weight_to_fp16, x = input_59_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_41_cast_fp16 = add(x = x_39_cast_fp16, y = linear_18_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1098_cast_fp16 = abs(x = x_41_cast_fp16)[name = string("op_1098_cast_fp16")];
            tensor<int32, [1]> var_1100_axes_0 = const()[name = string("op_1100_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1100_cast_fp16 = reduce_max(axes = var_1100_axes_0, keep_dims = var_128, x = var_1098_cast_fp16)[name = string("op_1100_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_21_cast_fp16 = maximum(x = var_1100_cast_fp16, y = var_129_to_fp16)[name = string("s_21_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_25_cast_fp16 = real_div(x = x_41_cast_fp16, y = s_21_cast_fp16)[name = string("y_25_cast_fp16")];
            fp16 var_130_promoted_9_to_fp16 = const()[name = string("op_130_promoted_9_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1103_cast_fp16 = pow(x = y_25_cast_fp16, y = var_130_promoted_9_to_fp16)[name = string("op_1103_cast_fp16")];
            tensor<int32, [1]> var_1105_axes_0 = const()[name = string("op_1105_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1105_keep_dims_0 = const()[name = string("op_1105_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1105_cast_fp16 = reduce_mean(axes = var_1105_axes_0, keep_dims = var_1105_keep_dims_0, x = var_1103_cast_fp16)[name = string("op_1105_cast_fp16")];
            fp32 var_1106_epsilon_0 = const()[name = string("op_1106_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1106_cast_fp16 = inverse(epsilon = var_1106_epsilon_0, x = s_21_cast_fp16)[name = string("op_1106_cast_fp16")];
            fp16 var_1107_to_fp16 = const()[name = string("op_1107_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1108_cast_fp16 = mul(x = var_1106_cast_fp16, y = var_1107_to_fp16)[name = string("op_1108_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1109_cast_fp16 = real_div(x = var_1108_cast_fp16, y = s_21_cast_fp16)[name = string("op_1109_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1110_cast_fp16 = add(x = var_1105_cast_fp16, y = var_1109_cast_fp16)[name = string("op_1110_cast_fp16")];
            fp32 var_1111_epsilon_0 = const()[name = string("op_1111_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1111_cast_fp16 = rsqrt(epsilon = var_1111_epsilon_0, x = var_1110_cast_fp16)[name = string("op_1111_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1112_cast_fp16 = mul(x = y_25_cast_fp16, y = var_1111_cast_fp16)[name = string("op_1112_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_3_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49098752)))];
            tensor<fp16, [1, 375, 768]> input_61_cast_fp16 = mul(x = core_backbone_layers_3_ffn_norm_weight_to_fp16, y = var_1112_cast_fp16)[name = string("input_61_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_3_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49100352)))];
            tensor<fp16, [1, 375, 1792]> linear_19_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w1_weight_to_fp16, x = input_61_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_3_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51852928)))];
            tensor<fp16, [1, 375, 1792]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w3_weight_to_fp16, x = input_61_cast_fp16)[name = string("linear_20_cast_fp16")];
            fp16 var_1121_to_fp16 = const()[name = string("op_1121_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1122_cast_fp16 = mul(x = linear_19_cast_fp16, y = var_1121_to_fp16)[name = string("op_1122_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1125_cast_fp16 = tanh(x = var_1122_cast_fp16)[name = string("op_1125_cast_fp16")];
            fp16 var_1126_to_fp16 = const()[name = string("op_1126_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1127_cast_fp16 = add(x = var_1125_cast_fp16, y = var_1126_to_fp16)[name = string("op_1127_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1128_cast_fp16 = mul(x = var_1122_cast_fp16, y = var_1127_cast_fp16)[name = string("op_1128_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_63_cast_fp16 = mul(x = var_1128_cast_fp16, y = linear_20_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_3_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54605504)))];
            tensor<fp16, [1, 375, 768]> linear_21_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w2_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_45_cast_fp16 = add(x = x_41_cast_fp16, y = linear_21_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1134_cast_fp16 = abs(x = x_45_cast_fp16)[name = string("op_1134_cast_fp16")];
            tensor<int32, [1]> var_1136_axes_0 = const()[name = string("op_1136_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1136_cast_fp16 = reduce_max(axes = var_1136_axes_0, keep_dims = var_128, x = var_1134_cast_fp16)[name = string("op_1136_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_23_cast_fp16 = maximum(x = var_1136_cast_fp16, y = var_129_to_fp16)[name = string("s_23_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_27_cast_fp16 = real_div(x = x_45_cast_fp16, y = s_23_cast_fp16)[name = string("y_27_cast_fp16")];
            fp16 var_130_promoted_10_to_fp16 = const()[name = string("op_130_promoted_10_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1139_cast_fp16 = pow(x = y_27_cast_fp16, y = var_130_promoted_10_to_fp16)[name = string("op_1139_cast_fp16")];
            tensor<int32, [1]> var_1141_axes_0 = const()[name = string("op_1141_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1141_keep_dims_0 = const()[name = string("op_1141_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1141_cast_fp16 = reduce_mean(axes = var_1141_axes_0, keep_dims = var_1141_keep_dims_0, x = var_1139_cast_fp16)[name = string("op_1141_cast_fp16")];
            fp32 var_1142_epsilon_0 = const()[name = string("op_1142_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1142_cast_fp16 = inverse(epsilon = var_1142_epsilon_0, x = s_23_cast_fp16)[name = string("op_1142_cast_fp16")];
            fp16 var_1143_to_fp16 = const()[name = string("op_1143_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1144_cast_fp16 = mul(x = var_1142_cast_fp16, y = var_1143_to_fp16)[name = string("op_1144_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1145_cast_fp16 = real_div(x = var_1144_cast_fp16, y = s_23_cast_fp16)[name = string("op_1145_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1146_cast_fp16 = add(x = var_1141_cast_fp16, y = var_1145_cast_fp16)[name = string("op_1146_cast_fp16")];
            fp32 var_1147_epsilon_0 = const()[name = string("op_1147_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1147_cast_fp16 = rsqrt(epsilon = var_1147_epsilon_0, x = var_1146_cast_fp16)[name = string("op_1147_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1148_cast_fp16 = mul(x = y_27_cast_fp16, y = var_1147_cast_fp16)[name = string("op_1148_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_4_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57358080)))];
            tensor<fp16, [1, 375, 768]> input_65_cast_fp16 = mul(x = core_backbone_layers_4_operator_norm_weight_to_fp16, y = var_1148_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_4_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57359680)))];
            tensor<fp16, [1, 375, 2304]> linear_22_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_4_conv_in_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [3]> bcx_7_perm_0 = const()[name = string("bcx_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_15_begin_0 = const()[name = string("b_15_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_15_end_0 = const()[name = string("b_15_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_15_end_mask_0 = const()[name = string("b_15_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_7_cast_fp16 = transpose(perm = bcx_7_perm_0, x = linear_22_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [1, 768, 375]> b_15_cast_fp16 = slice_by_index(begin = b_15_begin_0, end = b_15_end_0, end_mask = b_15_end_mask_0, x = bcx_7_cast_fp16)[name = string("b_15_cast_fp16")];
            tensor<int32, [3]> c_7_begin_0 = const()[name = string("c_7_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_7_end_0 = const()[name = string("c_7_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_7_end_mask_0 = const()[name = string("c_7_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_7_cast_fp16 = slice_by_index(begin = c_7_begin_0, end = c_7_end_0, end_mask = c_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("c_7_cast_fp16")];
            tensor<int32, [3]> xx_7_begin_0 = const()[name = string("xx_7_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_7_end_0 = const()[name = string("xx_7_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_7_end_mask_0 = const()[name = string("xx_7_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_7_cast_fp16 = slice_by_index(begin = xx_7_begin_0, end = xx_7_end_0, end_mask = xx_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("xx_7_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_11_cast_fp16 = mul(x = b_15_cast_fp16, y = xx_7_cast_fp16)[name = string("new_11_cast_fp16")];
            tensor<int32, [4]> input_69_begin_0 = const()[name = string("input_69_begin_0"), val = tensor<int32, [4]>([3, 0, 0, 0])];
            tensor<int32, [4]> input_69_end_0 = const()[name = string("input_69_end_0"), val = tensor<int32, [4]>([4, 1, 768, 2])];
            tensor<bool, [4]> input_69_end_mask_0 = const()[name = string("input_69_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_69_squeeze_mask_0 = const()[name = string("input_69_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_69_cast_fp16 = slice_by_index(begin = input_69_begin_0, end = input_69_end_0, end_mask = input_69_end_mask_0, squeeze_mask = input_69_squeeze_mask_0, x = conv)[name = string("input_69_cast_fp16")];
            tensor<int32, [6]> var_1163_pad_0 = const()[name = string("op_1163_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1163_mode_0 = const()[name = string("op_1163_mode_0"), val = string("constant")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1163_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = var_1163_mode_0, pad = var_1163_pad_0, x = new_11_cast_fp16)[name = string("op_1163_cast_fp16")];
            tensor<int32, [6]> var_1165_pad_0 = const()[name = string("op_1165_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_1165_mode_0 = const()[name = string("op_1165_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1165_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = var_1165_mode_0, pad = var_1165_pad_0, x = input_69_cast_fp16)[name = string("op_1165_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_13_cast_fp16 = add(x = var_1163_cast_fp16, y = var_1165_cast_fp16)[name = string("win_13_cast_fp16")];
            tensor<int32, [3]> var_1167_begin_0 = const()[name = string("op_1167_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_1167_end_0 = const()[name = string("op_1167_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_1167_end_mask_0 = const()[name = string("op_1167_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1167_cast_fp16 = slice_by_index(begin = var_1167_begin_0, end = var_1167_end_0, end_mask = var_1167_end_mask_0, x = win_13_cast_fp16)[name = string("op_1167_cast_fp16")];
            string var_1173_pad_type_0 = const()[name = string("op_1173_pad_type_0"), val = string("valid")];
            int32 var_1173_groups_0 = const()[name = string("op_1173_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1173_strides_0 = const()[name = string("op_1173_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1173_pad_0 = const()[name = string("op_1173_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1173_dilations_0 = const()[name = string("op_1173_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_4_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60898688)))];
            tensor<fp16, [1, 768, 375]> var_1173_cast_fp16 = conv(dilations = var_1173_dilations_0, groups = var_1173_groups_0, pad = var_1173_pad_0, pad_type = var_1173_pad_type_0, strides = var_1173_strides_0, weight = core_backbone_layers_4_conv_conv_weight_to_fp16, x = win_13_cast_fp16)[name = string("op_1173_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_1175_cast_fp16 = mul(x = c_7_cast_fp16, y = var_1173_cast_fp16)[name = string("op_1175_cast_fp16")];
            tensor<int32, [3]> input_71_perm_0 = const()[name = string("input_71_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_4_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60903360)))];
            tensor<fp16, [1, 375, 768]> input_71_cast_fp16 = transpose(perm = input_71_perm_0, x = var_1175_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [1, 375, 768]> linear_23_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_4_conv_out_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_47_cast_fp16 = add(x = x_45_cast_fp16, y = linear_23_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1181_cast_fp16 = abs(x = x_47_cast_fp16)[name = string("op_1181_cast_fp16")];
            tensor<int32, [1]> var_1183_axes_0 = const()[name = string("op_1183_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1183_cast_fp16 = reduce_max(axes = var_1183_axes_0, keep_dims = var_128, x = var_1181_cast_fp16)[name = string("op_1183_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_25_cast_fp16 = maximum(x = var_1183_cast_fp16, y = var_129_to_fp16)[name = string("s_25_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_31_cast_fp16 = real_div(x = x_47_cast_fp16, y = s_25_cast_fp16)[name = string("y_31_cast_fp16")];
            fp16 var_130_promoted_11_to_fp16 = const()[name = string("op_130_promoted_11_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1186_cast_fp16 = pow(x = y_31_cast_fp16, y = var_130_promoted_11_to_fp16)[name = string("op_1186_cast_fp16")];
            tensor<int32, [1]> var_1188_axes_0 = const()[name = string("op_1188_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1188_keep_dims_0 = const()[name = string("op_1188_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1188_cast_fp16 = reduce_mean(axes = var_1188_axes_0, keep_dims = var_1188_keep_dims_0, x = var_1186_cast_fp16)[name = string("op_1188_cast_fp16")];
            fp32 var_1189_epsilon_0 = const()[name = string("op_1189_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1189_cast_fp16 = inverse(epsilon = var_1189_epsilon_0, x = s_25_cast_fp16)[name = string("op_1189_cast_fp16")];
            fp16 var_1190_to_fp16 = const()[name = string("op_1190_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1191_cast_fp16 = mul(x = var_1189_cast_fp16, y = var_1190_to_fp16)[name = string("op_1191_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1192_cast_fp16 = real_div(x = var_1191_cast_fp16, y = s_25_cast_fp16)[name = string("op_1192_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1193_cast_fp16 = add(x = var_1188_cast_fp16, y = var_1192_cast_fp16)[name = string("op_1193_cast_fp16")];
            fp32 var_1194_epsilon_0 = const()[name = string("op_1194_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1194_cast_fp16 = rsqrt(epsilon = var_1194_epsilon_0, x = var_1193_cast_fp16)[name = string("op_1194_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1195_cast_fp16 = mul(x = y_31_cast_fp16, y = var_1194_cast_fp16)[name = string("op_1195_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_4_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62083072)))];
            tensor<fp16, [1, 375, 768]> input_73_cast_fp16 = mul(x = core_backbone_layers_4_ffn_norm_weight_to_fp16, y = var_1195_cast_fp16)[name = string("input_73_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_4_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62084672)))];
            tensor<fp16, [1, 375, 1792]> linear_24_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w1_weight_to_fp16, x = input_73_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_4_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64837248)))];
            tensor<fp16, [1, 375, 1792]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w3_weight_to_fp16, x = input_73_cast_fp16)[name = string("linear_25_cast_fp16")];
            fp16 var_1204_to_fp16 = const()[name = string("op_1204_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1205_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_1204_to_fp16)[name = string("op_1205_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1208_cast_fp16 = tanh(x = var_1205_cast_fp16)[name = string("op_1208_cast_fp16")];
            fp16 var_1209_to_fp16 = const()[name = string("op_1209_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1210_cast_fp16 = add(x = var_1208_cast_fp16, y = var_1209_to_fp16)[name = string("op_1210_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1211_cast_fp16 = mul(x = var_1205_cast_fp16, y = var_1210_cast_fp16)[name = string("op_1211_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_75_cast_fp16 = mul(x = var_1211_cast_fp16, y = linear_25_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_4_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67589824)))];
            tensor<fp16, [1, 375, 768]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w2_weight_to_fp16, x = input_75_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_51_cast_fp16 = add(x = x_47_cast_fp16, y = linear_26_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1217_cast_fp16 = abs(x = x_51_cast_fp16)[name = string("op_1217_cast_fp16")];
            tensor<int32, [1]> var_1219_axes_0 = const()[name = string("op_1219_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1219_cast_fp16 = reduce_max(axes = var_1219_axes_0, keep_dims = var_128, x = var_1217_cast_fp16)[name = string("op_1219_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_27_cast_fp16 = maximum(x = var_1219_cast_fp16, y = var_129_to_fp16)[name = string("s_27_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_33_cast_fp16 = real_div(x = x_51_cast_fp16, y = s_27_cast_fp16)[name = string("y_33_cast_fp16")];
            fp16 var_130_promoted_12_to_fp16 = const()[name = string("op_130_promoted_12_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1222_cast_fp16 = pow(x = y_33_cast_fp16, y = var_130_promoted_12_to_fp16)[name = string("op_1222_cast_fp16")];
            tensor<int32, [1]> var_1224_axes_0 = const()[name = string("op_1224_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1224_keep_dims_0 = const()[name = string("op_1224_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1224_cast_fp16 = reduce_mean(axes = var_1224_axes_0, keep_dims = var_1224_keep_dims_0, x = var_1222_cast_fp16)[name = string("op_1224_cast_fp16")];
            fp32 var_1225_epsilon_0 = const()[name = string("op_1225_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1225_cast_fp16 = inverse(epsilon = var_1225_epsilon_0, x = s_27_cast_fp16)[name = string("op_1225_cast_fp16")];
            fp16 var_1226_to_fp16 = const()[name = string("op_1226_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1227_cast_fp16 = mul(x = var_1225_cast_fp16, y = var_1226_to_fp16)[name = string("op_1227_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1228_cast_fp16 = real_div(x = var_1227_cast_fp16, y = s_27_cast_fp16)[name = string("op_1228_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1229_cast_fp16 = add(x = var_1224_cast_fp16, y = var_1228_cast_fp16)[name = string("op_1229_cast_fp16")];
            fp32 var_1230_epsilon_0 = const()[name = string("op_1230_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1230_cast_fp16 = rsqrt(epsilon = var_1230_epsilon_0, x = var_1229_cast_fp16)[name = string("op_1230_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1231_cast_fp16 = mul(x = y_33_cast_fp16, y = var_1230_cast_fp16)[name = string("op_1231_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_5_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70342400)))];
            tensor<fp16, [1, 375, 768]> input_77_cast_fp16 = mul(x = core_backbone_layers_5_operator_norm_weight_to_fp16, y = var_1231_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70344000)))];
            tensor<fp16, [1, 375, 768]> linear_27_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_q_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_27_cast_fp16")];
            tensor<int32, [4]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [4]>([1, 375, 8, 96])];
            tensor<fp16, [1, 375, 8, 96]> x_53_cast_fp16 = reshape(shape = concat_9, x = linear_27_cast_fp16)[name = string("x_53_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1238_cast_fp16 = abs(x = x_53_cast_fp16)[name = string("op_1238_cast_fp16")];
            tensor<int32, [1]> var_1240_axes_0 = const()[name = string("op_1240_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 8, 1]> var_1240_cast_fp16 = reduce_max(axes = var_1240_axes_0, keep_dims = var_128, x = var_1238_cast_fp16)[name = string("op_1240_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> s_29_cast_fp16 = maximum(x = var_1240_cast_fp16, y = var_129_to_fp16)[name = string("s_29_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> y_35_cast_fp16 = real_div(x = x_53_cast_fp16, y = s_29_cast_fp16)[name = string("y_35_cast_fp16")];
            fp16 var_130_promoted_13_to_fp16 = const()[name = string("op_130_promoted_13_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 8, 96]> var_1243_cast_fp16 = pow(x = y_35_cast_fp16, y = var_130_promoted_13_to_fp16)[name = string("op_1243_cast_fp16")];
            tensor<int32, [1]> var_1245_axes_0 = const()[name = string("op_1245_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1245_keep_dims_0 = const()[name = string("op_1245_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 8, 1]> var_1245_cast_fp16 = reduce_mean(axes = var_1245_axes_0, keep_dims = var_1245_keep_dims_0, x = var_1243_cast_fp16)[name = string("op_1245_cast_fp16")];
            fp32 var_1246_epsilon_0 = const()[name = string("op_1246_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 8, 1]> var_1246_cast_fp16 = inverse(epsilon = var_1246_epsilon_0, x = s_29_cast_fp16)[name = string("op_1246_cast_fp16")];
            fp16 var_1247_to_fp16 = const()[name = string("op_1247_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 8, 1]> var_1248_cast_fp16 = mul(x = var_1246_cast_fp16, y = var_1247_to_fp16)[name = string("op_1248_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_1249_cast_fp16 = real_div(x = var_1248_cast_fp16, y = s_29_cast_fp16)[name = string("op_1249_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_1250_cast_fp16 = add(x = var_1245_cast_fp16, y = var_1249_cast_fp16)[name = string("op_1250_cast_fp16")];
            fp32 var_1251_epsilon_0 = const()[name = string("op_1251_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 8, 1]> var_1251_cast_fp16 = rsqrt(epsilon = var_1251_epsilon_0, x = var_1250_cast_fp16)[name = string("op_1251_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1252_cast_fp16 = mul(x = y_35_cast_fp16, y = var_1251_cast_fp16)[name = string("op_1252_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_5_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71523712)))];
            tensor<fp16, [1, 375, 8, 96]> var_1253_cast_fp16 = mul(x = core_backbone_layers_5_self_attn_q_norm_weight_to_fp16, y = var_1252_cast_fp16)[name = string("op_1253_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71523968)))];
            tensor<fp16, [1, 375, 384]> linear_28_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_k_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [4]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> x_55_cast_fp16 = reshape(shape = concat_10, x = linear_28_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_1260_cast_fp16 = abs(x = x_55_cast_fp16)[name = string("op_1260_cast_fp16")];
            tensor<int32, [1]> var_1262_axes_0 = const()[name = string("op_1262_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 4, 1]> var_1262_cast_fp16 = reduce_max(axes = var_1262_axes_0, keep_dims = var_128, x = var_1260_cast_fp16)[name = string("op_1262_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> s_31_cast_fp16 = maximum(x = var_1262_cast_fp16, y = var_129_to_fp16)[name = string("s_31_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> y_37_cast_fp16 = real_div(x = x_55_cast_fp16, y = s_31_cast_fp16)[name = string("y_37_cast_fp16")];
            fp16 var_130_promoted_14_to_fp16 = const()[name = string("op_130_promoted_14_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 4, 96]> var_1265_cast_fp16 = pow(x = y_37_cast_fp16, y = var_130_promoted_14_to_fp16)[name = string("op_1265_cast_fp16")];
            tensor<int32, [1]> var_1267_axes_0 = const()[name = string("op_1267_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1267_keep_dims_0 = const()[name = string("op_1267_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 4, 1]> var_1267_cast_fp16 = reduce_mean(axes = var_1267_axes_0, keep_dims = var_1267_keep_dims_0, x = var_1265_cast_fp16)[name = string("op_1267_cast_fp16")];
            fp32 var_1268_epsilon_0 = const()[name = string("op_1268_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 4, 1]> var_1268_cast_fp16 = inverse(epsilon = var_1268_epsilon_0, x = s_31_cast_fp16)[name = string("op_1268_cast_fp16")];
            fp16 var_1269_to_fp16 = const()[name = string("op_1269_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 4, 1]> var_1270_cast_fp16 = mul(x = var_1268_cast_fp16, y = var_1269_to_fp16)[name = string("op_1270_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_1271_cast_fp16 = real_div(x = var_1270_cast_fp16, y = s_31_cast_fp16)[name = string("op_1271_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_1272_cast_fp16 = add(x = var_1267_cast_fp16, y = var_1271_cast_fp16)[name = string("op_1272_cast_fp16")];
            fp32 var_1273_epsilon_0 = const()[name = string("op_1273_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 4, 1]> var_1273_cast_fp16 = rsqrt(epsilon = var_1273_epsilon_0, x = var_1272_cast_fp16)[name = string("op_1273_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_1274_cast_fp16 = mul(x = y_37_cast_fp16, y = var_1273_cast_fp16)[name = string("op_1274_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_5_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72113856)))];
            tensor<fp16, [1, 375, 4, 96]> var_1275_cast_fp16 = mul(x = core_backbone_layers_5_self_attn_k_norm_weight_to_fp16, y = var_1274_cast_fp16)[name = string("op_1275_cast_fp16")];
            tensor<int32, [4]> new_13_perm_0 = const()[name = string("new_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72114112)))];
            tensor<fp16, [1, 375, 384]> linear_29_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_v_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<int32, [4]> concat_11 = const()[name = string("concat_11"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> var_1280_cast_fp16 = reshape(shape = concat_11, x = linear_29_cast_fp16)[name = string("op_1280_cast_fp16")];
            tensor<int32, [4]> new_15_perm_0 = const()[name = string("new_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_81_begin_0 = const()[name = string("input_81_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_81_end_0 = const()[name = string("input_81_end_0"), val = tensor<int32, [5]>([2, 1, 4, 31, 96])];
            tensor<bool, [5]> input_81_end_mask_0 = const()[name = string("input_81_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_81_squeeze_mask_0 = const()[name = string("input_81_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_81_cast_fp16 = slice_by_index(begin = input_81_begin_0, end = input_81_end_0, end_mask = input_81_end_mask_0, squeeze_mask = input_81_squeeze_mask_0, x = k)[name = string("input_81_cast_fp16")];
            tensor<int32, [8]> var_1285_pad_0 = const()[name = string("op_1285_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1285_mode_0 = const()[name = string("op_1285_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_13_cast_fp16 = transpose(perm = new_13_perm_0, x = var_1275_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [1, 4, 406, 96]> var_1285_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = var_1285_mode_0, pad = var_1285_pad_0, x = new_13_cast_fp16)[name = string("op_1285_cast_fp16")];
            tensor<int32, [8]> var_1287_pad_0 = const()[name = string("op_1287_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_1287_mode_0 = const()[name = string("op_1287_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_1287_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = var_1287_mode_0, pad = var_1287_pad_0, x = input_81_cast_fp16)[name = string("op_1287_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> x_59_cast_fp16 = add(x = var_1285_cast_fp16, y = var_1287_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<int32, [5]> input_85_begin_0 = const()[name = string("input_85_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_85_end_0 = const()[name = string("input_85_end_0"), val = tensor<int32, [5]>([2, 1, 4, 31, 96])];
            tensor<bool, [5]> input_85_end_mask_0 = const()[name = string("input_85_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_85_squeeze_mask_0 = const()[name = string("input_85_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_85_cast_fp16 = slice_by_index(begin = input_85_begin_0, end = input_85_end_0, end_mask = input_85_end_mask_0, squeeze_mask = input_85_squeeze_mask_0, x = v)[name = string("input_85_cast_fp16")];
            tensor<int32, [8]> var_1292_pad_0 = const()[name = string("op_1292_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1292_mode_0 = const()[name = string("op_1292_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_15_cast_fp16 = transpose(perm = new_15_perm_0, x = var_1280_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [1, 4, 406, 96]> var_1292_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = var_1292_mode_0, pad = var_1292_pad_0, x = new_15_cast_fp16)[name = string("op_1292_cast_fp16")];
            tensor<int32, [8]> var_1294_pad_0 = const()[name = string("op_1294_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_1294_mode_0 = const()[name = string("op_1294_mode_0"), val = string("constant")];
            fp16 const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_1294_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = var_1294_mode_0, pad = var_1294_pad_0, x = input_85_cast_fp16)[name = string("op_1294_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> v_7_cast_fp16 = add(x = var_1292_cast_fp16, y = var_1294_cast_fp16)[name = string("v_7_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1296_cast_fp16 = mul(x = var_1253_cast_fp16, y = const_66_to_fp16)[name = string("op_1296_cast_fp16")];
            tensor<int32, [2]> var_1297_split_sizes_0 = const()[name = string("op_1297_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_71 = const()[name = string("const_71"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 48]> var_1297_cast_fp16_0, tensor<fp16, [1, 375, 8, 48]> var_1297_cast_fp16_1 = split(axis = const_71, split_sizes = var_1297_split_sizes_0, x = var_1253_cast_fp16)[name = string("op_1297_cast_fp16")];
            fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 375, 8, 48]> var_1299_cast_fp16 = mul(x = var_1297_cast_fp16_1, y = const_22_promoted_to_fp16)[name = string("op_1299_cast_fp16")];
            bool var_1301_interleave_0 = const()[name = string("op_1301_interleave_0"), val = bool(false)];
            int32 const_72 = const()[name = string("const_72"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 96]> var_1301_cast_fp16 = concat(axis = const_72, interleave = var_1301_interleave_0, values = (var_1299_cast_fp16, var_1297_cast_fp16_0))[name = string("op_1301_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1302_cast_fp16 = mul(x = var_1301_cast_fp16, y = const_69_to_fp16)[name = string("op_1302_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> q_3_cast_fp16 = add(x = var_1296_cast_fp16, y = var_1302_cast_fp16)[name = string("q_3_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_1304_cast_fp16 = mul(x = x_59_cast_fp16, y = const_0_to_fp16)[name = string("op_1304_cast_fp16")];
            tensor<int32, [2]> var_1305_split_sizes_0 = const()[name = string("op_1305_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1305_axis_0 = const()[name = string("op_1305_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 406, 48]> var_1305_cast_fp16_0, tensor<fp16, [1, 4, 406, 48]> var_1305_cast_fp16_1 = split(axis = var_1305_axis_0, split_sizes = var_1305_split_sizes_0, x = x_59_cast_fp16)[name = string("op_1305_cast_fp16")];
            fp16 const_23_promoted_to_fp16 = const()[name = string("const_23_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 406, 48]> var_1307_cast_fp16 = mul(x = var_1305_cast_fp16_1, y = const_23_promoted_to_fp16)[name = string("op_1307_cast_fp16")];
            bool var_1309_interleave_0 = const()[name = string("op_1309_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 406, 96]> var_1309_cast_fp16 = concat(axis = var_127, interleave = var_1309_interleave_0, values = (var_1307_cast_fp16, var_1305_cast_fp16_0))[name = string("op_1309_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_1310_cast_fp16 = mul(x = var_1309_cast_fp16, y = const_1_to_fp16)[name = string("op_1310_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> k_7_cast_fp16 = add(x = var_1304_cast_fp16, y = var_1310_cast_fp16)[name = string("k_7_cast_fp16")];
            tensor<int32, [1]> var_1312_axes_0 = const()[name = string("op_1312_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_1312_cast_fp16 = expand_dims(axes = var_1312_axes_0, x = k_7_cast_fp16)[name = string("op_1312_cast_fp16")];
            tensor<int32, [5]> var_1314_reps_0 = const()[name = string("op_1314_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_1314_cast_fp16 = tile(reps = var_1314_reps_0, x = var_1312_cast_fp16)[name = string("op_1314_cast_fp16")];
            tensor<int32, [4]> concat_13 = const()[name = string("concat_13"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> k_9_cast_fp16 = reshape(shape = concat_13, x = var_1314_cast_fp16)[name = string("k_9_cast_fp16")];
            tensor<int32, [1]> var_1316_axes_0 = const()[name = string("op_1316_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_1316_cast_fp16 = expand_dims(axes = var_1316_axes_0, x = v_7_cast_fp16)[name = string("op_1316_cast_fp16")];
            tensor<int32, [5]> var_1318_reps_0 = const()[name = string("op_1318_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_1318_cast_fp16 = tile(reps = var_1318_reps_0, x = var_1316_cast_fp16)[name = string("op_1318_cast_fp16")];
            tensor<int32, [4]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> v_9_cast_fp16 = reshape(shape = concat_15, x = var_1318_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> transpose_2_perm_0 = const()[name = string("transpose_2_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1321_transpose_x_1 = const()[name = string("op_1321_transpose_x_1"), val = bool(false)];
            bool var_1321_transpose_y_1 = const()[name = string("op_1321_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 96]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = q_3_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [1, 8, 375, 406]> var_1321_cast_fp16 = matmul(transpose_x = var_1321_transpose_x_1, transpose_y = var_1321_transpose_y_1, x = transpose_2_cast_fp16, y = k_9_cast_fp16)[name = string("op_1321_cast_fp16")];
            fp16 var_1322_to_fp16 = const()[name = string("op_1322_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 375, 406]> s_33_cast_fp16 = mul(x = var_1321_cast_fp16, y = var_1322_to_fp16)[name = string("s_33_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_61_cast_fp16 = add(x = s_33_cast_fp16, y = mask)[name = string("x_61_cast_fp16")];
            tensor<int32, [1]> var_1326_axes_0 = const()[name = string("op_1326_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 375, 1]> var_1326_cast_fp16 = reduce_max(axes = var_1326_axes_0, keep_dims = var_128, x = x_61_cast_fp16)[name = string("op_1326_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_63_cast_fp16 = sub(x = x_61_cast_fp16, y = var_1326_cast_fp16)[name = string("x_63_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_65_cast_fp16 = exp(x = x_63_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [1]> var_1330_axes_0 = const()[name = string("op_1330_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1330_keep_dims_0 = const()[name = string("op_1330_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 1]> var_1330_cast_fp16 = reduce_sum(axes = var_1330_axes_0, keep_dims = var_1330_keep_dims_0, x = x_65_cast_fp16)[name = string("op_1330_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> var_1331_cast_fp16 = real_div(x = x_65_cast_fp16, y = var_1330_cast_fp16)[name = string("op_1331_cast_fp16")];
            bool o_5_transpose_x_0 = const()[name = string("o_5_transpose_x_0"), val = bool(false)];
            bool o_5_transpose_y_0 = const()[name = string("o_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 375, 96]> o_5_cast_fp16 = matmul(transpose_x = o_5_transpose_x_0, transpose_y = o_5_transpose_y_0, x = var_1331_cast_fp16, y = v_9_cast_fp16)[name = string("o_5_cast_fp16")];
            tensor<int32, [4]> o_7_perm_0 = const()[name = string("o_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [3]>([1, 375, 768])];
            tensor<fp16, [1, 375, 8, 96]> o_7_cast_fp16 = transpose(perm = o_7_perm_0, x = o_5_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [1, 375, 768]> input_87_cast_fp16 = reshape(shape = concat_16, x = o_7_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72704000)))];
            tensor<fp16, [1, 375, 768]> linear_30_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_o_proj_weight_to_fp16, x = input_87_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> var_1340_begin_0 = const()[name = string("op_1340_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_1340_end_0 = const()[name = string("op_1340_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_1340_end_mask_0 = const()[name = string("op_1340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1340_cast_fp16 = slice_by_index(begin = var_1340_begin_0, end = var_1340_end_0, end_mask = var_1340_end_mask_0, x = x_59_cast_fp16)[name = string("op_1340_cast_fp16")];
            tensor<int32, [4]> var_1343_begin_0 = const()[name = string("op_1343_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_1343_end_0 = const()[name = string("op_1343_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_1343_end_mask_0 = const()[name = string("op_1343_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1343_cast_fp16 = slice_by_index(begin = var_1343_begin_0, end = var_1343_end_0, end_mask = var_1343_end_mask_0, x = v_7_cast_fp16)[name = string("op_1343_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_67_cast_fp16 = add(x = x_51_cast_fp16, y = linear_30_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1346_cast_fp16 = abs(x = x_67_cast_fp16)[name = string("op_1346_cast_fp16")];
            tensor<int32, [1]> var_1348_axes_0 = const()[name = string("op_1348_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1348_cast_fp16 = reduce_max(axes = var_1348_axes_0, keep_dims = var_128, x = var_1346_cast_fp16)[name = string("op_1348_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_35_cast_fp16 = maximum(x = var_1348_cast_fp16, y = var_129_to_fp16)[name = string("s_35_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_39_cast_fp16 = real_div(x = x_67_cast_fp16, y = s_35_cast_fp16)[name = string("y_39_cast_fp16")];
            fp16 var_130_promoted_15_to_fp16 = const()[name = string("op_130_promoted_15_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1351_cast_fp16 = pow(x = y_39_cast_fp16, y = var_130_promoted_15_to_fp16)[name = string("op_1351_cast_fp16")];
            tensor<int32, [1]> var_1353_axes_0 = const()[name = string("op_1353_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1353_keep_dims_0 = const()[name = string("op_1353_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1353_cast_fp16 = reduce_mean(axes = var_1353_axes_0, keep_dims = var_1353_keep_dims_0, x = var_1351_cast_fp16)[name = string("op_1353_cast_fp16")];
            fp32 var_1354_epsilon_0 = const()[name = string("op_1354_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1354_cast_fp16 = inverse(epsilon = var_1354_epsilon_0, x = s_35_cast_fp16)[name = string("op_1354_cast_fp16")];
            fp16 var_1355_to_fp16 = const()[name = string("op_1355_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1356_cast_fp16 = mul(x = var_1354_cast_fp16, y = var_1355_to_fp16)[name = string("op_1356_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1357_cast_fp16 = real_div(x = var_1356_cast_fp16, y = s_35_cast_fp16)[name = string("op_1357_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1358_cast_fp16 = add(x = var_1353_cast_fp16, y = var_1357_cast_fp16)[name = string("op_1358_cast_fp16")];
            fp32 var_1359_epsilon_0 = const()[name = string("op_1359_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1359_cast_fp16 = rsqrt(epsilon = var_1359_epsilon_0, x = var_1358_cast_fp16)[name = string("op_1359_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1360_cast_fp16 = mul(x = y_39_cast_fp16, y = var_1359_cast_fp16)[name = string("op_1360_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_5_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73883712)))];
            tensor<fp16, [1, 375, 768]> input_89_cast_fp16 = mul(x = core_backbone_layers_5_ffn_norm_weight_to_fp16, y = var_1360_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_5_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73885312)))];
            tensor<fp16, [1, 375, 1792]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w1_weight_to_fp16, x = input_89_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_5_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76637888)))];
            tensor<fp16, [1, 375, 1792]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w3_weight_to_fp16, x = input_89_cast_fp16)[name = string("linear_32_cast_fp16")];
            fp16 var_1369_to_fp16 = const()[name = string("op_1369_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1370_cast_fp16 = mul(x = linear_31_cast_fp16, y = var_1369_to_fp16)[name = string("op_1370_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1373_cast_fp16 = tanh(x = var_1370_cast_fp16)[name = string("op_1373_cast_fp16")];
            fp16 var_1374_to_fp16 = const()[name = string("op_1374_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1375_cast_fp16 = add(x = var_1373_cast_fp16, y = var_1374_to_fp16)[name = string("op_1375_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1376_cast_fp16 = mul(x = var_1370_cast_fp16, y = var_1375_cast_fp16)[name = string("op_1376_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_91_cast_fp16 = mul(x = var_1376_cast_fp16, y = linear_32_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_5_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79390464)))];
            tensor<fp16, [1, 375, 768]> linear_33_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w2_weight_to_fp16, x = input_91_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_71_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1382_cast_fp16 = abs(x = x_71_cast_fp16)[name = string("op_1382_cast_fp16")];
            tensor<int32, [1]> var_1384_axes_0 = const()[name = string("op_1384_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1384_cast_fp16 = reduce_max(axes = var_1384_axes_0, keep_dims = var_128, x = var_1382_cast_fp16)[name = string("op_1384_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_37_cast_fp16 = maximum(x = var_1384_cast_fp16, y = var_129_to_fp16)[name = string("s_37_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_41_cast_fp16 = real_div(x = x_71_cast_fp16, y = s_37_cast_fp16)[name = string("y_41_cast_fp16")];
            fp16 var_130_promoted_16_to_fp16 = const()[name = string("op_130_promoted_16_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1387_cast_fp16 = pow(x = y_41_cast_fp16, y = var_130_promoted_16_to_fp16)[name = string("op_1387_cast_fp16")];
            tensor<int32, [1]> var_1389_axes_0 = const()[name = string("op_1389_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1389_keep_dims_0 = const()[name = string("op_1389_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1389_cast_fp16 = reduce_mean(axes = var_1389_axes_0, keep_dims = var_1389_keep_dims_0, x = var_1387_cast_fp16)[name = string("op_1389_cast_fp16")];
            fp32 var_1390_epsilon_0 = const()[name = string("op_1390_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1390_cast_fp16 = inverse(epsilon = var_1390_epsilon_0, x = s_37_cast_fp16)[name = string("op_1390_cast_fp16")];
            fp16 var_1391_to_fp16 = const()[name = string("op_1391_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1392_cast_fp16 = mul(x = var_1390_cast_fp16, y = var_1391_to_fp16)[name = string("op_1392_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1393_cast_fp16 = real_div(x = var_1392_cast_fp16, y = s_37_cast_fp16)[name = string("op_1393_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1394_cast_fp16 = add(x = var_1389_cast_fp16, y = var_1393_cast_fp16)[name = string("op_1394_cast_fp16")];
            fp32 var_1395_epsilon_0 = const()[name = string("op_1395_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1395_cast_fp16 = rsqrt(epsilon = var_1395_epsilon_0, x = var_1394_cast_fp16)[name = string("op_1395_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1396_cast_fp16 = mul(x = y_41_cast_fp16, y = var_1395_cast_fp16)[name = string("op_1396_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_6_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82143040)))];
            tensor<fp16, [1, 375, 768]> input_93_cast_fp16 = mul(x = core_backbone_layers_6_operator_norm_weight_to_fp16, y = var_1396_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_6_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82144640)))];
            tensor<fp16, [1, 375, 2304]> linear_34_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_6_conv_in_proj_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<int32, [3]> bcx_9_perm_0 = const()[name = string("bcx_9_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_21_begin_0 = const()[name = string("b_21_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_21_end_0 = const()[name = string("b_21_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_21_end_mask_0 = const()[name = string("b_21_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_9_cast_fp16 = transpose(perm = bcx_9_perm_0, x = linear_34_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [1, 768, 375]> b_21_cast_fp16 = slice_by_index(begin = b_21_begin_0, end = b_21_end_0, end_mask = b_21_end_mask_0, x = bcx_9_cast_fp16)[name = string("b_21_cast_fp16")];
            tensor<int32, [3]> c_9_begin_0 = const()[name = string("c_9_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_9_end_0 = const()[name = string("c_9_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_9_end_mask_0 = const()[name = string("c_9_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_9_cast_fp16 = slice_by_index(begin = c_9_begin_0, end = c_9_end_0, end_mask = c_9_end_mask_0, x = bcx_9_cast_fp16)[name = string("c_9_cast_fp16")];
            tensor<int32, [3]> xx_9_begin_0 = const()[name = string("xx_9_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_9_end_0 = const()[name = string("xx_9_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_9_end_mask_0 = const()[name = string("xx_9_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_9_cast_fp16 = slice_by_index(begin = xx_9_begin_0, end = xx_9_end_0, end_mask = xx_9_end_mask_0, x = bcx_9_cast_fp16)[name = string("xx_9_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_17_cast_fp16 = mul(x = b_21_cast_fp16, y = xx_9_cast_fp16)[name = string("new_17_cast_fp16")];
            tensor<int32, [4]> input_97_begin_0 = const()[name = string("input_97_begin_0"), val = tensor<int32, [4]>([4, 0, 0, 0])];
            tensor<int32, [4]> input_97_end_0 = const()[name = string("input_97_end_0"), val = tensor<int32, [4]>([5, 1, 768, 2])];
            tensor<bool, [4]> input_97_end_mask_0 = const()[name = string("input_97_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_97_squeeze_mask_0 = const()[name = string("input_97_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_97_cast_fp16 = slice_by_index(begin = input_97_begin_0, end = input_97_end_0, end_mask = input_97_end_mask_0, squeeze_mask = input_97_squeeze_mask_0, x = conv)[name = string("input_97_cast_fp16")];
            tensor<int32, [6]> var_1411_pad_0 = const()[name = string("op_1411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1411_mode_0 = const()[name = string("op_1411_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1411_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_1411_mode_0, pad = var_1411_pad_0, x = new_17_cast_fp16)[name = string("op_1411_cast_fp16")];
            tensor<int32, [6]> var_1413_pad_0 = const()[name = string("op_1413_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_1413_mode_0 = const()[name = string("op_1413_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1413_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = var_1413_mode_0, pad = var_1413_pad_0, x = input_97_cast_fp16)[name = string("op_1413_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_17_cast_fp16 = add(x = var_1411_cast_fp16, y = var_1413_cast_fp16)[name = string("win_17_cast_fp16")];
            tensor<int32, [3]> var_1415_begin_0 = const()[name = string("op_1415_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_1415_end_0 = const()[name = string("op_1415_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_1415_end_mask_0 = const()[name = string("op_1415_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1415_cast_fp16 = slice_by_index(begin = var_1415_begin_0, end = var_1415_end_0, end_mask = var_1415_end_mask_0, x = win_17_cast_fp16)[name = string("op_1415_cast_fp16")];
            string var_1421_pad_type_0 = const()[name = string("op_1421_pad_type_0"), val = string("valid")];
            int32 var_1421_groups_0 = const()[name = string("op_1421_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1421_strides_0 = const()[name = string("op_1421_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1421_pad_0 = const()[name = string("op_1421_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1421_dilations_0 = const()[name = string("op_1421_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_6_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85683648)))];
            tensor<fp16, [1, 768, 375]> var_1421_cast_fp16 = conv(dilations = var_1421_dilations_0, groups = var_1421_groups_0, pad = var_1421_pad_0, pad_type = var_1421_pad_type_0, strides = var_1421_strides_0, weight = core_backbone_layers_6_conv_conv_weight_to_fp16, x = win_17_cast_fp16)[name = string("op_1421_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_1423_cast_fp16 = mul(x = c_9_cast_fp16, y = var_1421_cast_fp16)[name = string("op_1423_cast_fp16")];
            tensor<int32, [3]> input_99_perm_0 = const()[name = string("input_99_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_6_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85688320)))];
            tensor<fp16, [1, 375, 768]> input_99_cast_fp16 = transpose(perm = input_99_perm_0, x = var_1423_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [1, 375, 768]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_6_conv_out_proj_weight_to_fp16, x = input_99_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_73_cast_fp16 = add(x = x_71_cast_fp16, y = linear_35_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1429_cast_fp16 = abs(x = x_73_cast_fp16)[name = string("op_1429_cast_fp16")];
            tensor<int32, [1]> var_1431_axes_0 = const()[name = string("op_1431_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1431_cast_fp16 = reduce_max(axes = var_1431_axes_0, keep_dims = var_128, x = var_1429_cast_fp16)[name = string("op_1431_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_39_cast_fp16 = maximum(x = var_1431_cast_fp16, y = var_129_to_fp16)[name = string("s_39_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_45_cast_fp16 = real_div(x = x_73_cast_fp16, y = s_39_cast_fp16)[name = string("y_45_cast_fp16")];
            fp16 var_130_promoted_17_to_fp16 = const()[name = string("op_130_promoted_17_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1434_cast_fp16 = pow(x = y_45_cast_fp16, y = var_130_promoted_17_to_fp16)[name = string("op_1434_cast_fp16")];
            tensor<int32, [1]> var_1436_axes_0 = const()[name = string("op_1436_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1436_keep_dims_0 = const()[name = string("op_1436_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1436_cast_fp16 = reduce_mean(axes = var_1436_axes_0, keep_dims = var_1436_keep_dims_0, x = var_1434_cast_fp16)[name = string("op_1436_cast_fp16")];
            fp32 var_1437_epsilon_0 = const()[name = string("op_1437_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1437_cast_fp16 = inverse(epsilon = var_1437_epsilon_0, x = s_39_cast_fp16)[name = string("op_1437_cast_fp16")];
            fp16 var_1438_to_fp16 = const()[name = string("op_1438_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1439_cast_fp16 = mul(x = var_1437_cast_fp16, y = var_1438_to_fp16)[name = string("op_1439_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1440_cast_fp16 = real_div(x = var_1439_cast_fp16, y = s_39_cast_fp16)[name = string("op_1440_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1441_cast_fp16 = add(x = var_1436_cast_fp16, y = var_1440_cast_fp16)[name = string("op_1441_cast_fp16")];
            fp32 var_1442_epsilon_0 = const()[name = string("op_1442_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1442_cast_fp16 = rsqrt(epsilon = var_1442_epsilon_0, x = var_1441_cast_fp16)[name = string("op_1442_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1443_cast_fp16 = mul(x = y_45_cast_fp16, y = var_1442_cast_fp16)[name = string("op_1443_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_6_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86868032)))];
            tensor<fp16, [1, 375, 768]> input_101_cast_fp16 = mul(x = core_backbone_layers_6_ffn_norm_weight_to_fp16, y = var_1443_cast_fp16)[name = string("input_101_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_6_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86869632)))];
            tensor<fp16, [1, 375, 1792]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w1_weight_to_fp16, x = input_101_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_6_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89622208)))];
            tensor<fp16, [1, 375, 1792]> linear_37_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w3_weight_to_fp16, x = input_101_cast_fp16)[name = string("linear_37_cast_fp16")];
            fp16 var_1452_to_fp16 = const()[name = string("op_1452_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1453_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_1452_to_fp16)[name = string("op_1453_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1456_cast_fp16 = tanh(x = var_1453_cast_fp16)[name = string("op_1456_cast_fp16")];
            fp16 var_1457_to_fp16 = const()[name = string("op_1457_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1458_cast_fp16 = add(x = var_1456_cast_fp16, y = var_1457_to_fp16)[name = string("op_1458_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1459_cast_fp16 = mul(x = var_1453_cast_fp16, y = var_1458_cast_fp16)[name = string("op_1459_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_103_cast_fp16 = mul(x = var_1459_cast_fp16, y = linear_37_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_6_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92374784)))];
            tensor<fp16, [1, 375, 768]> linear_38_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w2_weight_to_fp16, x = input_103_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_77_cast_fp16 = add(x = x_73_cast_fp16, y = linear_38_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1465_cast_fp16 = abs(x = x_77_cast_fp16)[name = string("op_1465_cast_fp16")];
            tensor<int32, [1]> var_1467_axes_0 = const()[name = string("op_1467_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1467_cast_fp16 = reduce_max(axes = var_1467_axes_0, keep_dims = var_128, x = var_1465_cast_fp16)[name = string("op_1467_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_41_cast_fp16 = maximum(x = var_1467_cast_fp16, y = var_129_to_fp16)[name = string("s_41_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_47_cast_fp16 = real_div(x = x_77_cast_fp16, y = s_41_cast_fp16)[name = string("y_47_cast_fp16")];
            fp16 var_130_promoted_18_to_fp16 = const()[name = string("op_130_promoted_18_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1470_cast_fp16 = pow(x = y_47_cast_fp16, y = var_130_promoted_18_to_fp16)[name = string("op_1470_cast_fp16")];
            tensor<int32, [1]> var_1472_axes_0 = const()[name = string("op_1472_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1472_keep_dims_0 = const()[name = string("op_1472_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1472_cast_fp16 = reduce_mean(axes = var_1472_axes_0, keep_dims = var_1472_keep_dims_0, x = var_1470_cast_fp16)[name = string("op_1472_cast_fp16")];
            fp32 var_1473_epsilon_0 = const()[name = string("op_1473_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1473_cast_fp16 = inverse(epsilon = var_1473_epsilon_0, x = s_41_cast_fp16)[name = string("op_1473_cast_fp16")];
            fp16 var_1474_to_fp16 = const()[name = string("op_1474_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1475_cast_fp16 = mul(x = var_1473_cast_fp16, y = var_1474_to_fp16)[name = string("op_1475_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1476_cast_fp16 = real_div(x = var_1475_cast_fp16, y = s_41_cast_fp16)[name = string("op_1476_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1477_cast_fp16 = add(x = var_1472_cast_fp16, y = var_1476_cast_fp16)[name = string("op_1477_cast_fp16")];
            fp32 var_1478_epsilon_0 = const()[name = string("op_1478_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1478_cast_fp16 = rsqrt(epsilon = var_1478_epsilon_0, x = var_1477_cast_fp16)[name = string("op_1478_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1479_cast_fp16 = mul(x = y_47_cast_fp16, y = var_1478_cast_fp16)[name = string("op_1479_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_7_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95127360)))];
            tensor<fp16, [1, 375, 768]> input_105_cast_fp16 = mul(x = core_backbone_layers_7_operator_norm_weight_to_fp16, y = var_1479_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_7_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95128960)))];
            tensor<fp16, [1, 375, 2304]> linear_39_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_7_conv_in_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [3]> bcx_11_perm_0 = const()[name = string("bcx_11_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_25_begin_0 = const()[name = string("b_25_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_25_end_0 = const()[name = string("b_25_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_25_end_mask_0 = const()[name = string("b_25_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_11_cast_fp16 = transpose(perm = bcx_11_perm_0, x = linear_39_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [1, 768, 375]> b_25_cast_fp16 = slice_by_index(begin = b_25_begin_0, end = b_25_end_0, end_mask = b_25_end_mask_0, x = bcx_11_cast_fp16)[name = string("b_25_cast_fp16")];
            tensor<int32, [3]> c_11_begin_0 = const()[name = string("c_11_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_11_end_0 = const()[name = string("c_11_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_11_end_mask_0 = const()[name = string("c_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_11_cast_fp16 = slice_by_index(begin = c_11_begin_0, end = c_11_end_0, end_mask = c_11_end_mask_0, x = bcx_11_cast_fp16)[name = string("c_11_cast_fp16")];
            tensor<int32, [3]> xx_11_begin_0 = const()[name = string("xx_11_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_11_end_0 = const()[name = string("xx_11_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_11_end_mask_0 = const()[name = string("xx_11_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_11_cast_fp16 = slice_by_index(begin = xx_11_begin_0, end = xx_11_end_0, end_mask = xx_11_end_mask_0, x = bcx_11_cast_fp16)[name = string("xx_11_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_19_cast_fp16 = mul(x = b_25_cast_fp16, y = xx_11_cast_fp16)[name = string("new_19_cast_fp16")];
            tensor<int32, [4]> input_109_begin_0 = const()[name = string("input_109_begin_0"), val = tensor<int32, [4]>([5, 0, 0, 0])];
            tensor<int32, [4]> input_109_end_0 = const()[name = string("input_109_end_0"), val = tensor<int32, [4]>([6, 1, 768, 2])];
            tensor<bool, [4]> input_109_end_mask_0 = const()[name = string("input_109_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_109_squeeze_mask_0 = const()[name = string("input_109_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_109_cast_fp16 = slice_by_index(begin = input_109_begin_0, end = input_109_end_0, end_mask = input_109_end_mask_0, squeeze_mask = input_109_squeeze_mask_0, x = conv)[name = string("input_109_cast_fp16")];
            tensor<int32, [6]> var_1494_pad_0 = const()[name = string("op_1494_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1494_mode_0 = const()[name = string("op_1494_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1494_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = var_1494_mode_0, pad = var_1494_pad_0, x = new_19_cast_fp16)[name = string("op_1494_cast_fp16")];
            tensor<int32, [6]> var_1496_pad_0 = const()[name = string("op_1496_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_1496_mode_0 = const()[name = string("op_1496_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1496_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_1496_mode_0, pad = var_1496_pad_0, x = input_109_cast_fp16)[name = string("op_1496_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_21_cast_fp16 = add(x = var_1494_cast_fp16, y = var_1496_cast_fp16)[name = string("win_21_cast_fp16")];
            tensor<int32, [3]> var_1498_begin_0 = const()[name = string("op_1498_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_1498_end_0 = const()[name = string("op_1498_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_1498_end_mask_0 = const()[name = string("op_1498_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1498_cast_fp16 = slice_by_index(begin = var_1498_begin_0, end = var_1498_end_0, end_mask = var_1498_end_mask_0, x = win_21_cast_fp16)[name = string("op_1498_cast_fp16")];
            string var_1504_pad_type_0 = const()[name = string("op_1504_pad_type_0"), val = string("valid")];
            int32 var_1504_groups_0 = const()[name = string("op_1504_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1504_strides_0 = const()[name = string("op_1504_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1504_pad_0 = const()[name = string("op_1504_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1504_dilations_0 = const()[name = string("op_1504_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_7_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98667968)))];
            tensor<fp16, [1, 768, 375]> var_1504_cast_fp16 = conv(dilations = var_1504_dilations_0, groups = var_1504_groups_0, pad = var_1504_pad_0, pad_type = var_1504_pad_type_0, strides = var_1504_strides_0, weight = core_backbone_layers_7_conv_conv_weight_to_fp16, x = win_21_cast_fp16)[name = string("op_1504_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_1506_cast_fp16 = mul(x = c_11_cast_fp16, y = var_1504_cast_fp16)[name = string("op_1506_cast_fp16")];
            tensor<int32, [3]> input_111_perm_0 = const()[name = string("input_111_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_7_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98672640)))];
            tensor<fp16, [1, 375, 768]> input_111_cast_fp16 = transpose(perm = input_111_perm_0, x = var_1506_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [1, 375, 768]> linear_40_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_conv_out_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_79_cast_fp16 = add(x = x_77_cast_fp16, y = linear_40_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1512_cast_fp16 = abs(x = x_79_cast_fp16)[name = string("op_1512_cast_fp16")];
            tensor<int32, [1]> var_1514_axes_0 = const()[name = string("op_1514_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1514_cast_fp16 = reduce_max(axes = var_1514_axes_0, keep_dims = var_128, x = var_1512_cast_fp16)[name = string("op_1514_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_43_cast_fp16 = maximum(x = var_1514_cast_fp16, y = var_129_to_fp16)[name = string("s_43_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_51_cast_fp16 = real_div(x = x_79_cast_fp16, y = s_43_cast_fp16)[name = string("y_51_cast_fp16")];
            fp16 var_130_promoted_19_to_fp16 = const()[name = string("op_130_promoted_19_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1517_cast_fp16 = pow(x = y_51_cast_fp16, y = var_130_promoted_19_to_fp16)[name = string("op_1517_cast_fp16")];
            tensor<int32, [1]> var_1519_axes_0 = const()[name = string("op_1519_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1519_keep_dims_0 = const()[name = string("op_1519_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1519_cast_fp16 = reduce_mean(axes = var_1519_axes_0, keep_dims = var_1519_keep_dims_0, x = var_1517_cast_fp16)[name = string("op_1519_cast_fp16")];
            fp32 var_1520_epsilon_0 = const()[name = string("op_1520_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1520_cast_fp16 = inverse(epsilon = var_1520_epsilon_0, x = s_43_cast_fp16)[name = string("op_1520_cast_fp16")];
            fp16 var_1521_to_fp16 = const()[name = string("op_1521_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1522_cast_fp16 = mul(x = var_1520_cast_fp16, y = var_1521_to_fp16)[name = string("op_1522_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1523_cast_fp16 = real_div(x = var_1522_cast_fp16, y = s_43_cast_fp16)[name = string("op_1523_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1524_cast_fp16 = add(x = var_1519_cast_fp16, y = var_1523_cast_fp16)[name = string("op_1524_cast_fp16")];
            fp32 var_1525_epsilon_0 = const()[name = string("op_1525_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1525_cast_fp16 = rsqrt(epsilon = var_1525_epsilon_0, x = var_1524_cast_fp16)[name = string("op_1525_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1526_cast_fp16 = mul(x = y_51_cast_fp16, y = var_1525_cast_fp16)[name = string("op_1526_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_7_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99852352)))];
            tensor<fp16, [1, 375, 768]> input_113_cast_fp16 = mul(x = core_backbone_layers_7_ffn_norm_weight_to_fp16, y = var_1526_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_7_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99853952)))];
            tensor<fp16, [1, 375, 1792]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w1_weight_to_fp16, x = input_113_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_7_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102606528)))];
            tensor<fp16, [1, 375, 1792]> linear_42_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w3_weight_to_fp16, x = input_113_cast_fp16)[name = string("linear_42_cast_fp16")];
            fp16 var_1535_to_fp16 = const()[name = string("op_1535_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1536_cast_fp16 = mul(x = linear_41_cast_fp16, y = var_1535_to_fp16)[name = string("op_1536_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1539_cast_fp16 = tanh(x = var_1536_cast_fp16)[name = string("op_1539_cast_fp16")];
            fp16 var_1540_to_fp16 = const()[name = string("op_1540_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1541_cast_fp16 = add(x = var_1539_cast_fp16, y = var_1540_to_fp16)[name = string("op_1541_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1542_cast_fp16 = mul(x = var_1536_cast_fp16, y = var_1541_cast_fp16)[name = string("op_1542_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_115_cast_fp16 = mul(x = var_1542_cast_fp16, y = linear_42_cast_fp16)[name = string("input_115_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_7_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105359104)))];
            tensor<fp16, [1, 375, 768]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w2_weight_to_fp16, x = input_115_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_83_cast_fp16 = add(x = x_79_cast_fp16, y = linear_43_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1548_cast_fp16 = abs(x = x_83_cast_fp16)[name = string("op_1548_cast_fp16")];
            tensor<int32, [1]> var_1550_axes_0 = const()[name = string("op_1550_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1550_cast_fp16 = reduce_max(axes = var_1550_axes_0, keep_dims = var_128, x = var_1548_cast_fp16)[name = string("op_1550_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_45_cast_fp16 = maximum(x = var_1550_cast_fp16, y = var_129_to_fp16)[name = string("s_45_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_53_cast_fp16 = real_div(x = x_83_cast_fp16, y = s_45_cast_fp16)[name = string("y_53_cast_fp16")];
            fp16 var_130_promoted_20_to_fp16 = const()[name = string("op_130_promoted_20_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1553_cast_fp16 = pow(x = y_53_cast_fp16, y = var_130_promoted_20_to_fp16)[name = string("op_1553_cast_fp16")];
            tensor<int32, [1]> var_1555_axes_0 = const()[name = string("op_1555_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1555_keep_dims_0 = const()[name = string("op_1555_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1555_cast_fp16 = reduce_mean(axes = var_1555_axes_0, keep_dims = var_1555_keep_dims_0, x = var_1553_cast_fp16)[name = string("op_1555_cast_fp16")];
            fp32 var_1556_epsilon_0 = const()[name = string("op_1556_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1556_cast_fp16 = inverse(epsilon = var_1556_epsilon_0, x = s_45_cast_fp16)[name = string("op_1556_cast_fp16")];
            fp16 var_1557_to_fp16 = const()[name = string("op_1557_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1558_cast_fp16 = mul(x = var_1556_cast_fp16, y = var_1557_to_fp16)[name = string("op_1558_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1559_cast_fp16 = real_div(x = var_1558_cast_fp16, y = s_45_cast_fp16)[name = string("op_1559_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1560_cast_fp16 = add(x = var_1555_cast_fp16, y = var_1559_cast_fp16)[name = string("op_1560_cast_fp16")];
            fp32 var_1561_epsilon_0 = const()[name = string("op_1561_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1561_cast_fp16 = rsqrt(epsilon = var_1561_epsilon_0, x = var_1560_cast_fp16)[name = string("op_1561_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1562_cast_fp16 = mul(x = y_53_cast_fp16, y = var_1561_cast_fp16)[name = string("op_1562_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_8_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108111680)))];
            tensor<fp16, [1, 375, 768]> input_117_cast_fp16 = mul(x = core_backbone_layers_8_operator_norm_weight_to_fp16, y = var_1562_cast_fp16)[name = string("input_117_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108113280)))];
            tensor<fp16, [1, 375, 768]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_q_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<int32, [4]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [4]>([1, 375, 8, 96])];
            tensor<fp16, [1, 375, 8, 96]> x_85_cast_fp16 = reshape(shape = concat_17, x = linear_44_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1569_cast_fp16 = abs(x = x_85_cast_fp16)[name = string("op_1569_cast_fp16")];
            tensor<int32, [1]> var_1571_axes_0 = const()[name = string("op_1571_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 8, 1]> var_1571_cast_fp16 = reduce_max(axes = var_1571_axes_0, keep_dims = var_128, x = var_1569_cast_fp16)[name = string("op_1571_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> s_47_cast_fp16 = maximum(x = var_1571_cast_fp16, y = var_129_to_fp16)[name = string("s_47_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> y_55_cast_fp16 = real_div(x = x_85_cast_fp16, y = s_47_cast_fp16)[name = string("y_55_cast_fp16")];
            fp16 var_130_promoted_21_to_fp16 = const()[name = string("op_130_promoted_21_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 8, 96]> var_1574_cast_fp16 = pow(x = y_55_cast_fp16, y = var_130_promoted_21_to_fp16)[name = string("op_1574_cast_fp16")];
            tensor<int32, [1]> var_1576_axes_0 = const()[name = string("op_1576_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1576_keep_dims_0 = const()[name = string("op_1576_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 8, 1]> var_1576_cast_fp16 = reduce_mean(axes = var_1576_axes_0, keep_dims = var_1576_keep_dims_0, x = var_1574_cast_fp16)[name = string("op_1576_cast_fp16")];
            fp32 var_1577_epsilon_0 = const()[name = string("op_1577_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 8, 1]> var_1577_cast_fp16 = inverse(epsilon = var_1577_epsilon_0, x = s_47_cast_fp16)[name = string("op_1577_cast_fp16")];
            fp16 var_1578_to_fp16 = const()[name = string("op_1578_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 8, 1]> var_1579_cast_fp16 = mul(x = var_1577_cast_fp16, y = var_1578_to_fp16)[name = string("op_1579_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_1580_cast_fp16 = real_div(x = var_1579_cast_fp16, y = s_47_cast_fp16)[name = string("op_1580_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_1581_cast_fp16 = add(x = var_1576_cast_fp16, y = var_1580_cast_fp16)[name = string("op_1581_cast_fp16")];
            fp32 var_1582_epsilon_0 = const()[name = string("op_1582_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 8, 1]> var_1582_cast_fp16 = rsqrt(epsilon = var_1582_epsilon_0, x = var_1581_cast_fp16)[name = string("op_1582_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1583_cast_fp16 = mul(x = y_55_cast_fp16, y = var_1582_cast_fp16)[name = string("op_1583_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_8_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109292992)))];
            tensor<fp16, [1, 375, 8, 96]> var_1584_cast_fp16 = mul(x = core_backbone_layers_8_self_attn_q_norm_weight_to_fp16, y = var_1583_cast_fp16)[name = string("op_1584_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109293248)))];
            tensor<fp16, [1, 375, 384]> linear_45_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_k_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<int32, [4]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> x_87_cast_fp16 = reshape(shape = concat_18, x = linear_45_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_1591_cast_fp16 = abs(x = x_87_cast_fp16)[name = string("op_1591_cast_fp16")];
            tensor<int32, [1]> var_1593_axes_0 = const()[name = string("op_1593_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 4, 1]> var_1593_cast_fp16 = reduce_max(axes = var_1593_axes_0, keep_dims = var_128, x = var_1591_cast_fp16)[name = string("op_1593_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> s_49_cast_fp16 = maximum(x = var_1593_cast_fp16, y = var_129_to_fp16)[name = string("s_49_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> y_57_cast_fp16 = real_div(x = x_87_cast_fp16, y = s_49_cast_fp16)[name = string("y_57_cast_fp16")];
            fp16 var_130_promoted_22_to_fp16 = const()[name = string("op_130_promoted_22_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 4, 96]> var_1596_cast_fp16 = pow(x = y_57_cast_fp16, y = var_130_promoted_22_to_fp16)[name = string("op_1596_cast_fp16")];
            tensor<int32, [1]> var_1598_axes_0 = const()[name = string("op_1598_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1598_keep_dims_0 = const()[name = string("op_1598_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 4, 1]> var_1598_cast_fp16 = reduce_mean(axes = var_1598_axes_0, keep_dims = var_1598_keep_dims_0, x = var_1596_cast_fp16)[name = string("op_1598_cast_fp16")];
            fp32 var_1599_epsilon_0 = const()[name = string("op_1599_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 4, 1]> var_1599_cast_fp16 = inverse(epsilon = var_1599_epsilon_0, x = s_49_cast_fp16)[name = string("op_1599_cast_fp16")];
            fp16 var_1600_to_fp16 = const()[name = string("op_1600_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 4, 1]> var_1601_cast_fp16 = mul(x = var_1599_cast_fp16, y = var_1600_to_fp16)[name = string("op_1601_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_1602_cast_fp16 = real_div(x = var_1601_cast_fp16, y = s_49_cast_fp16)[name = string("op_1602_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_1603_cast_fp16 = add(x = var_1598_cast_fp16, y = var_1602_cast_fp16)[name = string("op_1603_cast_fp16")];
            fp32 var_1604_epsilon_0 = const()[name = string("op_1604_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 4, 1]> var_1604_cast_fp16 = rsqrt(epsilon = var_1604_epsilon_0, x = var_1603_cast_fp16)[name = string("op_1604_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_1605_cast_fp16 = mul(x = y_57_cast_fp16, y = var_1604_cast_fp16)[name = string("op_1605_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_8_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109883136)))];
            tensor<fp16, [1, 375, 4, 96]> var_1606_cast_fp16 = mul(x = core_backbone_layers_8_self_attn_k_norm_weight_to_fp16, y = var_1605_cast_fp16)[name = string("op_1606_cast_fp16")];
            tensor<int32, [4]> new_21_perm_0 = const()[name = string("new_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109883392)))];
            tensor<fp16, [1, 375, 384]> linear_46_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_v_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<int32, [4]> concat_19 = const()[name = string("concat_19"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> var_1611_cast_fp16 = reshape(shape = concat_19, x = linear_46_cast_fp16)[name = string("op_1611_cast_fp16")];
            tensor<int32, [4]> new_23_perm_0 = const()[name = string("new_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_121_begin_0 = const()[name = string("input_121_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_121_end_0 = const()[name = string("input_121_end_0"), val = tensor<int32, [5]>([3, 1, 4, 31, 96])];
            tensor<bool, [5]> input_121_end_mask_0 = const()[name = string("input_121_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_121_squeeze_mask_0 = const()[name = string("input_121_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_121_cast_fp16 = slice_by_index(begin = input_121_begin_0, end = input_121_end_0, end_mask = input_121_end_mask_0, squeeze_mask = input_121_squeeze_mask_0, x = k)[name = string("input_121_cast_fp16")];
            tensor<int32, [8]> var_1616_pad_0 = const()[name = string("op_1616_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1616_mode_0 = const()[name = string("op_1616_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_21_cast_fp16 = transpose(perm = new_21_perm_0, x = var_1606_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [1, 4, 406, 96]> var_1616_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = var_1616_mode_0, pad = var_1616_pad_0, x = new_21_cast_fp16)[name = string("op_1616_cast_fp16")];
            tensor<int32, [8]> var_1618_pad_0 = const()[name = string("op_1618_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_1618_mode_0 = const()[name = string("op_1618_mode_0"), val = string("constant")];
            fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_1618_cast_fp16 = pad(constant_val = const_29_to_fp16, mode = var_1618_mode_0, pad = var_1618_pad_0, x = input_121_cast_fp16)[name = string("op_1618_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> x_91_cast_fp16 = add(x = var_1616_cast_fp16, y = var_1618_cast_fp16)[name = string("x_91_cast_fp16")];
            tensor<int32, [5]> input_125_begin_0 = const()[name = string("input_125_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_125_end_0 = const()[name = string("input_125_end_0"), val = tensor<int32, [5]>([3, 1, 4, 31, 96])];
            tensor<bool, [5]> input_125_end_mask_0 = const()[name = string("input_125_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_125_squeeze_mask_0 = const()[name = string("input_125_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_125_cast_fp16 = slice_by_index(begin = input_125_begin_0, end = input_125_end_0, end_mask = input_125_end_mask_0, squeeze_mask = input_125_squeeze_mask_0, x = v)[name = string("input_125_cast_fp16")];
            tensor<int32, [8]> var_1623_pad_0 = const()[name = string("op_1623_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1623_mode_0 = const()[name = string("op_1623_mode_0"), val = string("constant")];
            fp16 const_30_to_fp16 = const()[name = string("const_30_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_23_cast_fp16 = transpose(perm = new_23_perm_0, x = var_1611_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [1, 4, 406, 96]> var_1623_cast_fp16 = pad(constant_val = const_30_to_fp16, mode = var_1623_mode_0, pad = var_1623_pad_0, x = new_23_cast_fp16)[name = string("op_1623_cast_fp16")];
            tensor<int32, [8]> var_1625_pad_0 = const()[name = string("op_1625_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_1625_mode_0 = const()[name = string("op_1625_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_1625_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = var_1625_mode_0, pad = var_1625_pad_0, x = input_125_cast_fp16)[name = string("op_1625_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> v_11_cast_fp16 = add(x = var_1623_cast_fp16, y = var_1625_cast_fp16)[name = string("v_11_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1627_cast_fp16 = mul(x = var_1584_cast_fp16, y = const_66_to_fp16)[name = string("op_1627_cast_fp16")];
            tensor<int32, [2]> var_1628_split_sizes_0 = const()[name = string("op_1628_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_75 = const()[name = string("const_75"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 48]> var_1628_cast_fp16_0, tensor<fp16, [1, 375, 8, 48]> var_1628_cast_fp16_1 = split(axis = const_75, split_sizes = var_1628_split_sizes_0, x = var_1584_cast_fp16)[name = string("op_1628_cast_fp16")];
            fp16 const_32_promoted_to_fp16 = const()[name = string("const_32_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 375, 8, 48]> var_1630_cast_fp16 = mul(x = var_1628_cast_fp16_1, y = const_32_promoted_to_fp16)[name = string("op_1630_cast_fp16")];
            bool var_1632_interleave_0 = const()[name = string("op_1632_interleave_0"), val = bool(false)];
            int32 const_76 = const()[name = string("const_76"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 96]> var_1632_cast_fp16 = concat(axis = const_76, interleave = var_1632_interleave_0, values = (var_1630_cast_fp16, var_1628_cast_fp16_0))[name = string("op_1632_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1633_cast_fp16 = mul(x = var_1632_cast_fp16, y = const_69_to_fp16)[name = string("op_1633_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> q_5_cast_fp16 = add(x = var_1627_cast_fp16, y = var_1633_cast_fp16)[name = string("q_5_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_1635_cast_fp16 = mul(x = x_91_cast_fp16, y = const_0_to_fp16)[name = string("op_1635_cast_fp16")];
            tensor<int32, [2]> var_1636_split_sizes_0 = const()[name = string("op_1636_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1636_axis_0 = const()[name = string("op_1636_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 406, 48]> var_1636_cast_fp16_0, tensor<fp16, [1, 4, 406, 48]> var_1636_cast_fp16_1 = split(axis = var_1636_axis_0, split_sizes = var_1636_split_sizes_0, x = x_91_cast_fp16)[name = string("op_1636_cast_fp16")];
            fp16 const_33_promoted_to_fp16 = const()[name = string("const_33_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 406, 48]> var_1638_cast_fp16 = mul(x = var_1636_cast_fp16_1, y = const_33_promoted_to_fp16)[name = string("op_1638_cast_fp16")];
            bool var_1640_interleave_0 = const()[name = string("op_1640_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 406, 96]> var_1640_cast_fp16 = concat(axis = var_127, interleave = var_1640_interleave_0, values = (var_1638_cast_fp16, var_1636_cast_fp16_0))[name = string("op_1640_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_1641_cast_fp16 = mul(x = var_1640_cast_fp16, y = const_1_to_fp16)[name = string("op_1641_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> k_11_cast_fp16 = add(x = var_1635_cast_fp16, y = var_1641_cast_fp16)[name = string("k_11_cast_fp16")];
            tensor<int32, [1]> var_1643_axes_0 = const()[name = string("op_1643_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_1643_cast_fp16 = expand_dims(axes = var_1643_axes_0, x = k_11_cast_fp16)[name = string("op_1643_cast_fp16")];
            tensor<int32, [5]> var_1645_reps_0 = const()[name = string("op_1645_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_1645_cast_fp16 = tile(reps = var_1645_reps_0, x = var_1643_cast_fp16)[name = string("op_1645_cast_fp16")];
            tensor<int32, [4]> concat_21 = const()[name = string("concat_21"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> k_13_cast_fp16 = reshape(shape = concat_21, x = var_1645_cast_fp16)[name = string("k_13_cast_fp16")];
            tensor<int32, [1]> var_1647_axes_0 = const()[name = string("op_1647_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_1647_cast_fp16 = expand_dims(axes = var_1647_axes_0, x = v_11_cast_fp16)[name = string("op_1647_cast_fp16")];
            tensor<int32, [5]> var_1649_reps_0 = const()[name = string("op_1649_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_1649_cast_fp16 = tile(reps = var_1649_reps_0, x = var_1647_cast_fp16)[name = string("op_1649_cast_fp16")];
            tensor<int32, [4]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> v_13_cast_fp16 = reshape(shape = concat_23, x = var_1649_cast_fp16)[name = string("v_13_cast_fp16")];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = string("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1652_transpose_x_1 = const()[name = string("op_1652_transpose_x_1"), val = bool(false)];
            bool var_1652_transpose_y_1 = const()[name = string("op_1652_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 96]> transpose_4_cast_fp16 = transpose(perm = transpose_4_perm_0, x = q_5_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [1, 8, 375, 406]> var_1652_cast_fp16 = matmul(transpose_x = var_1652_transpose_x_1, transpose_y = var_1652_transpose_y_1, x = transpose_4_cast_fp16, y = k_13_cast_fp16)[name = string("op_1652_cast_fp16")];
            fp16 var_1653_to_fp16 = const()[name = string("op_1653_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 375, 406]> s_51_cast_fp16 = mul(x = var_1652_cast_fp16, y = var_1653_to_fp16)[name = string("s_51_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_93_cast_fp16 = add(x = s_51_cast_fp16, y = mask)[name = string("x_93_cast_fp16")];
            tensor<int32, [1]> var_1657_axes_0 = const()[name = string("op_1657_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 375, 1]> var_1657_cast_fp16 = reduce_max(axes = var_1657_axes_0, keep_dims = var_128, x = x_93_cast_fp16)[name = string("op_1657_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_95_cast_fp16 = sub(x = x_93_cast_fp16, y = var_1657_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_97_cast_fp16 = exp(x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<int32, [1]> var_1661_axes_0 = const()[name = string("op_1661_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1661_keep_dims_0 = const()[name = string("op_1661_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 1]> var_1661_cast_fp16 = reduce_sum(axes = var_1661_axes_0, keep_dims = var_1661_keep_dims_0, x = x_97_cast_fp16)[name = string("op_1661_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> var_1662_cast_fp16 = real_div(x = x_97_cast_fp16, y = var_1661_cast_fp16)[name = string("op_1662_cast_fp16")];
            bool o_9_transpose_x_0 = const()[name = string("o_9_transpose_x_0"), val = bool(false)];
            bool o_9_transpose_y_0 = const()[name = string("o_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 375, 96]> o_9_cast_fp16 = matmul(transpose_x = o_9_transpose_x_0, transpose_y = o_9_transpose_y_0, x = var_1662_cast_fp16, y = v_13_cast_fp16)[name = string("o_9_cast_fp16")];
            tensor<int32, [4]> o_11_perm_0 = const()[name = string("o_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_24 = const()[name = string("concat_24"), val = tensor<int32, [3]>([1, 375, 768])];
            tensor<fp16, [1, 375, 8, 96]> o_11_cast_fp16 = transpose(perm = o_11_perm_0, x = o_9_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [1, 375, 768]> input_127_cast_fp16 = reshape(shape = concat_24, x = o_11_cast_fp16)[name = string("input_127_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110473280)))];
            tensor<fp16, [1, 375, 768]> linear_47_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_o_proj_weight_to_fp16, x = input_127_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<int32, [4]> var_1671_begin_0 = const()[name = string("op_1671_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_1671_end_0 = const()[name = string("op_1671_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_1671_end_mask_0 = const()[name = string("op_1671_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1671_cast_fp16 = slice_by_index(begin = var_1671_begin_0, end = var_1671_end_0, end_mask = var_1671_end_mask_0, x = x_91_cast_fp16)[name = string("op_1671_cast_fp16")];
            tensor<int32, [4]> var_1674_begin_0 = const()[name = string("op_1674_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_1674_end_0 = const()[name = string("op_1674_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_1674_end_mask_0 = const()[name = string("op_1674_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1674_cast_fp16 = slice_by_index(begin = var_1674_begin_0, end = var_1674_end_0, end_mask = var_1674_end_mask_0, x = v_11_cast_fp16)[name = string("op_1674_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_99_cast_fp16 = add(x = x_83_cast_fp16, y = linear_47_cast_fp16)[name = string("x_99_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1677_cast_fp16 = abs(x = x_99_cast_fp16)[name = string("op_1677_cast_fp16")];
            tensor<int32, [1]> var_1679_axes_0 = const()[name = string("op_1679_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1679_cast_fp16 = reduce_max(axes = var_1679_axes_0, keep_dims = var_128, x = var_1677_cast_fp16)[name = string("op_1679_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_53_cast_fp16 = maximum(x = var_1679_cast_fp16, y = var_129_to_fp16)[name = string("s_53_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_59_cast_fp16 = real_div(x = x_99_cast_fp16, y = s_53_cast_fp16)[name = string("y_59_cast_fp16")];
            fp16 var_130_promoted_23_to_fp16 = const()[name = string("op_130_promoted_23_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1682_cast_fp16 = pow(x = y_59_cast_fp16, y = var_130_promoted_23_to_fp16)[name = string("op_1682_cast_fp16")];
            tensor<int32, [1]> var_1684_axes_0 = const()[name = string("op_1684_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1684_keep_dims_0 = const()[name = string("op_1684_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1684_cast_fp16 = reduce_mean(axes = var_1684_axes_0, keep_dims = var_1684_keep_dims_0, x = var_1682_cast_fp16)[name = string("op_1684_cast_fp16")];
            fp32 var_1685_epsilon_0 = const()[name = string("op_1685_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1685_cast_fp16 = inverse(epsilon = var_1685_epsilon_0, x = s_53_cast_fp16)[name = string("op_1685_cast_fp16")];
            fp16 var_1686_to_fp16 = const()[name = string("op_1686_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1687_cast_fp16 = mul(x = var_1685_cast_fp16, y = var_1686_to_fp16)[name = string("op_1687_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1688_cast_fp16 = real_div(x = var_1687_cast_fp16, y = s_53_cast_fp16)[name = string("op_1688_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1689_cast_fp16 = add(x = var_1684_cast_fp16, y = var_1688_cast_fp16)[name = string("op_1689_cast_fp16")];
            fp32 var_1690_epsilon_0 = const()[name = string("op_1690_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1690_cast_fp16 = rsqrt(epsilon = var_1690_epsilon_0, x = var_1689_cast_fp16)[name = string("op_1690_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1691_cast_fp16 = mul(x = y_59_cast_fp16, y = var_1690_cast_fp16)[name = string("op_1691_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_8_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111652992)))];
            tensor<fp16, [1, 375, 768]> input_129_cast_fp16 = mul(x = core_backbone_layers_8_ffn_norm_weight_to_fp16, y = var_1691_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_8_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111654592)))];
            tensor<fp16, [1, 375, 1792]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w1_weight_to_fp16, x = input_129_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_8_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114407168)))];
            tensor<fp16, [1, 375, 1792]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w3_weight_to_fp16, x = input_129_cast_fp16)[name = string("linear_49_cast_fp16")];
            fp16 var_1700_to_fp16 = const()[name = string("op_1700_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1701_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_1700_to_fp16)[name = string("op_1701_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1704_cast_fp16 = tanh(x = var_1701_cast_fp16)[name = string("op_1704_cast_fp16")];
            fp16 var_1705_to_fp16 = const()[name = string("op_1705_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1706_cast_fp16 = add(x = var_1704_cast_fp16, y = var_1705_to_fp16)[name = string("op_1706_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1707_cast_fp16 = mul(x = var_1701_cast_fp16, y = var_1706_cast_fp16)[name = string("op_1707_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_131_cast_fp16 = mul(x = var_1707_cast_fp16, y = linear_49_cast_fp16)[name = string("input_131_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_8_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117159744)))];
            tensor<fp16, [1, 375, 768]> linear_50_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w2_weight_to_fp16, x = input_131_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_103_cast_fp16 = add(x = x_99_cast_fp16, y = linear_50_cast_fp16)[name = string("x_103_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1713_cast_fp16 = abs(x = x_103_cast_fp16)[name = string("op_1713_cast_fp16")];
            tensor<int32, [1]> var_1715_axes_0 = const()[name = string("op_1715_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1715_cast_fp16 = reduce_max(axes = var_1715_axes_0, keep_dims = var_128, x = var_1713_cast_fp16)[name = string("op_1715_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_55_cast_fp16 = maximum(x = var_1715_cast_fp16, y = var_129_to_fp16)[name = string("s_55_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_61_cast_fp16 = real_div(x = x_103_cast_fp16, y = s_55_cast_fp16)[name = string("y_61_cast_fp16")];
            fp16 var_130_promoted_24_to_fp16 = const()[name = string("op_130_promoted_24_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1718_cast_fp16 = pow(x = y_61_cast_fp16, y = var_130_promoted_24_to_fp16)[name = string("op_1718_cast_fp16")];
            tensor<int32, [1]> var_1720_axes_0 = const()[name = string("op_1720_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1720_keep_dims_0 = const()[name = string("op_1720_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1720_cast_fp16 = reduce_mean(axes = var_1720_axes_0, keep_dims = var_1720_keep_dims_0, x = var_1718_cast_fp16)[name = string("op_1720_cast_fp16")];
            fp32 var_1721_epsilon_0 = const()[name = string("op_1721_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1721_cast_fp16 = inverse(epsilon = var_1721_epsilon_0, x = s_55_cast_fp16)[name = string("op_1721_cast_fp16")];
            fp16 var_1722_to_fp16 = const()[name = string("op_1722_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1723_cast_fp16 = mul(x = var_1721_cast_fp16, y = var_1722_to_fp16)[name = string("op_1723_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1724_cast_fp16 = real_div(x = var_1723_cast_fp16, y = s_55_cast_fp16)[name = string("op_1724_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1725_cast_fp16 = add(x = var_1720_cast_fp16, y = var_1724_cast_fp16)[name = string("op_1725_cast_fp16")];
            fp32 var_1726_epsilon_0 = const()[name = string("op_1726_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1726_cast_fp16 = rsqrt(epsilon = var_1726_epsilon_0, x = var_1725_cast_fp16)[name = string("op_1726_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1727_cast_fp16 = mul(x = y_61_cast_fp16, y = var_1726_cast_fp16)[name = string("op_1727_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_9_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_9_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119912320)))];
            tensor<fp16, [1, 375, 768]> input_133_cast_fp16 = mul(x = core_backbone_layers_9_operator_norm_weight_to_fp16, y = var_1727_cast_fp16)[name = string("input_133_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_9_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119913920)))];
            tensor<fp16, [1, 375, 2304]> linear_51_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_9_conv_in_proj_weight_to_fp16, x = input_133_cast_fp16)[name = string("linear_51_cast_fp16")];
            tensor<int32, [3]> bcx_13_perm_0 = const()[name = string("bcx_13_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_31_begin_0 = const()[name = string("b_31_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_31_end_0 = const()[name = string("b_31_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_31_end_mask_0 = const()[name = string("b_31_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_13_cast_fp16 = transpose(perm = bcx_13_perm_0, x = linear_51_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 768, 375]> b_31_cast_fp16 = slice_by_index(begin = b_31_begin_0, end = b_31_end_0, end_mask = b_31_end_mask_0, x = bcx_13_cast_fp16)[name = string("b_31_cast_fp16")];
            tensor<int32, [3]> c_13_begin_0 = const()[name = string("c_13_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_13_end_0 = const()[name = string("c_13_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_13_end_mask_0 = const()[name = string("c_13_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_13_cast_fp16 = slice_by_index(begin = c_13_begin_0, end = c_13_end_0, end_mask = c_13_end_mask_0, x = bcx_13_cast_fp16)[name = string("c_13_cast_fp16")];
            tensor<int32, [3]> xx_13_begin_0 = const()[name = string("xx_13_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_13_end_0 = const()[name = string("xx_13_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_13_end_mask_0 = const()[name = string("xx_13_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_13_cast_fp16 = slice_by_index(begin = xx_13_begin_0, end = xx_13_end_0, end_mask = xx_13_end_mask_0, x = bcx_13_cast_fp16)[name = string("xx_13_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_25_cast_fp16 = mul(x = b_31_cast_fp16, y = xx_13_cast_fp16)[name = string("new_25_cast_fp16")];
            tensor<int32, [4]> input_137_begin_0 = const()[name = string("input_137_begin_0"), val = tensor<int32, [4]>([6, 0, 0, 0])];
            tensor<int32, [4]> input_137_end_0 = const()[name = string("input_137_end_0"), val = tensor<int32, [4]>([7, 1, 768, 2])];
            tensor<bool, [4]> input_137_end_mask_0 = const()[name = string("input_137_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_137_squeeze_mask_0 = const()[name = string("input_137_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_137_cast_fp16 = slice_by_index(begin = input_137_begin_0, end = input_137_end_0, end_mask = input_137_end_mask_0, squeeze_mask = input_137_squeeze_mask_0, x = conv)[name = string("input_137_cast_fp16")];
            tensor<int32, [6]> var_1742_pad_0 = const()[name = string("op_1742_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1742_mode_0 = const()[name = string("op_1742_mode_0"), val = string("constant")];
            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1742_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = var_1742_mode_0, pad = var_1742_pad_0, x = new_25_cast_fp16)[name = string("op_1742_cast_fp16")];
            tensor<int32, [6]> var_1744_pad_0 = const()[name = string("op_1744_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_1744_mode_0 = const()[name = string("op_1744_mode_0"), val = string("constant")];
            fp16 const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1744_cast_fp16 = pad(constant_val = const_35_to_fp16, mode = var_1744_mode_0, pad = var_1744_pad_0, x = input_137_cast_fp16)[name = string("op_1744_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_25_cast_fp16 = add(x = var_1742_cast_fp16, y = var_1744_cast_fp16)[name = string("win_25_cast_fp16")];
            tensor<int32, [3]> var_1746_begin_0 = const()[name = string("op_1746_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_1746_end_0 = const()[name = string("op_1746_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_1746_end_mask_0 = const()[name = string("op_1746_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1746_cast_fp16 = slice_by_index(begin = var_1746_begin_0, end = var_1746_end_0, end_mask = var_1746_end_mask_0, x = win_25_cast_fp16)[name = string("op_1746_cast_fp16")];
            string var_1752_pad_type_0 = const()[name = string("op_1752_pad_type_0"), val = string("valid")];
            int32 var_1752_groups_0 = const()[name = string("op_1752_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1752_strides_0 = const()[name = string("op_1752_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1752_pad_0 = const()[name = string("op_1752_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1752_dilations_0 = const()[name = string("op_1752_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_9_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123452928)))];
            tensor<fp16, [1, 768, 375]> var_1752_cast_fp16 = conv(dilations = var_1752_dilations_0, groups = var_1752_groups_0, pad = var_1752_pad_0, pad_type = var_1752_pad_type_0, strides = var_1752_strides_0, weight = core_backbone_layers_9_conv_conv_weight_to_fp16, x = win_25_cast_fp16)[name = string("op_1752_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_1754_cast_fp16 = mul(x = c_13_cast_fp16, y = var_1752_cast_fp16)[name = string("op_1754_cast_fp16")];
            tensor<int32, [3]> input_139_perm_0 = const()[name = string("input_139_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_9_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123457600)))];
            tensor<fp16, [1, 375, 768]> input_139_cast_fp16 = transpose(perm = input_139_perm_0, x = var_1754_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 375, 768]> linear_52_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_9_conv_out_proj_weight_to_fp16, x = input_139_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_105_cast_fp16 = add(x = x_103_cast_fp16, y = linear_52_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1760_cast_fp16 = abs(x = x_105_cast_fp16)[name = string("op_1760_cast_fp16")];
            tensor<int32, [1]> var_1762_axes_0 = const()[name = string("op_1762_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1762_cast_fp16 = reduce_max(axes = var_1762_axes_0, keep_dims = var_128, x = var_1760_cast_fp16)[name = string("op_1762_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_57_cast_fp16 = maximum(x = var_1762_cast_fp16, y = var_129_to_fp16)[name = string("s_57_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_65_cast_fp16 = real_div(x = x_105_cast_fp16, y = s_57_cast_fp16)[name = string("y_65_cast_fp16")];
            fp16 var_130_promoted_25_to_fp16 = const()[name = string("op_130_promoted_25_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1765_cast_fp16 = pow(x = y_65_cast_fp16, y = var_130_promoted_25_to_fp16)[name = string("op_1765_cast_fp16")];
            tensor<int32, [1]> var_1767_axes_0 = const()[name = string("op_1767_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1767_keep_dims_0 = const()[name = string("op_1767_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1767_cast_fp16 = reduce_mean(axes = var_1767_axes_0, keep_dims = var_1767_keep_dims_0, x = var_1765_cast_fp16)[name = string("op_1767_cast_fp16")];
            fp32 var_1768_epsilon_0 = const()[name = string("op_1768_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1768_cast_fp16 = inverse(epsilon = var_1768_epsilon_0, x = s_57_cast_fp16)[name = string("op_1768_cast_fp16")];
            fp16 var_1769_to_fp16 = const()[name = string("op_1769_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1770_cast_fp16 = mul(x = var_1768_cast_fp16, y = var_1769_to_fp16)[name = string("op_1770_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1771_cast_fp16 = real_div(x = var_1770_cast_fp16, y = s_57_cast_fp16)[name = string("op_1771_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1772_cast_fp16 = add(x = var_1767_cast_fp16, y = var_1771_cast_fp16)[name = string("op_1772_cast_fp16")];
            fp32 var_1773_epsilon_0 = const()[name = string("op_1773_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1773_cast_fp16 = rsqrt(epsilon = var_1773_epsilon_0, x = var_1772_cast_fp16)[name = string("op_1773_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1774_cast_fp16 = mul(x = y_65_cast_fp16, y = var_1773_cast_fp16)[name = string("op_1774_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_9_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_9_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124637312)))];
            tensor<fp16, [1, 375, 768]> input_141_cast_fp16 = mul(x = core_backbone_layers_9_ffn_norm_weight_to_fp16, y = var_1774_cast_fp16)[name = string("input_141_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_9_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124638912)))];
            tensor<fp16, [1, 375, 1792]> linear_53_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w1_weight_to_fp16, x = input_141_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_9_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127391488)))];
            tensor<fp16, [1, 375, 1792]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w3_weight_to_fp16, x = input_141_cast_fp16)[name = string("linear_54_cast_fp16")];
            fp16 var_1783_to_fp16 = const()[name = string("op_1783_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1784_cast_fp16 = mul(x = linear_53_cast_fp16, y = var_1783_to_fp16)[name = string("op_1784_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1787_cast_fp16 = tanh(x = var_1784_cast_fp16)[name = string("op_1787_cast_fp16")];
            fp16 var_1788_to_fp16 = const()[name = string("op_1788_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1789_cast_fp16 = add(x = var_1787_cast_fp16, y = var_1788_to_fp16)[name = string("op_1789_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1790_cast_fp16 = mul(x = var_1784_cast_fp16, y = var_1789_cast_fp16)[name = string("op_1790_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_143_cast_fp16 = mul(x = var_1790_cast_fp16, y = linear_54_cast_fp16)[name = string("input_143_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_9_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130144064)))];
            tensor<fp16, [1, 375, 768]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w2_weight_to_fp16, x = input_143_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_109_cast_fp16 = add(x = x_105_cast_fp16, y = linear_55_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1796_cast_fp16 = abs(x = x_109_cast_fp16)[name = string("op_1796_cast_fp16")];
            tensor<int32, [1]> var_1798_axes_0 = const()[name = string("op_1798_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1798_cast_fp16 = reduce_max(axes = var_1798_axes_0, keep_dims = var_128, x = var_1796_cast_fp16)[name = string("op_1798_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_59_cast_fp16 = maximum(x = var_1798_cast_fp16, y = var_129_to_fp16)[name = string("s_59_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_67_cast_fp16 = real_div(x = x_109_cast_fp16, y = s_59_cast_fp16)[name = string("y_67_cast_fp16")];
            fp16 var_130_promoted_26_to_fp16 = const()[name = string("op_130_promoted_26_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1801_cast_fp16 = pow(x = y_67_cast_fp16, y = var_130_promoted_26_to_fp16)[name = string("op_1801_cast_fp16")];
            tensor<int32, [1]> var_1803_axes_0 = const()[name = string("op_1803_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1803_keep_dims_0 = const()[name = string("op_1803_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1803_cast_fp16 = reduce_mean(axes = var_1803_axes_0, keep_dims = var_1803_keep_dims_0, x = var_1801_cast_fp16)[name = string("op_1803_cast_fp16")];
            fp32 var_1804_epsilon_0 = const()[name = string("op_1804_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1804_cast_fp16 = inverse(epsilon = var_1804_epsilon_0, x = s_59_cast_fp16)[name = string("op_1804_cast_fp16")];
            fp16 var_1805_to_fp16 = const()[name = string("op_1805_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1806_cast_fp16 = mul(x = var_1804_cast_fp16, y = var_1805_to_fp16)[name = string("op_1806_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1807_cast_fp16 = real_div(x = var_1806_cast_fp16, y = s_59_cast_fp16)[name = string("op_1807_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1808_cast_fp16 = add(x = var_1803_cast_fp16, y = var_1807_cast_fp16)[name = string("op_1808_cast_fp16")];
            fp32 var_1809_epsilon_0 = const()[name = string("op_1809_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1809_cast_fp16 = rsqrt(epsilon = var_1809_epsilon_0, x = var_1808_cast_fp16)[name = string("op_1809_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1810_cast_fp16 = mul(x = y_67_cast_fp16, y = var_1809_cast_fp16)[name = string("op_1810_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_10_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_10_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132896640)))];
            tensor<fp16, [1, 375, 768]> input_145_cast_fp16 = mul(x = core_backbone_layers_10_operator_norm_weight_to_fp16, y = var_1810_cast_fp16)[name = string("input_145_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_10_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132898240)))];
            tensor<fp16, [1, 375, 2304]> linear_56_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_10_conv_in_proj_weight_to_fp16, x = input_145_cast_fp16)[name = string("linear_56_cast_fp16")];
            tensor<int32, [3]> bcx_15_perm_0 = const()[name = string("bcx_15_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_35_begin_0 = const()[name = string("b_35_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_35_end_0 = const()[name = string("b_35_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_35_end_mask_0 = const()[name = string("b_35_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_15_cast_fp16 = transpose(perm = bcx_15_perm_0, x = linear_56_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [1, 768, 375]> b_35_cast_fp16 = slice_by_index(begin = b_35_begin_0, end = b_35_end_0, end_mask = b_35_end_mask_0, x = bcx_15_cast_fp16)[name = string("b_35_cast_fp16")];
            tensor<int32, [3]> c_15_begin_0 = const()[name = string("c_15_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_15_end_0 = const()[name = string("c_15_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_15_end_mask_0 = const()[name = string("c_15_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_15_cast_fp16 = slice_by_index(begin = c_15_begin_0, end = c_15_end_0, end_mask = c_15_end_mask_0, x = bcx_15_cast_fp16)[name = string("c_15_cast_fp16")];
            tensor<int32, [3]> xx_15_begin_0 = const()[name = string("xx_15_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_15_end_0 = const()[name = string("xx_15_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_15_end_mask_0 = const()[name = string("xx_15_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_15_cast_fp16 = slice_by_index(begin = xx_15_begin_0, end = xx_15_end_0, end_mask = xx_15_end_mask_0, x = bcx_15_cast_fp16)[name = string("xx_15_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_27_cast_fp16 = mul(x = b_35_cast_fp16, y = xx_15_cast_fp16)[name = string("new_27_cast_fp16")];
            tensor<int32, [4]> input_149_begin_0 = const()[name = string("input_149_begin_0"), val = tensor<int32, [4]>([7, 0, 0, 0])];
            tensor<int32, [4]> input_149_end_0 = const()[name = string("input_149_end_0"), val = tensor<int32, [4]>([8, 1, 768, 2])];
            tensor<bool, [4]> input_149_end_mask_0 = const()[name = string("input_149_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_149_squeeze_mask_0 = const()[name = string("input_149_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_149_cast_fp16 = slice_by_index(begin = input_149_begin_0, end = input_149_end_0, end_mask = input_149_end_mask_0, squeeze_mask = input_149_squeeze_mask_0, x = conv)[name = string("input_149_cast_fp16")];
            tensor<int32, [6]> var_1825_pad_0 = const()[name = string("op_1825_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1825_mode_0 = const()[name = string("op_1825_mode_0"), val = string("constant")];
            fp16 const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1825_cast_fp16 = pad(constant_val = const_36_to_fp16, mode = var_1825_mode_0, pad = var_1825_pad_0, x = new_27_cast_fp16)[name = string("op_1825_cast_fp16")];
            tensor<int32, [6]> var_1827_pad_0 = const()[name = string("op_1827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_1827_mode_0 = const()[name = string("op_1827_mode_0"), val = string("constant")];
            fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_1827_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = var_1827_mode_0, pad = var_1827_pad_0, x = input_149_cast_fp16)[name = string("op_1827_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_29_cast_fp16 = add(x = var_1825_cast_fp16, y = var_1827_cast_fp16)[name = string("win_29_cast_fp16")];
            tensor<int32, [3]> var_1829_begin_0 = const()[name = string("op_1829_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_1829_end_0 = const()[name = string("op_1829_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_1829_end_mask_0 = const()[name = string("op_1829_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1829_cast_fp16 = slice_by_index(begin = var_1829_begin_0, end = var_1829_end_0, end_mask = var_1829_end_mask_0, x = win_29_cast_fp16)[name = string("op_1829_cast_fp16")];
            string var_1835_pad_type_0 = const()[name = string("op_1835_pad_type_0"), val = string("valid")];
            int32 var_1835_groups_0 = const()[name = string("op_1835_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1835_strides_0 = const()[name = string("op_1835_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1835_pad_0 = const()[name = string("op_1835_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1835_dilations_0 = const()[name = string("op_1835_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_10_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136437248)))];
            tensor<fp16, [1, 768, 375]> var_1835_cast_fp16 = conv(dilations = var_1835_dilations_0, groups = var_1835_groups_0, pad = var_1835_pad_0, pad_type = var_1835_pad_type_0, strides = var_1835_strides_0, weight = core_backbone_layers_10_conv_conv_weight_to_fp16, x = win_29_cast_fp16)[name = string("op_1835_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_1837_cast_fp16 = mul(x = c_15_cast_fp16, y = var_1835_cast_fp16)[name = string("op_1837_cast_fp16")];
            tensor<int32, [3]> input_151_perm_0 = const()[name = string("input_151_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_10_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136441920)))];
            tensor<fp16, [1, 375, 768]> input_151_cast_fp16 = transpose(perm = input_151_perm_0, x = var_1837_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 375, 768]> linear_57_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_10_conv_out_proj_weight_to_fp16, x = input_151_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_111_cast_fp16 = add(x = x_109_cast_fp16, y = linear_57_cast_fp16)[name = string("x_111_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1843_cast_fp16 = abs(x = x_111_cast_fp16)[name = string("op_1843_cast_fp16")];
            tensor<int32, [1]> var_1845_axes_0 = const()[name = string("op_1845_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1845_cast_fp16 = reduce_max(axes = var_1845_axes_0, keep_dims = var_128, x = var_1843_cast_fp16)[name = string("op_1845_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_61_cast_fp16 = maximum(x = var_1845_cast_fp16, y = var_129_to_fp16)[name = string("s_61_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_71_cast_fp16 = real_div(x = x_111_cast_fp16, y = s_61_cast_fp16)[name = string("y_71_cast_fp16")];
            fp16 var_130_promoted_27_to_fp16 = const()[name = string("op_130_promoted_27_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1848_cast_fp16 = pow(x = y_71_cast_fp16, y = var_130_promoted_27_to_fp16)[name = string("op_1848_cast_fp16")];
            tensor<int32, [1]> var_1850_axes_0 = const()[name = string("op_1850_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1850_keep_dims_0 = const()[name = string("op_1850_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1850_cast_fp16 = reduce_mean(axes = var_1850_axes_0, keep_dims = var_1850_keep_dims_0, x = var_1848_cast_fp16)[name = string("op_1850_cast_fp16")];
            fp32 var_1851_epsilon_0 = const()[name = string("op_1851_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1851_cast_fp16 = inverse(epsilon = var_1851_epsilon_0, x = s_61_cast_fp16)[name = string("op_1851_cast_fp16")];
            fp16 var_1852_to_fp16 = const()[name = string("op_1852_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1853_cast_fp16 = mul(x = var_1851_cast_fp16, y = var_1852_to_fp16)[name = string("op_1853_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1854_cast_fp16 = real_div(x = var_1853_cast_fp16, y = s_61_cast_fp16)[name = string("op_1854_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1855_cast_fp16 = add(x = var_1850_cast_fp16, y = var_1854_cast_fp16)[name = string("op_1855_cast_fp16")];
            fp32 var_1856_epsilon_0 = const()[name = string("op_1856_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1856_cast_fp16 = rsqrt(epsilon = var_1856_epsilon_0, x = var_1855_cast_fp16)[name = string("op_1856_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1857_cast_fp16 = mul(x = y_71_cast_fp16, y = var_1856_cast_fp16)[name = string("op_1857_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_10_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_10_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137621632)))];
            tensor<fp16, [1, 375, 768]> input_153_cast_fp16 = mul(x = core_backbone_layers_10_ffn_norm_weight_to_fp16, y = var_1857_cast_fp16)[name = string("input_153_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_10_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137623232)))];
            tensor<fp16, [1, 375, 1792]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w1_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_58_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_10_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140375808)))];
            tensor<fp16, [1, 375, 1792]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w3_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_59_cast_fp16")];
            fp16 var_1866_to_fp16 = const()[name = string("op_1866_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_1867_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_1866_to_fp16)[name = string("op_1867_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1870_cast_fp16 = tanh(x = var_1867_cast_fp16)[name = string("op_1870_cast_fp16")];
            fp16 var_1871_to_fp16 = const()[name = string("op_1871_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_1872_cast_fp16 = add(x = var_1870_cast_fp16, y = var_1871_to_fp16)[name = string("op_1872_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_1873_cast_fp16 = mul(x = var_1867_cast_fp16, y = var_1872_cast_fp16)[name = string("op_1873_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_155_cast_fp16 = mul(x = var_1873_cast_fp16, y = linear_59_cast_fp16)[name = string("input_155_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_10_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143128384)))];
            tensor<fp16, [1, 375, 768]> linear_60_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w2_weight_to_fp16, x = input_155_cast_fp16)[name = string("linear_60_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_115_cast_fp16 = add(x = x_111_cast_fp16, y = linear_60_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1879_cast_fp16 = abs(x = x_115_cast_fp16)[name = string("op_1879_cast_fp16")];
            tensor<int32, [1]> var_1881_axes_0 = const()[name = string("op_1881_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_1881_cast_fp16 = reduce_max(axes = var_1881_axes_0, keep_dims = var_128, x = var_1879_cast_fp16)[name = string("op_1881_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_63_cast_fp16 = maximum(x = var_1881_cast_fp16, y = var_129_to_fp16)[name = string("s_63_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_73_cast_fp16 = real_div(x = x_115_cast_fp16, y = s_63_cast_fp16)[name = string("y_73_cast_fp16")];
            fp16 var_130_promoted_28_to_fp16 = const()[name = string("op_130_promoted_28_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_1884_cast_fp16 = pow(x = y_73_cast_fp16, y = var_130_promoted_28_to_fp16)[name = string("op_1884_cast_fp16")];
            tensor<int32, [1]> var_1886_axes_0 = const()[name = string("op_1886_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1886_keep_dims_0 = const()[name = string("op_1886_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_1886_cast_fp16 = reduce_mean(axes = var_1886_axes_0, keep_dims = var_1886_keep_dims_0, x = var_1884_cast_fp16)[name = string("op_1886_cast_fp16")];
            fp32 var_1887_epsilon_0 = const()[name = string("op_1887_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_1887_cast_fp16 = inverse(epsilon = var_1887_epsilon_0, x = s_63_cast_fp16)[name = string("op_1887_cast_fp16")];
            fp16 var_1888_to_fp16 = const()[name = string("op_1888_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_1889_cast_fp16 = mul(x = var_1887_cast_fp16, y = var_1888_to_fp16)[name = string("op_1889_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1890_cast_fp16 = real_div(x = var_1889_cast_fp16, y = s_63_cast_fp16)[name = string("op_1890_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_1891_cast_fp16 = add(x = var_1886_cast_fp16, y = var_1890_cast_fp16)[name = string("op_1891_cast_fp16")];
            fp32 var_1892_epsilon_0 = const()[name = string("op_1892_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_1892_cast_fp16 = rsqrt(epsilon = var_1892_epsilon_0, x = var_1891_cast_fp16)[name = string("op_1892_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_1893_cast_fp16 = mul(x = y_73_cast_fp16, y = var_1892_cast_fp16)[name = string("op_1893_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_11_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145880960)))];
            tensor<fp16, [1, 375, 768]> input_157_cast_fp16 = mul(x = core_backbone_layers_11_operator_norm_weight_to_fp16, y = var_1893_cast_fp16)[name = string("input_157_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145882560)))];
            tensor<fp16, [1, 375, 768]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_q_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<int32, [4]> concat_25 = const()[name = string("concat_25"), val = tensor<int32, [4]>([1, 375, 8, 96])];
            tensor<fp16, [1, 375, 8, 96]> x_117_cast_fp16 = reshape(shape = concat_25, x = linear_61_cast_fp16)[name = string("x_117_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1900_cast_fp16 = abs(x = x_117_cast_fp16)[name = string("op_1900_cast_fp16")];
            tensor<int32, [1]> var_1902_axes_0 = const()[name = string("op_1902_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 8, 1]> var_1902_cast_fp16 = reduce_max(axes = var_1902_axes_0, keep_dims = var_128, x = var_1900_cast_fp16)[name = string("op_1902_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> s_65_cast_fp16 = maximum(x = var_1902_cast_fp16, y = var_129_to_fp16)[name = string("s_65_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> y_75_cast_fp16 = real_div(x = x_117_cast_fp16, y = s_65_cast_fp16)[name = string("y_75_cast_fp16")];
            fp16 var_130_promoted_29_to_fp16 = const()[name = string("op_130_promoted_29_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 8, 96]> var_1905_cast_fp16 = pow(x = y_75_cast_fp16, y = var_130_promoted_29_to_fp16)[name = string("op_1905_cast_fp16")];
            tensor<int32, [1]> var_1907_axes_0 = const()[name = string("op_1907_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1907_keep_dims_0 = const()[name = string("op_1907_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 8, 1]> var_1907_cast_fp16 = reduce_mean(axes = var_1907_axes_0, keep_dims = var_1907_keep_dims_0, x = var_1905_cast_fp16)[name = string("op_1907_cast_fp16")];
            fp32 var_1908_epsilon_0 = const()[name = string("op_1908_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 8, 1]> var_1908_cast_fp16 = inverse(epsilon = var_1908_epsilon_0, x = s_65_cast_fp16)[name = string("op_1908_cast_fp16")];
            fp16 var_1909_to_fp16 = const()[name = string("op_1909_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 8, 1]> var_1910_cast_fp16 = mul(x = var_1908_cast_fp16, y = var_1909_to_fp16)[name = string("op_1910_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_1911_cast_fp16 = real_div(x = var_1910_cast_fp16, y = s_65_cast_fp16)[name = string("op_1911_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_1912_cast_fp16 = add(x = var_1907_cast_fp16, y = var_1911_cast_fp16)[name = string("op_1912_cast_fp16")];
            fp32 var_1913_epsilon_0 = const()[name = string("op_1913_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 8, 1]> var_1913_cast_fp16 = rsqrt(epsilon = var_1913_epsilon_0, x = var_1912_cast_fp16)[name = string("op_1913_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1914_cast_fp16 = mul(x = y_75_cast_fp16, y = var_1913_cast_fp16)[name = string("op_1914_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_11_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147062272)))];
            tensor<fp16, [1, 375, 8, 96]> var_1915_cast_fp16 = mul(x = core_backbone_layers_11_self_attn_q_norm_weight_to_fp16, y = var_1914_cast_fp16)[name = string("op_1915_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147062528)))];
            tensor<fp16, [1, 375, 384]> linear_62_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_k_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_62_cast_fp16")];
            tensor<int32, [4]> concat_26 = const()[name = string("concat_26"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> x_119_cast_fp16 = reshape(shape = concat_26, x = linear_62_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_1922_cast_fp16 = abs(x = x_119_cast_fp16)[name = string("op_1922_cast_fp16")];
            tensor<int32, [1]> var_1924_axes_0 = const()[name = string("op_1924_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 4, 1]> var_1924_cast_fp16 = reduce_max(axes = var_1924_axes_0, keep_dims = var_128, x = var_1922_cast_fp16)[name = string("op_1924_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> s_67_cast_fp16 = maximum(x = var_1924_cast_fp16, y = var_129_to_fp16)[name = string("s_67_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> y_77_cast_fp16 = real_div(x = x_119_cast_fp16, y = s_67_cast_fp16)[name = string("y_77_cast_fp16")];
            fp16 var_130_promoted_30_to_fp16 = const()[name = string("op_130_promoted_30_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 4, 96]> var_1927_cast_fp16 = pow(x = y_77_cast_fp16, y = var_130_promoted_30_to_fp16)[name = string("op_1927_cast_fp16")];
            tensor<int32, [1]> var_1929_axes_0 = const()[name = string("op_1929_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1929_keep_dims_0 = const()[name = string("op_1929_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 4, 1]> var_1929_cast_fp16 = reduce_mean(axes = var_1929_axes_0, keep_dims = var_1929_keep_dims_0, x = var_1927_cast_fp16)[name = string("op_1929_cast_fp16")];
            fp32 var_1930_epsilon_0 = const()[name = string("op_1930_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 4, 1]> var_1930_cast_fp16 = inverse(epsilon = var_1930_epsilon_0, x = s_67_cast_fp16)[name = string("op_1930_cast_fp16")];
            fp16 var_1931_to_fp16 = const()[name = string("op_1931_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 4, 1]> var_1932_cast_fp16 = mul(x = var_1930_cast_fp16, y = var_1931_to_fp16)[name = string("op_1932_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_1933_cast_fp16 = real_div(x = var_1932_cast_fp16, y = s_67_cast_fp16)[name = string("op_1933_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_1934_cast_fp16 = add(x = var_1929_cast_fp16, y = var_1933_cast_fp16)[name = string("op_1934_cast_fp16")];
            fp32 var_1935_epsilon_0 = const()[name = string("op_1935_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 4, 1]> var_1935_cast_fp16 = rsqrt(epsilon = var_1935_epsilon_0, x = var_1934_cast_fp16)[name = string("op_1935_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_1936_cast_fp16 = mul(x = y_77_cast_fp16, y = var_1935_cast_fp16)[name = string("op_1936_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_11_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147652416)))];
            tensor<fp16, [1, 375, 4, 96]> var_1937_cast_fp16 = mul(x = core_backbone_layers_11_self_attn_k_norm_weight_to_fp16, y = var_1936_cast_fp16)[name = string("op_1937_cast_fp16")];
            tensor<int32, [4]> new_29_perm_0 = const()[name = string("new_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147652672)))];
            tensor<fp16, [1, 375, 384]> linear_63_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_v_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_63_cast_fp16")];
            tensor<int32, [4]> concat_27 = const()[name = string("concat_27"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> var_1942_cast_fp16 = reshape(shape = concat_27, x = linear_63_cast_fp16)[name = string("op_1942_cast_fp16")];
            tensor<int32, [4]> new_31_perm_0 = const()[name = string("new_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_161_begin_0 = const()[name = string("input_161_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])];
            tensor<int32, [5]> input_161_end_0 = const()[name = string("input_161_end_0"), val = tensor<int32, [5]>([4, 1, 4, 31, 96])];
            tensor<bool, [5]> input_161_end_mask_0 = const()[name = string("input_161_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_161_squeeze_mask_0 = const()[name = string("input_161_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_161_cast_fp16 = slice_by_index(begin = input_161_begin_0, end = input_161_end_0, end_mask = input_161_end_mask_0, squeeze_mask = input_161_squeeze_mask_0, x = k)[name = string("input_161_cast_fp16")];
            tensor<int32, [8]> var_1947_pad_0 = const()[name = string("op_1947_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1947_mode_0 = const()[name = string("op_1947_mode_0"), val = string("constant")];
            fp16 const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_29_cast_fp16 = transpose(perm = new_29_perm_0, x = var_1937_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [1, 4, 406, 96]> var_1947_cast_fp16 = pad(constant_val = const_38_to_fp16, mode = var_1947_mode_0, pad = var_1947_pad_0, x = new_29_cast_fp16)[name = string("op_1947_cast_fp16")];
            tensor<int32, [8]> var_1949_pad_0 = const()[name = string("op_1949_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_1949_mode_0 = const()[name = string("op_1949_mode_0"), val = string("constant")];
            fp16 const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_1949_cast_fp16 = pad(constant_val = const_39_to_fp16, mode = var_1949_mode_0, pad = var_1949_pad_0, x = input_161_cast_fp16)[name = string("op_1949_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> x_123_cast_fp16 = add(x = var_1947_cast_fp16, y = var_1949_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<int32, [5]> input_165_begin_0 = const()[name = string("input_165_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])];
            tensor<int32, [5]> input_165_end_0 = const()[name = string("input_165_end_0"), val = tensor<int32, [5]>([4, 1, 4, 31, 96])];
            tensor<bool, [5]> input_165_end_mask_0 = const()[name = string("input_165_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_165_squeeze_mask_0 = const()[name = string("input_165_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_165_cast_fp16 = slice_by_index(begin = input_165_begin_0, end = input_165_end_0, end_mask = input_165_end_mask_0, squeeze_mask = input_165_squeeze_mask_0, x = v)[name = string("input_165_cast_fp16")];
            tensor<int32, [8]> var_1954_pad_0 = const()[name = string("op_1954_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1954_mode_0 = const()[name = string("op_1954_mode_0"), val = string("constant")];
            fp16 const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_31_cast_fp16 = transpose(perm = new_31_perm_0, x = var_1942_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [1, 4, 406, 96]> var_1954_cast_fp16 = pad(constant_val = const_40_to_fp16, mode = var_1954_mode_0, pad = var_1954_pad_0, x = new_31_cast_fp16)[name = string("op_1954_cast_fp16")];
            tensor<int32, [8]> var_1956_pad_0 = const()[name = string("op_1956_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_1956_mode_0 = const()[name = string("op_1956_mode_0"), val = string("constant")];
            fp16 const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_1956_cast_fp16 = pad(constant_val = const_41_to_fp16, mode = var_1956_mode_0, pad = var_1956_pad_0, x = input_165_cast_fp16)[name = string("op_1956_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> v_15_cast_fp16 = add(x = var_1954_cast_fp16, y = var_1956_cast_fp16)[name = string("v_15_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1958_cast_fp16 = mul(x = var_1915_cast_fp16, y = const_66_to_fp16)[name = string("op_1958_cast_fp16")];
            tensor<int32, [2]> var_1959_split_sizes_0 = const()[name = string("op_1959_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_79 = const()[name = string("const_79"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 48]> var_1959_cast_fp16_0, tensor<fp16, [1, 375, 8, 48]> var_1959_cast_fp16_1 = split(axis = const_79, split_sizes = var_1959_split_sizes_0, x = var_1915_cast_fp16)[name = string("op_1959_cast_fp16")];
            fp16 const_42_promoted_to_fp16 = const()[name = string("const_42_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 375, 8, 48]> var_1961_cast_fp16 = mul(x = var_1959_cast_fp16_1, y = const_42_promoted_to_fp16)[name = string("op_1961_cast_fp16")];
            bool var_1963_interleave_0 = const()[name = string("op_1963_interleave_0"), val = bool(false)];
            int32 const_80 = const()[name = string("const_80"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 96]> var_1963_cast_fp16 = concat(axis = const_80, interleave = var_1963_interleave_0, values = (var_1961_cast_fp16, var_1959_cast_fp16_0))[name = string("op_1963_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_1964_cast_fp16 = mul(x = var_1963_cast_fp16, y = const_69_to_fp16)[name = string("op_1964_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> q_7_cast_fp16 = add(x = var_1958_cast_fp16, y = var_1964_cast_fp16)[name = string("q_7_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_1966_cast_fp16 = mul(x = x_123_cast_fp16, y = const_0_to_fp16)[name = string("op_1966_cast_fp16")];
            tensor<int32, [2]> var_1967_split_sizes_0 = const()[name = string("op_1967_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1967_axis_0 = const()[name = string("op_1967_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 406, 48]> var_1967_cast_fp16_0, tensor<fp16, [1, 4, 406, 48]> var_1967_cast_fp16_1 = split(axis = var_1967_axis_0, split_sizes = var_1967_split_sizes_0, x = x_123_cast_fp16)[name = string("op_1967_cast_fp16")];
            fp16 const_43_promoted_to_fp16 = const()[name = string("const_43_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 406, 48]> var_1969_cast_fp16 = mul(x = var_1967_cast_fp16_1, y = const_43_promoted_to_fp16)[name = string("op_1969_cast_fp16")];
            bool var_1971_interleave_0 = const()[name = string("op_1971_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 406, 96]> var_1971_cast_fp16 = concat(axis = var_127, interleave = var_1971_interleave_0, values = (var_1969_cast_fp16, var_1967_cast_fp16_0))[name = string("op_1971_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_1972_cast_fp16 = mul(x = var_1971_cast_fp16, y = const_1_to_fp16)[name = string("op_1972_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> k_15_cast_fp16 = add(x = var_1966_cast_fp16, y = var_1972_cast_fp16)[name = string("k_15_cast_fp16")];
            tensor<int32, [1]> var_1974_axes_0 = const()[name = string("op_1974_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_1974_cast_fp16 = expand_dims(axes = var_1974_axes_0, x = k_15_cast_fp16)[name = string("op_1974_cast_fp16")];
            tensor<int32, [5]> var_1976_reps_0 = const()[name = string("op_1976_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_1976_cast_fp16 = tile(reps = var_1976_reps_0, x = var_1974_cast_fp16)[name = string("op_1976_cast_fp16")];
            tensor<int32, [4]> concat_29 = const()[name = string("concat_29"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> k_17_cast_fp16 = reshape(shape = concat_29, x = var_1976_cast_fp16)[name = string("k_17_cast_fp16")];
            tensor<int32, [1]> var_1978_axes_0 = const()[name = string("op_1978_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_1978_cast_fp16 = expand_dims(axes = var_1978_axes_0, x = v_15_cast_fp16)[name = string("op_1978_cast_fp16")];
            tensor<int32, [5]> var_1980_reps_0 = const()[name = string("op_1980_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_1980_cast_fp16 = tile(reps = var_1980_reps_0, x = var_1978_cast_fp16)[name = string("op_1980_cast_fp16")];
            tensor<int32, [4]> concat_31 = const()[name = string("concat_31"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> v_17_cast_fp16 = reshape(shape = concat_31, x = var_1980_cast_fp16)[name = string("v_17_cast_fp16")];
            tensor<int32, [4]> transpose_6_perm_0 = const()[name = string("transpose_6_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1983_transpose_x_1 = const()[name = string("op_1983_transpose_x_1"), val = bool(false)];
            bool var_1983_transpose_y_1 = const()[name = string("op_1983_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 96]> transpose_6_cast_fp16 = transpose(perm = transpose_6_perm_0, x = q_7_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 8, 375, 406]> var_1983_cast_fp16 = matmul(transpose_x = var_1983_transpose_x_1, transpose_y = var_1983_transpose_y_1, x = transpose_6_cast_fp16, y = k_17_cast_fp16)[name = string("op_1983_cast_fp16")];
            fp16 var_1984_to_fp16 = const()[name = string("op_1984_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 375, 406]> s_69_cast_fp16 = mul(x = var_1983_cast_fp16, y = var_1984_to_fp16)[name = string("s_69_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_125_cast_fp16 = add(x = s_69_cast_fp16, y = mask)[name = string("x_125_cast_fp16")];
            tensor<int32, [1]> var_1988_axes_0 = const()[name = string("op_1988_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 375, 1]> var_1988_cast_fp16 = reduce_max(axes = var_1988_axes_0, keep_dims = var_128, x = x_125_cast_fp16)[name = string("op_1988_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_127_cast_fp16 = sub(x = x_125_cast_fp16, y = var_1988_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_129_cast_fp16 = exp(x = x_127_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<int32, [1]> var_1992_axes_0 = const()[name = string("op_1992_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1992_keep_dims_0 = const()[name = string("op_1992_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 1]> var_1992_cast_fp16 = reduce_sum(axes = var_1992_axes_0, keep_dims = var_1992_keep_dims_0, x = x_129_cast_fp16)[name = string("op_1992_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> var_1993_cast_fp16 = real_div(x = x_129_cast_fp16, y = var_1992_cast_fp16)[name = string("op_1993_cast_fp16")];
            bool o_13_transpose_x_0 = const()[name = string("o_13_transpose_x_0"), val = bool(false)];
            bool o_13_transpose_y_0 = const()[name = string("o_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 375, 96]> o_13_cast_fp16 = matmul(transpose_x = o_13_transpose_x_0, transpose_y = o_13_transpose_y_0, x = var_1993_cast_fp16, y = v_17_cast_fp16)[name = string("o_13_cast_fp16")];
            tensor<int32, [4]> o_15_perm_0 = const()[name = string("o_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_32 = const()[name = string("concat_32"), val = tensor<int32, [3]>([1, 375, 768])];
            tensor<fp16, [1, 375, 8, 96]> o_15_cast_fp16 = transpose(perm = o_15_perm_0, x = o_13_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 375, 768]> input_167_cast_fp16 = reshape(shape = concat_32, x = o_15_cast_fp16)[name = string("input_167_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148242560)))];
            tensor<fp16, [1, 375, 768]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_o_proj_weight_to_fp16, x = input_167_cast_fp16)[name = string("linear_64_cast_fp16")];
            tensor<int32, [4]> var_2002_begin_0 = const()[name = string("op_2002_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_2002_end_0 = const()[name = string("op_2002_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_2002_end_mask_0 = const()[name = string("op_2002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2002_cast_fp16 = slice_by_index(begin = var_2002_begin_0, end = var_2002_end_0, end_mask = var_2002_end_mask_0, x = x_123_cast_fp16)[name = string("op_2002_cast_fp16")];
            tensor<int32, [4]> var_2005_begin_0 = const()[name = string("op_2005_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_2005_end_0 = const()[name = string("op_2005_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_2005_end_mask_0 = const()[name = string("op_2005_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2005_cast_fp16 = slice_by_index(begin = var_2005_begin_0, end = var_2005_end_0, end_mask = var_2005_end_mask_0, x = v_15_cast_fp16)[name = string("op_2005_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_131_cast_fp16 = add(x = x_115_cast_fp16, y = linear_64_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2008_cast_fp16 = abs(x = x_131_cast_fp16)[name = string("op_2008_cast_fp16")];
            tensor<int32, [1]> var_2010_axes_0 = const()[name = string("op_2010_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2010_cast_fp16 = reduce_max(axes = var_2010_axes_0, keep_dims = var_128, x = var_2008_cast_fp16)[name = string("op_2010_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_71_cast_fp16 = maximum(x = var_2010_cast_fp16, y = var_129_to_fp16)[name = string("s_71_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_79_cast_fp16 = real_div(x = x_131_cast_fp16, y = s_71_cast_fp16)[name = string("y_79_cast_fp16")];
            fp16 var_130_promoted_31_to_fp16 = const()[name = string("op_130_promoted_31_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2013_cast_fp16 = pow(x = y_79_cast_fp16, y = var_130_promoted_31_to_fp16)[name = string("op_2013_cast_fp16")];
            tensor<int32, [1]> var_2015_axes_0 = const()[name = string("op_2015_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2015_keep_dims_0 = const()[name = string("op_2015_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2015_cast_fp16 = reduce_mean(axes = var_2015_axes_0, keep_dims = var_2015_keep_dims_0, x = var_2013_cast_fp16)[name = string("op_2015_cast_fp16")];
            fp32 var_2016_epsilon_0 = const()[name = string("op_2016_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2016_cast_fp16 = inverse(epsilon = var_2016_epsilon_0, x = s_71_cast_fp16)[name = string("op_2016_cast_fp16")];
            fp16 var_2017_to_fp16 = const()[name = string("op_2017_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2018_cast_fp16 = mul(x = var_2016_cast_fp16, y = var_2017_to_fp16)[name = string("op_2018_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2019_cast_fp16 = real_div(x = var_2018_cast_fp16, y = s_71_cast_fp16)[name = string("op_2019_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2020_cast_fp16 = add(x = var_2015_cast_fp16, y = var_2019_cast_fp16)[name = string("op_2020_cast_fp16")];
            fp32 var_2021_epsilon_0 = const()[name = string("op_2021_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2021_cast_fp16 = rsqrt(epsilon = var_2021_epsilon_0, x = var_2020_cast_fp16)[name = string("op_2021_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2022_cast_fp16 = mul(x = y_79_cast_fp16, y = var_2021_cast_fp16)[name = string("op_2022_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_11_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149422272)))];
            tensor<fp16, [1, 375, 768]> input_169_cast_fp16 = mul(x = core_backbone_layers_11_ffn_norm_weight_to_fp16, y = var_2022_cast_fp16)[name = string("input_169_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_11_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149423872)))];
            tensor<fp16, [1, 375, 1792]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w1_weight_to_fp16, x = input_169_cast_fp16)[name = string("linear_65_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_11_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152176448)))];
            tensor<fp16, [1, 375, 1792]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w3_weight_to_fp16, x = input_169_cast_fp16)[name = string("linear_66_cast_fp16")];
            fp16 var_2031_to_fp16 = const()[name = string("op_2031_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_2032_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_2031_to_fp16)[name = string("op_2032_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2035_cast_fp16 = tanh(x = var_2032_cast_fp16)[name = string("op_2035_cast_fp16")];
            fp16 var_2036_to_fp16 = const()[name = string("op_2036_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_2037_cast_fp16 = add(x = var_2035_cast_fp16, y = var_2036_to_fp16)[name = string("op_2037_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2038_cast_fp16 = mul(x = var_2032_cast_fp16, y = var_2037_cast_fp16)[name = string("op_2038_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_171_cast_fp16 = mul(x = var_2038_cast_fp16, y = linear_66_cast_fp16)[name = string("input_171_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_11_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154929024)))];
            tensor<fp16, [1, 375, 768]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w2_weight_to_fp16, x = input_171_cast_fp16)[name = string("linear_67_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_135_cast_fp16 = add(x = x_131_cast_fp16, y = linear_67_cast_fp16)[name = string("x_135_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2044_cast_fp16 = abs(x = x_135_cast_fp16)[name = string("op_2044_cast_fp16")];
            tensor<int32, [1]> var_2046_axes_0 = const()[name = string("op_2046_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2046_cast_fp16 = reduce_max(axes = var_2046_axes_0, keep_dims = var_128, x = var_2044_cast_fp16)[name = string("op_2046_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_73_cast_fp16 = maximum(x = var_2046_cast_fp16, y = var_129_to_fp16)[name = string("s_73_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_81_cast_fp16 = real_div(x = x_135_cast_fp16, y = s_73_cast_fp16)[name = string("y_81_cast_fp16")];
            fp16 var_130_promoted_32_to_fp16 = const()[name = string("op_130_promoted_32_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2049_cast_fp16 = pow(x = y_81_cast_fp16, y = var_130_promoted_32_to_fp16)[name = string("op_2049_cast_fp16")];
            tensor<int32, [1]> var_2051_axes_0 = const()[name = string("op_2051_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2051_keep_dims_0 = const()[name = string("op_2051_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2051_cast_fp16 = reduce_mean(axes = var_2051_axes_0, keep_dims = var_2051_keep_dims_0, x = var_2049_cast_fp16)[name = string("op_2051_cast_fp16")];
            fp32 var_2052_epsilon_0 = const()[name = string("op_2052_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2052_cast_fp16 = inverse(epsilon = var_2052_epsilon_0, x = s_73_cast_fp16)[name = string("op_2052_cast_fp16")];
            fp16 var_2053_to_fp16 = const()[name = string("op_2053_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2054_cast_fp16 = mul(x = var_2052_cast_fp16, y = var_2053_to_fp16)[name = string("op_2054_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2055_cast_fp16 = real_div(x = var_2054_cast_fp16, y = s_73_cast_fp16)[name = string("op_2055_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2056_cast_fp16 = add(x = var_2051_cast_fp16, y = var_2055_cast_fp16)[name = string("op_2056_cast_fp16")];
            fp32 var_2057_epsilon_0 = const()[name = string("op_2057_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2057_cast_fp16 = rsqrt(epsilon = var_2057_epsilon_0, x = var_2056_cast_fp16)[name = string("op_2057_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2058_cast_fp16 = mul(x = y_81_cast_fp16, y = var_2057_cast_fp16)[name = string("op_2058_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_12_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_12_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157681600)))];
            tensor<fp16, [1, 375, 768]> input_173_cast_fp16 = mul(x = core_backbone_layers_12_operator_norm_weight_to_fp16, y = var_2058_cast_fp16)[name = string("input_173_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_12_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157683200)))];
            tensor<fp16, [1, 375, 2304]> linear_68_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_12_conv_in_proj_weight_to_fp16, x = input_173_cast_fp16)[name = string("linear_68_cast_fp16")];
            tensor<int32, [3]> bcx_17_perm_0 = const()[name = string("bcx_17_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_41_begin_0 = const()[name = string("b_41_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_41_end_0 = const()[name = string("b_41_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_41_end_mask_0 = const()[name = string("b_41_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_17_cast_fp16 = transpose(perm = bcx_17_perm_0, x = linear_68_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 768, 375]> b_41_cast_fp16 = slice_by_index(begin = b_41_begin_0, end = b_41_end_0, end_mask = b_41_end_mask_0, x = bcx_17_cast_fp16)[name = string("b_41_cast_fp16")];
            tensor<int32, [3]> c_17_begin_0 = const()[name = string("c_17_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_17_end_0 = const()[name = string("c_17_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_17_end_mask_0 = const()[name = string("c_17_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_17_cast_fp16 = slice_by_index(begin = c_17_begin_0, end = c_17_end_0, end_mask = c_17_end_mask_0, x = bcx_17_cast_fp16)[name = string("c_17_cast_fp16")];
            tensor<int32, [3]> xx_17_begin_0 = const()[name = string("xx_17_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_17_end_0 = const()[name = string("xx_17_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_17_end_mask_0 = const()[name = string("xx_17_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_17_cast_fp16 = slice_by_index(begin = xx_17_begin_0, end = xx_17_end_0, end_mask = xx_17_end_mask_0, x = bcx_17_cast_fp16)[name = string("xx_17_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_33_cast_fp16 = mul(x = b_41_cast_fp16, y = xx_17_cast_fp16)[name = string("new_33_cast_fp16")];
            tensor<int32, [4]> input_177_begin_0 = const()[name = string("input_177_begin_0"), val = tensor<int32, [4]>([8, 0, 0, 0])];
            tensor<int32, [4]> input_177_end_0 = const()[name = string("input_177_end_0"), val = tensor<int32, [4]>([9, 1, 768, 2])];
            tensor<bool, [4]> input_177_end_mask_0 = const()[name = string("input_177_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_177_squeeze_mask_0 = const()[name = string("input_177_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_177_cast_fp16 = slice_by_index(begin = input_177_begin_0, end = input_177_end_0, end_mask = input_177_end_mask_0, squeeze_mask = input_177_squeeze_mask_0, x = conv)[name = string("input_177_cast_fp16")];
            tensor<int32, [6]> var_2073_pad_0 = const()[name = string("op_2073_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_2073_mode_0 = const()[name = string("op_2073_mode_0"), val = string("constant")];
            fp16 const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_2073_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = var_2073_mode_0, pad = var_2073_pad_0, x = new_33_cast_fp16)[name = string("op_2073_cast_fp16")];
            tensor<int32, [6]> var_2075_pad_0 = const()[name = string("op_2075_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_2075_mode_0 = const()[name = string("op_2075_mode_0"), val = string("constant")];
            fp16 const_45_to_fp16 = const()[name = string("const_45_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_2075_cast_fp16 = pad(constant_val = const_45_to_fp16, mode = var_2075_mode_0, pad = var_2075_pad_0, x = input_177_cast_fp16)[name = string("op_2075_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_33_cast_fp16 = add(x = var_2073_cast_fp16, y = var_2075_cast_fp16)[name = string("win_33_cast_fp16")];
            tensor<int32, [3]> var_2077_begin_0 = const()[name = string("op_2077_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_2077_end_0 = const()[name = string("op_2077_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_2077_end_mask_0 = const()[name = string("op_2077_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_2077_cast_fp16 = slice_by_index(begin = var_2077_begin_0, end = var_2077_end_0, end_mask = var_2077_end_mask_0, x = win_33_cast_fp16)[name = string("op_2077_cast_fp16")];
            string var_2083_pad_type_0 = const()[name = string("op_2083_pad_type_0"), val = string("valid")];
            int32 var_2083_groups_0 = const()[name = string("op_2083_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_2083_strides_0 = const()[name = string("op_2083_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2083_pad_0 = const()[name = string("op_2083_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2083_dilations_0 = const()[name = string("op_2083_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_12_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161222208)))];
            tensor<fp16, [1, 768, 375]> var_2083_cast_fp16 = conv(dilations = var_2083_dilations_0, groups = var_2083_groups_0, pad = var_2083_pad_0, pad_type = var_2083_pad_type_0, strides = var_2083_strides_0, weight = core_backbone_layers_12_conv_conv_weight_to_fp16, x = win_33_cast_fp16)[name = string("op_2083_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_2085_cast_fp16 = mul(x = c_17_cast_fp16, y = var_2083_cast_fp16)[name = string("op_2085_cast_fp16")];
            tensor<int32, [3]> input_179_perm_0 = const()[name = string("input_179_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_12_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161226880)))];
            tensor<fp16, [1, 375, 768]> input_179_cast_fp16 = transpose(perm = input_179_perm_0, x = var_2085_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [1, 375, 768]> linear_69_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_12_conv_out_proj_weight_to_fp16, x = input_179_cast_fp16)[name = string("linear_69_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_137_cast_fp16 = add(x = x_135_cast_fp16, y = linear_69_cast_fp16)[name = string("x_137_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2091_cast_fp16 = abs(x = x_137_cast_fp16)[name = string("op_2091_cast_fp16")];
            tensor<int32, [1]> var_2093_axes_0 = const()[name = string("op_2093_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2093_cast_fp16 = reduce_max(axes = var_2093_axes_0, keep_dims = var_128, x = var_2091_cast_fp16)[name = string("op_2093_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_75_cast_fp16 = maximum(x = var_2093_cast_fp16, y = var_129_to_fp16)[name = string("s_75_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_85_cast_fp16 = real_div(x = x_137_cast_fp16, y = s_75_cast_fp16)[name = string("y_85_cast_fp16")];
            fp16 var_130_promoted_33_to_fp16 = const()[name = string("op_130_promoted_33_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2096_cast_fp16 = pow(x = y_85_cast_fp16, y = var_130_promoted_33_to_fp16)[name = string("op_2096_cast_fp16")];
            tensor<int32, [1]> var_2098_axes_0 = const()[name = string("op_2098_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2098_keep_dims_0 = const()[name = string("op_2098_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2098_cast_fp16 = reduce_mean(axes = var_2098_axes_0, keep_dims = var_2098_keep_dims_0, x = var_2096_cast_fp16)[name = string("op_2098_cast_fp16")];
            fp32 var_2099_epsilon_0 = const()[name = string("op_2099_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2099_cast_fp16 = inverse(epsilon = var_2099_epsilon_0, x = s_75_cast_fp16)[name = string("op_2099_cast_fp16")];
            fp16 var_2100_to_fp16 = const()[name = string("op_2100_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2101_cast_fp16 = mul(x = var_2099_cast_fp16, y = var_2100_to_fp16)[name = string("op_2101_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2102_cast_fp16 = real_div(x = var_2101_cast_fp16, y = s_75_cast_fp16)[name = string("op_2102_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2103_cast_fp16 = add(x = var_2098_cast_fp16, y = var_2102_cast_fp16)[name = string("op_2103_cast_fp16")];
            fp32 var_2104_epsilon_0 = const()[name = string("op_2104_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2104_cast_fp16 = rsqrt(epsilon = var_2104_epsilon_0, x = var_2103_cast_fp16)[name = string("op_2104_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2105_cast_fp16 = mul(x = y_85_cast_fp16, y = var_2104_cast_fp16)[name = string("op_2105_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_12_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_12_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162406592)))];
            tensor<fp16, [1, 375, 768]> input_181_cast_fp16 = mul(x = core_backbone_layers_12_ffn_norm_weight_to_fp16, y = var_2105_cast_fp16)[name = string("input_181_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_12_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162408192)))];
            tensor<fp16, [1, 375, 1792]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w1_weight_to_fp16, x = input_181_cast_fp16)[name = string("linear_70_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_12_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165160768)))];
            tensor<fp16, [1, 375, 1792]> linear_71_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w3_weight_to_fp16, x = input_181_cast_fp16)[name = string("linear_71_cast_fp16")];
            fp16 var_2114_to_fp16 = const()[name = string("op_2114_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_2115_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_2114_to_fp16)[name = string("op_2115_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2118_cast_fp16 = tanh(x = var_2115_cast_fp16)[name = string("op_2118_cast_fp16")];
            fp16 var_2119_to_fp16 = const()[name = string("op_2119_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_2120_cast_fp16 = add(x = var_2118_cast_fp16, y = var_2119_to_fp16)[name = string("op_2120_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2121_cast_fp16 = mul(x = var_2115_cast_fp16, y = var_2120_cast_fp16)[name = string("op_2121_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_183_cast_fp16 = mul(x = var_2121_cast_fp16, y = linear_71_cast_fp16)[name = string("input_183_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_12_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167913344)))];
            tensor<fp16, [1, 375, 768]> linear_72_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w2_weight_to_fp16, x = input_183_cast_fp16)[name = string("linear_72_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_141_cast_fp16 = add(x = x_137_cast_fp16, y = linear_72_cast_fp16)[name = string("x_141_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2127_cast_fp16 = abs(x = x_141_cast_fp16)[name = string("op_2127_cast_fp16")];
            tensor<int32, [1]> var_2129_axes_0 = const()[name = string("op_2129_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2129_cast_fp16 = reduce_max(axes = var_2129_axes_0, keep_dims = var_128, x = var_2127_cast_fp16)[name = string("op_2129_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_77_cast_fp16 = maximum(x = var_2129_cast_fp16, y = var_129_to_fp16)[name = string("s_77_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_87_cast_fp16 = real_div(x = x_141_cast_fp16, y = s_77_cast_fp16)[name = string("y_87_cast_fp16")];
            fp16 var_130_promoted_34_to_fp16 = const()[name = string("op_130_promoted_34_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2132_cast_fp16 = pow(x = y_87_cast_fp16, y = var_130_promoted_34_to_fp16)[name = string("op_2132_cast_fp16")];
            tensor<int32, [1]> var_2134_axes_0 = const()[name = string("op_2134_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2134_keep_dims_0 = const()[name = string("op_2134_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2134_cast_fp16 = reduce_mean(axes = var_2134_axes_0, keep_dims = var_2134_keep_dims_0, x = var_2132_cast_fp16)[name = string("op_2134_cast_fp16")];
            fp32 var_2135_epsilon_0 = const()[name = string("op_2135_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2135_cast_fp16 = inverse(epsilon = var_2135_epsilon_0, x = s_77_cast_fp16)[name = string("op_2135_cast_fp16")];
            fp16 var_2136_to_fp16 = const()[name = string("op_2136_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2137_cast_fp16 = mul(x = var_2135_cast_fp16, y = var_2136_to_fp16)[name = string("op_2137_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2138_cast_fp16 = real_div(x = var_2137_cast_fp16, y = s_77_cast_fp16)[name = string("op_2138_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2139_cast_fp16 = add(x = var_2134_cast_fp16, y = var_2138_cast_fp16)[name = string("op_2139_cast_fp16")];
            fp32 var_2140_epsilon_0 = const()[name = string("op_2140_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2140_cast_fp16 = rsqrt(epsilon = var_2140_epsilon_0, x = var_2139_cast_fp16)[name = string("op_2140_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2141_cast_fp16 = mul(x = y_87_cast_fp16, y = var_2140_cast_fp16)[name = string("op_2141_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_13_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_13_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170665920)))];
            tensor<fp16, [1, 375, 768]> input_185_cast_fp16 = mul(x = core_backbone_layers_13_operator_norm_weight_to_fp16, y = var_2141_cast_fp16)[name = string("input_185_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_13_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170667520)))];
            tensor<fp16, [1, 375, 2304]> linear_73_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_13_conv_in_proj_weight_to_fp16, x = input_185_cast_fp16)[name = string("linear_73_cast_fp16")];
            tensor<int32, [3]> bcx_perm_0 = const()[name = string("bcx_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_45_begin_0 = const()[name = string("b_45_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_45_end_0 = const()[name = string("b_45_end_0"), val = tensor<int32, [3]>([1, 768, 375])];
            tensor<bool, [3]> b_45_end_mask_0 = const()[name = string("b_45_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 375]> bcx_cast_fp16 = transpose(perm = bcx_perm_0, x = linear_73_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 768, 375]> b_45_cast_fp16 = slice_by_index(begin = b_45_begin_0, end = b_45_end_0, end_mask = b_45_end_mask_0, x = bcx_cast_fp16)[name = string("b_45_cast_fp16")];
            tensor<int32, [3]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [3]>([1, 1536, 375])];
            tensor<bool, [3]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 375]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = bcx_cast_fp16)[name = string("c_cast_fp16")];
            tensor<int32, [3]> xx_begin_0 = const()[name = string("xx_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_end_0 = const()[name = string("xx_end_0"), val = tensor<int32, [3]>([1, 2304, 375])];
            tensor<bool, [3]> xx_end_mask_0 = const()[name = string("xx_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 375]> xx_cast_fp16 = slice_by_index(begin = xx_begin_0, end = xx_end_0, end_mask = xx_end_mask_0, x = bcx_cast_fp16)[name = string("xx_cast_fp16")];
            tensor<fp16, [1, 768, 375]> new_35_cast_fp16 = mul(x = b_45_cast_fp16, y = xx_cast_fp16)[name = string("new_35_cast_fp16")];
            tensor<int32, [4]> input_189_begin_0 = const()[name = string("input_189_begin_0"), val = tensor<int32, [4]>([9, 0, 0, 0])];
            tensor<int32, [4]> input_189_end_0 = const()[name = string("input_189_end_0"), val = tensor<int32, [4]>([10, 1, 768, 2])];
            tensor<bool, [4]> input_189_end_mask_0 = const()[name = string("input_189_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_189_squeeze_mask_0 = const()[name = string("input_189_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_189_cast_fp16 = slice_by_index(begin = input_189_begin_0, end = input_189_end_0, end_mask = input_189_end_mask_0, squeeze_mask = input_189_squeeze_mask_0, x = conv)[name = string("input_189_cast_fp16")];
            tensor<int32, [6]> var_2156_pad_0 = const()[name = string("op_2156_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_2156_mode_0 = const()[name = string("op_2156_mode_0"), val = string("constant")];
            fp16 const_46_to_fp16 = const()[name = string("const_46_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_2156_cast_fp16 = pad(constant_val = const_46_to_fp16, mode = var_2156_mode_0, pad = var_2156_pad_0, x = new_35_cast_fp16)[name = string("op_2156_cast_fp16")];
            tensor<int32, [6]> var_2158_pad_0 = const()[name = string("op_2158_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 375])];
            string var_2158_mode_0 = const()[name = string("op_2158_mode_0"), val = string("constant")];
            fp16 const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 377]> var_2158_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = var_2158_mode_0, pad = var_2158_pad_0, x = input_189_cast_fp16)[name = string("op_2158_cast_fp16")];
            tensor<fp16, [1, 768, 377]> win_37_cast_fp16 = add(x = var_2156_cast_fp16, y = var_2158_cast_fp16)[name = string("win_37_cast_fp16")];
            tensor<int32, [3]> var_2160_begin_0 = const()[name = string("op_2160_begin_0"), val = tensor<int32, [3]>([0, 0, 375])];
            tensor<int32, [3]> var_2160_end_0 = const()[name = string("op_2160_end_0"), val = tensor<int32, [3]>([1, 768, 377])];
            tensor<bool, [3]> var_2160_end_mask_0 = const()[name = string("op_2160_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_2160_cast_fp16 = slice_by_index(begin = var_2160_begin_0, end = var_2160_end_0, end_mask = var_2160_end_mask_0, x = win_37_cast_fp16)[name = string("op_2160_cast_fp16")];
            string var_2166_pad_type_0 = const()[name = string("op_2166_pad_type_0"), val = string("valid")];
            int32 var_2166_groups_0 = const()[name = string("op_2166_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_2166_strides_0 = const()[name = string("op_2166_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2166_pad_0 = const()[name = string("op_2166_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2166_dilations_0 = const()[name = string("op_2166_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_13_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174206528)))];
            tensor<fp16, [1, 768, 375]> var_2166_cast_fp16 = conv(dilations = var_2166_dilations_0, groups = var_2166_groups_0, pad = var_2166_pad_0, pad_type = var_2166_pad_type_0, strides = var_2166_strides_0, weight = core_backbone_layers_13_conv_conv_weight_to_fp16, x = win_37_cast_fp16)[name = string("op_2166_cast_fp16")];
            tensor<fp16, [1, 768, 375]> var_2168_cast_fp16 = mul(x = c_cast_fp16, y = var_2166_cast_fp16)[name = string("op_2168_cast_fp16")];
            tensor<int32, [3]> input_191_perm_0 = const()[name = string("input_191_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_13_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174211200)))];
            tensor<fp16, [1, 375, 768]> input_191_cast_fp16 = transpose(perm = input_191_perm_0, x = var_2168_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 375, 768]> linear_74_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_13_conv_out_proj_weight_to_fp16, x = input_191_cast_fp16)[name = string("linear_74_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_143_cast_fp16 = add(x = x_141_cast_fp16, y = linear_74_cast_fp16)[name = string("x_143_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2174_cast_fp16 = abs(x = x_143_cast_fp16)[name = string("op_2174_cast_fp16")];
            tensor<int32, [1]> var_2176_axes_0 = const()[name = string("op_2176_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2176_cast_fp16 = reduce_max(axes = var_2176_axes_0, keep_dims = var_128, x = var_2174_cast_fp16)[name = string("op_2176_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_79_cast_fp16 = maximum(x = var_2176_cast_fp16, y = var_129_to_fp16)[name = string("s_79_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_91_cast_fp16 = real_div(x = x_143_cast_fp16, y = s_79_cast_fp16)[name = string("y_91_cast_fp16")];
            fp16 var_130_promoted_35_to_fp16 = const()[name = string("op_130_promoted_35_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2179_cast_fp16 = pow(x = y_91_cast_fp16, y = var_130_promoted_35_to_fp16)[name = string("op_2179_cast_fp16")];
            tensor<int32, [1]> var_2181_axes_0 = const()[name = string("op_2181_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2181_keep_dims_0 = const()[name = string("op_2181_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2181_cast_fp16 = reduce_mean(axes = var_2181_axes_0, keep_dims = var_2181_keep_dims_0, x = var_2179_cast_fp16)[name = string("op_2181_cast_fp16")];
            fp32 var_2182_epsilon_0 = const()[name = string("op_2182_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2182_cast_fp16 = inverse(epsilon = var_2182_epsilon_0, x = s_79_cast_fp16)[name = string("op_2182_cast_fp16")];
            fp16 var_2183_to_fp16 = const()[name = string("op_2183_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2184_cast_fp16 = mul(x = var_2182_cast_fp16, y = var_2183_to_fp16)[name = string("op_2184_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2185_cast_fp16 = real_div(x = var_2184_cast_fp16, y = s_79_cast_fp16)[name = string("op_2185_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2186_cast_fp16 = add(x = var_2181_cast_fp16, y = var_2185_cast_fp16)[name = string("op_2186_cast_fp16")];
            fp32 var_2187_epsilon_0 = const()[name = string("op_2187_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2187_cast_fp16 = rsqrt(epsilon = var_2187_epsilon_0, x = var_2186_cast_fp16)[name = string("op_2187_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2188_cast_fp16 = mul(x = y_91_cast_fp16, y = var_2187_cast_fp16)[name = string("op_2188_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_13_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_13_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175390912)))];
            tensor<fp16, [1, 375, 768]> input_193_cast_fp16 = mul(x = core_backbone_layers_13_ffn_norm_weight_to_fp16, y = var_2188_cast_fp16)[name = string("input_193_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_13_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175392512)))];
            tensor<fp16, [1, 375, 1792]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w1_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_75_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_13_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178145088)))];
            tensor<fp16, [1, 375, 1792]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w3_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_76_cast_fp16")];
            fp16 var_2197_to_fp16 = const()[name = string("op_2197_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_2198_cast_fp16 = mul(x = linear_75_cast_fp16, y = var_2197_to_fp16)[name = string("op_2198_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2201_cast_fp16 = tanh(x = var_2198_cast_fp16)[name = string("op_2201_cast_fp16")];
            fp16 var_2202_to_fp16 = const()[name = string("op_2202_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_2203_cast_fp16 = add(x = var_2201_cast_fp16, y = var_2202_to_fp16)[name = string("op_2203_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2204_cast_fp16 = mul(x = var_2198_cast_fp16, y = var_2203_cast_fp16)[name = string("op_2204_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_195_cast_fp16 = mul(x = var_2204_cast_fp16, y = linear_76_cast_fp16)[name = string("input_195_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_13_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180897664)))];
            tensor<fp16, [1, 375, 768]> linear_77_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w2_weight_to_fp16, x = input_195_cast_fp16)[name = string("linear_77_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_147_cast_fp16 = add(x = x_143_cast_fp16, y = linear_77_cast_fp16)[name = string("x_147_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2210_cast_fp16 = abs(x = x_147_cast_fp16)[name = string("op_2210_cast_fp16")];
            tensor<int32, [1]> var_2212_axes_0 = const()[name = string("op_2212_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2212_cast_fp16 = reduce_max(axes = var_2212_axes_0, keep_dims = var_128, x = var_2210_cast_fp16)[name = string("op_2212_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_81_cast_fp16 = maximum(x = var_2212_cast_fp16, y = var_129_to_fp16)[name = string("s_81_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_93_cast_fp16 = real_div(x = x_147_cast_fp16, y = s_81_cast_fp16)[name = string("y_93_cast_fp16")];
            fp16 var_130_promoted_36_to_fp16 = const()[name = string("op_130_promoted_36_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2215_cast_fp16 = pow(x = y_93_cast_fp16, y = var_130_promoted_36_to_fp16)[name = string("op_2215_cast_fp16")];
            tensor<int32, [1]> var_2217_axes_0 = const()[name = string("op_2217_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2217_keep_dims_0 = const()[name = string("op_2217_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2217_cast_fp16 = reduce_mean(axes = var_2217_axes_0, keep_dims = var_2217_keep_dims_0, x = var_2215_cast_fp16)[name = string("op_2217_cast_fp16")];
            fp32 var_2218_epsilon_0 = const()[name = string("op_2218_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2218_cast_fp16 = inverse(epsilon = var_2218_epsilon_0, x = s_81_cast_fp16)[name = string("op_2218_cast_fp16")];
            fp16 var_2219_to_fp16 = const()[name = string("op_2219_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2220_cast_fp16 = mul(x = var_2218_cast_fp16, y = var_2219_to_fp16)[name = string("op_2220_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2221_cast_fp16 = real_div(x = var_2220_cast_fp16, y = s_81_cast_fp16)[name = string("op_2221_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2222_cast_fp16 = add(x = var_2217_cast_fp16, y = var_2221_cast_fp16)[name = string("op_2222_cast_fp16")];
            fp32 var_2223_epsilon_0 = const()[name = string("op_2223_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2223_cast_fp16 = rsqrt(epsilon = var_2223_epsilon_0, x = var_2222_cast_fp16)[name = string("op_2223_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2224_cast_fp16 = mul(x = y_93_cast_fp16, y = var_2223_cast_fp16)[name = string("op_2224_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_14_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183650240)))];
            tensor<fp16, [1, 375, 768]> input_197_cast_fp16 = mul(x = core_backbone_layers_14_operator_norm_weight_to_fp16, y = var_2224_cast_fp16)[name = string("input_197_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183651840)))];
            tensor<fp16, [1, 375, 768]> linear_78_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_q_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_78_cast_fp16")];
            tensor<int32, [4]> concat_33 = const()[name = string("concat_33"), val = tensor<int32, [4]>([1, 375, 8, 96])];
            tensor<fp16, [1, 375, 8, 96]> x_149_cast_fp16 = reshape(shape = concat_33, x = linear_78_cast_fp16)[name = string("x_149_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2231_cast_fp16 = abs(x = x_149_cast_fp16)[name = string("op_2231_cast_fp16")];
            tensor<int32, [1]> var_2233_axes_0 = const()[name = string("op_2233_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 8, 1]> var_2233_cast_fp16 = reduce_max(axes = var_2233_axes_0, keep_dims = var_128, x = var_2231_cast_fp16)[name = string("op_2233_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> s_83_cast_fp16 = maximum(x = var_2233_cast_fp16, y = var_129_to_fp16)[name = string("s_83_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> y_95_cast_fp16 = real_div(x = x_149_cast_fp16, y = s_83_cast_fp16)[name = string("y_95_cast_fp16")];
            fp16 var_130_promoted_37_to_fp16 = const()[name = string("op_130_promoted_37_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 8, 96]> var_2236_cast_fp16 = pow(x = y_95_cast_fp16, y = var_130_promoted_37_to_fp16)[name = string("op_2236_cast_fp16")];
            tensor<int32, [1]> var_2238_axes_0 = const()[name = string("op_2238_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2238_keep_dims_0 = const()[name = string("op_2238_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 8, 1]> var_2238_cast_fp16 = reduce_mean(axes = var_2238_axes_0, keep_dims = var_2238_keep_dims_0, x = var_2236_cast_fp16)[name = string("op_2238_cast_fp16")];
            fp32 var_2239_epsilon_0 = const()[name = string("op_2239_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 8, 1]> var_2239_cast_fp16 = inverse(epsilon = var_2239_epsilon_0, x = s_83_cast_fp16)[name = string("op_2239_cast_fp16")];
            fp16 var_2240_to_fp16 = const()[name = string("op_2240_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 8, 1]> var_2241_cast_fp16 = mul(x = var_2239_cast_fp16, y = var_2240_to_fp16)[name = string("op_2241_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_2242_cast_fp16 = real_div(x = var_2241_cast_fp16, y = s_83_cast_fp16)[name = string("op_2242_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_2243_cast_fp16 = add(x = var_2238_cast_fp16, y = var_2242_cast_fp16)[name = string("op_2243_cast_fp16")];
            fp32 var_2244_epsilon_0 = const()[name = string("op_2244_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 8, 1]> var_2244_cast_fp16 = rsqrt(epsilon = var_2244_epsilon_0, x = var_2243_cast_fp16)[name = string("op_2244_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2245_cast_fp16 = mul(x = y_95_cast_fp16, y = var_2244_cast_fp16)[name = string("op_2245_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_14_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184831552)))];
            tensor<fp16, [1, 375, 8, 96]> var_2246_cast_fp16 = mul(x = core_backbone_layers_14_self_attn_q_norm_weight_to_fp16, y = var_2245_cast_fp16)[name = string("op_2246_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184831808)))];
            tensor<fp16, [1, 375, 384]> linear_79_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_k_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_79_cast_fp16")];
            tensor<int32, [4]> concat_34 = const()[name = string("concat_34"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> x_151_cast_fp16 = reshape(shape = concat_34, x = linear_79_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_2253_cast_fp16 = abs(x = x_151_cast_fp16)[name = string("op_2253_cast_fp16")];
            tensor<int32, [1]> var_2255_axes_0 = const()[name = string("op_2255_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 4, 1]> var_2255_cast_fp16 = reduce_max(axes = var_2255_axes_0, keep_dims = var_128, x = var_2253_cast_fp16)[name = string("op_2255_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> s_85_cast_fp16 = maximum(x = var_2255_cast_fp16, y = var_129_to_fp16)[name = string("s_85_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> y_97_cast_fp16 = real_div(x = x_151_cast_fp16, y = s_85_cast_fp16)[name = string("y_97_cast_fp16")];
            fp16 var_130_promoted_38_to_fp16 = const()[name = string("op_130_promoted_38_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 4, 96]> var_2258_cast_fp16 = pow(x = y_97_cast_fp16, y = var_130_promoted_38_to_fp16)[name = string("op_2258_cast_fp16")];
            tensor<int32, [1]> var_2260_axes_0 = const()[name = string("op_2260_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2260_keep_dims_0 = const()[name = string("op_2260_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 4, 1]> var_2260_cast_fp16 = reduce_mean(axes = var_2260_axes_0, keep_dims = var_2260_keep_dims_0, x = var_2258_cast_fp16)[name = string("op_2260_cast_fp16")];
            fp32 var_2261_epsilon_0 = const()[name = string("op_2261_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 4, 1]> var_2261_cast_fp16 = inverse(epsilon = var_2261_epsilon_0, x = s_85_cast_fp16)[name = string("op_2261_cast_fp16")];
            fp16 var_2262_to_fp16 = const()[name = string("op_2262_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 4, 1]> var_2263_cast_fp16 = mul(x = var_2261_cast_fp16, y = var_2262_to_fp16)[name = string("op_2263_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_2264_cast_fp16 = real_div(x = var_2263_cast_fp16, y = s_85_cast_fp16)[name = string("op_2264_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_2265_cast_fp16 = add(x = var_2260_cast_fp16, y = var_2264_cast_fp16)[name = string("op_2265_cast_fp16")];
            fp32 var_2266_epsilon_0 = const()[name = string("op_2266_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 4, 1]> var_2266_cast_fp16 = rsqrt(epsilon = var_2266_epsilon_0, x = var_2265_cast_fp16)[name = string("op_2266_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_2267_cast_fp16 = mul(x = y_97_cast_fp16, y = var_2266_cast_fp16)[name = string("op_2267_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_14_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185421696)))];
            tensor<fp16, [1, 375, 4, 96]> var_2268_cast_fp16 = mul(x = core_backbone_layers_14_self_attn_k_norm_weight_to_fp16, y = var_2267_cast_fp16)[name = string("op_2268_cast_fp16")];
            tensor<int32, [4]> new_37_perm_0 = const()[name = string("new_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185421952)))];
            tensor<fp16, [1, 375, 384]> linear_80_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_v_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_80_cast_fp16")];
            tensor<int32, [4]> concat_35 = const()[name = string("concat_35"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> var_2273_cast_fp16 = reshape(shape = concat_35, x = linear_80_cast_fp16)[name = string("op_2273_cast_fp16")];
            tensor<int32, [4]> new_39_perm_0 = const()[name = string("new_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_201_begin_0 = const()[name = string("input_201_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])];
            tensor<int32, [5]> input_201_end_0 = const()[name = string("input_201_end_0"), val = tensor<int32, [5]>([5, 1, 4, 31, 96])];
            tensor<bool, [5]> input_201_end_mask_0 = const()[name = string("input_201_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_201_squeeze_mask_0 = const()[name = string("input_201_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_201_cast_fp16 = slice_by_index(begin = input_201_begin_0, end = input_201_end_0, end_mask = input_201_end_mask_0, squeeze_mask = input_201_squeeze_mask_0, x = k)[name = string("input_201_cast_fp16")];
            tensor<int32, [8]> var_2278_pad_0 = const()[name = string("op_2278_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2278_mode_0 = const()[name = string("op_2278_mode_0"), val = string("constant")];
            fp16 const_48_to_fp16 = const()[name = string("const_48_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_37_cast_fp16 = transpose(perm = new_37_perm_0, x = var_2268_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 4, 406, 96]> var_2278_cast_fp16 = pad(constant_val = const_48_to_fp16, mode = var_2278_mode_0, pad = var_2278_pad_0, x = new_37_cast_fp16)[name = string("op_2278_cast_fp16")];
            tensor<int32, [8]> var_2280_pad_0 = const()[name = string("op_2280_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_2280_mode_0 = const()[name = string("op_2280_mode_0"), val = string("constant")];
            fp16 const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_2280_cast_fp16 = pad(constant_val = const_49_to_fp16, mode = var_2280_mode_0, pad = var_2280_pad_0, x = input_201_cast_fp16)[name = string("op_2280_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> x_155_cast_fp16 = add(x = var_2278_cast_fp16, y = var_2280_cast_fp16)[name = string("x_155_cast_fp16")];
            tensor<int32, [5]> input_205_begin_0 = const()[name = string("input_205_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])];
            tensor<int32, [5]> input_205_end_0 = const()[name = string("input_205_end_0"), val = tensor<int32, [5]>([5, 1, 4, 31, 96])];
            tensor<bool, [5]> input_205_end_mask_0 = const()[name = string("input_205_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_205_squeeze_mask_0 = const()[name = string("input_205_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_205_cast_fp16 = slice_by_index(begin = input_205_begin_0, end = input_205_end_0, end_mask = input_205_end_mask_0, squeeze_mask = input_205_squeeze_mask_0, x = v)[name = string("input_205_cast_fp16")];
            tensor<int32, [8]> var_2285_pad_0 = const()[name = string("op_2285_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2285_mode_0 = const()[name = string("op_2285_mode_0"), val = string("constant")];
            fp16 const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_39_cast_fp16 = transpose(perm = new_39_perm_0, x = var_2273_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 4, 406, 96]> var_2285_cast_fp16 = pad(constant_val = const_50_to_fp16, mode = var_2285_mode_0, pad = var_2285_pad_0, x = new_39_cast_fp16)[name = string("op_2285_cast_fp16")];
            tensor<int32, [8]> var_2287_pad_0 = const()[name = string("op_2287_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_2287_mode_0 = const()[name = string("op_2287_mode_0"), val = string("constant")];
            fp16 const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_2287_cast_fp16 = pad(constant_val = const_51_to_fp16, mode = var_2287_mode_0, pad = var_2287_pad_0, x = input_205_cast_fp16)[name = string("op_2287_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> v_19_cast_fp16 = add(x = var_2285_cast_fp16, y = var_2287_cast_fp16)[name = string("v_19_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2289_cast_fp16 = mul(x = var_2246_cast_fp16, y = const_66_to_fp16)[name = string("op_2289_cast_fp16")];
            tensor<int32, [2]> var_2290_split_sizes_0 = const()[name = string("op_2290_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_83 = const()[name = string("const_83"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 48]> var_2290_cast_fp16_0, tensor<fp16, [1, 375, 8, 48]> var_2290_cast_fp16_1 = split(axis = const_83, split_sizes = var_2290_split_sizes_0, x = var_2246_cast_fp16)[name = string("op_2290_cast_fp16")];
            fp16 const_52_promoted_to_fp16 = const()[name = string("const_52_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 375, 8, 48]> var_2292_cast_fp16 = mul(x = var_2290_cast_fp16_1, y = const_52_promoted_to_fp16)[name = string("op_2292_cast_fp16")];
            bool var_2294_interleave_0 = const()[name = string("op_2294_interleave_0"), val = bool(false)];
            int32 const_84 = const()[name = string("const_84"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 96]> var_2294_cast_fp16 = concat(axis = const_84, interleave = var_2294_interleave_0, values = (var_2292_cast_fp16, var_2290_cast_fp16_0))[name = string("op_2294_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2295_cast_fp16 = mul(x = var_2294_cast_fp16, y = const_69_to_fp16)[name = string("op_2295_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> q_9_cast_fp16 = add(x = var_2289_cast_fp16, y = var_2295_cast_fp16)[name = string("q_9_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_2297_cast_fp16 = mul(x = x_155_cast_fp16, y = const_0_to_fp16)[name = string("op_2297_cast_fp16")];
            tensor<int32, [2]> var_2298_split_sizes_0 = const()[name = string("op_2298_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2298_axis_0 = const()[name = string("op_2298_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 406, 48]> var_2298_cast_fp16_0, tensor<fp16, [1, 4, 406, 48]> var_2298_cast_fp16_1 = split(axis = var_2298_axis_0, split_sizes = var_2298_split_sizes_0, x = x_155_cast_fp16)[name = string("op_2298_cast_fp16")];
            fp16 const_53_promoted_to_fp16 = const()[name = string("const_53_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 406, 48]> var_2300_cast_fp16 = mul(x = var_2298_cast_fp16_1, y = const_53_promoted_to_fp16)[name = string("op_2300_cast_fp16")];
            bool var_2302_interleave_0 = const()[name = string("op_2302_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 406, 96]> var_2302_cast_fp16 = concat(axis = var_127, interleave = var_2302_interleave_0, values = (var_2300_cast_fp16, var_2298_cast_fp16_0))[name = string("op_2302_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_2303_cast_fp16 = mul(x = var_2302_cast_fp16, y = const_1_to_fp16)[name = string("op_2303_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> k_19_cast_fp16 = add(x = var_2297_cast_fp16, y = var_2303_cast_fp16)[name = string("k_19_cast_fp16")];
            tensor<int32, [1]> var_2305_axes_0 = const()[name = string("op_2305_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_2305_cast_fp16 = expand_dims(axes = var_2305_axes_0, x = k_19_cast_fp16)[name = string("op_2305_cast_fp16")];
            tensor<int32, [5]> var_2307_reps_0 = const()[name = string("op_2307_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_2307_cast_fp16 = tile(reps = var_2307_reps_0, x = var_2305_cast_fp16)[name = string("op_2307_cast_fp16")];
            tensor<int32, [4]> concat_37 = const()[name = string("concat_37"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> k_21_cast_fp16 = reshape(shape = concat_37, x = var_2307_cast_fp16)[name = string("k_21_cast_fp16")];
            tensor<int32, [1]> var_2309_axes_0 = const()[name = string("op_2309_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_2309_cast_fp16 = expand_dims(axes = var_2309_axes_0, x = v_19_cast_fp16)[name = string("op_2309_cast_fp16")];
            tensor<int32, [5]> var_2311_reps_0 = const()[name = string("op_2311_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_2311_cast_fp16 = tile(reps = var_2311_reps_0, x = var_2309_cast_fp16)[name = string("op_2311_cast_fp16")];
            tensor<int32, [4]> concat_39 = const()[name = string("concat_39"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> v_21_cast_fp16 = reshape(shape = concat_39, x = var_2311_cast_fp16)[name = string("v_21_cast_fp16")];
            tensor<int32, [4]> transpose_8_perm_0 = const()[name = string("transpose_8_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2314_transpose_x_1 = const()[name = string("op_2314_transpose_x_1"), val = bool(false)];
            bool var_2314_transpose_y_1 = const()[name = string("op_2314_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 96]> transpose_8_cast_fp16 = transpose(perm = transpose_8_perm_0, x = q_9_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 8, 375, 406]> var_2314_cast_fp16 = matmul(transpose_x = var_2314_transpose_x_1, transpose_y = var_2314_transpose_y_1, x = transpose_8_cast_fp16, y = k_21_cast_fp16)[name = string("op_2314_cast_fp16")];
            fp16 var_2315_to_fp16 = const()[name = string("op_2315_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 375, 406]> s_87_cast_fp16 = mul(x = var_2314_cast_fp16, y = var_2315_to_fp16)[name = string("s_87_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_157_cast_fp16 = add(x = s_87_cast_fp16, y = mask)[name = string("x_157_cast_fp16")];
            tensor<int32, [1]> var_2319_axes_0 = const()[name = string("op_2319_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 375, 1]> var_2319_cast_fp16 = reduce_max(axes = var_2319_axes_0, keep_dims = var_128, x = x_157_cast_fp16)[name = string("op_2319_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_159_cast_fp16 = sub(x = x_157_cast_fp16, y = var_2319_cast_fp16)[name = string("x_159_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_161_cast_fp16 = exp(x = x_159_cast_fp16)[name = string("x_161_cast_fp16")];
            tensor<int32, [1]> var_2323_axes_0 = const()[name = string("op_2323_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2323_keep_dims_0 = const()[name = string("op_2323_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 1]> var_2323_cast_fp16 = reduce_sum(axes = var_2323_axes_0, keep_dims = var_2323_keep_dims_0, x = x_161_cast_fp16)[name = string("op_2323_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> var_2324_cast_fp16 = real_div(x = x_161_cast_fp16, y = var_2323_cast_fp16)[name = string("op_2324_cast_fp16")];
            bool o_17_transpose_x_0 = const()[name = string("o_17_transpose_x_0"), val = bool(false)];
            bool o_17_transpose_y_0 = const()[name = string("o_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 375, 96]> o_17_cast_fp16 = matmul(transpose_x = o_17_transpose_x_0, transpose_y = o_17_transpose_y_0, x = var_2324_cast_fp16, y = v_21_cast_fp16)[name = string("o_17_cast_fp16")];
            tensor<int32, [4]> o_19_perm_0 = const()[name = string("o_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_40 = const()[name = string("concat_40"), val = tensor<int32, [3]>([1, 375, 768])];
            tensor<fp16, [1, 375, 8, 96]> o_19_cast_fp16 = transpose(perm = o_19_perm_0, x = o_17_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 375, 768]> input_207_cast_fp16 = reshape(shape = concat_40, x = o_19_cast_fp16)[name = string("input_207_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186011840)))];
            tensor<fp16, [1, 375, 768]> linear_81_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_o_proj_weight_to_fp16, x = input_207_cast_fp16)[name = string("linear_81_cast_fp16")];
            tensor<int32, [4]> var_2333_begin_0 = const()[name = string("op_2333_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_2333_end_0 = const()[name = string("op_2333_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_2333_end_mask_0 = const()[name = string("op_2333_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2333_cast_fp16 = slice_by_index(begin = var_2333_begin_0, end = var_2333_end_0, end_mask = var_2333_end_mask_0, x = x_155_cast_fp16)[name = string("op_2333_cast_fp16")];
            tensor<int32, [4]> var_2336_begin_0 = const()[name = string("op_2336_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_2336_end_0 = const()[name = string("op_2336_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_2336_end_mask_0 = const()[name = string("op_2336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2336_cast_fp16 = slice_by_index(begin = var_2336_begin_0, end = var_2336_end_0, end_mask = var_2336_end_mask_0, x = v_19_cast_fp16)[name = string("op_2336_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_163_cast_fp16 = add(x = x_147_cast_fp16, y = linear_81_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2339_cast_fp16 = abs(x = x_163_cast_fp16)[name = string("op_2339_cast_fp16")];
            tensor<int32, [1]> var_2341_axes_0 = const()[name = string("op_2341_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2341_cast_fp16 = reduce_max(axes = var_2341_axes_0, keep_dims = var_128, x = var_2339_cast_fp16)[name = string("op_2341_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_89_cast_fp16 = maximum(x = var_2341_cast_fp16, y = var_129_to_fp16)[name = string("s_89_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_99_cast_fp16 = real_div(x = x_163_cast_fp16, y = s_89_cast_fp16)[name = string("y_99_cast_fp16")];
            fp16 var_130_promoted_39_to_fp16 = const()[name = string("op_130_promoted_39_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2344_cast_fp16 = pow(x = y_99_cast_fp16, y = var_130_promoted_39_to_fp16)[name = string("op_2344_cast_fp16")];
            tensor<int32, [1]> var_2346_axes_0 = const()[name = string("op_2346_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2346_keep_dims_0 = const()[name = string("op_2346_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2346_cast_fp16 = reduce_mean(axes = var_2346_axes_0, keep_dims = var_2346_keep_dims_0, x = var_2344_cast_fp16)[name = string("op_2346_cast_fp16")];
            fp32 var_2347_epsilon_0 = const()[name = string("op_2347_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2347_cast_fp16 = inverse(epsilon = var_2347_epsilon_0, x = s_89_cast_fp16)[name = string("op_2347_cast_fp16")];
            fp16 var_2348_to_fp16 = const()[name = string("op_2348_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2349_cast_fp16 = mul(x = var_2347_cast_fp16, y = var_2348_to_fp16)[name = string("op_2349_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2350_cast_fp16 = real_div(x = var_2349_cast_fp16, y = s_89_cast_fp16)[name = string("op_2350_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2351_cast_fp16 = add(x = var_2346_cast_fp16, y = var_2350_cast_fp16)[name = string("op_2351_cast_fp16")];
            fp32 var_2352_epsilon_0 = const()[name = string("op_2352_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2352_cast_fp16 = rsqrt(epsilon = var_2352_epsilon_0, x = var_2351_cast_fp16)[name = string("op_2352_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2353_cast_fp16 = mul(x = y_99_cast_fp16, y = var_2352_cast_fp16)[name = string("op_2353_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_14_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187191552)))];
            tensor<fp16, [1, 375, 768]> input_209_cast_fp16 = mul(x = core_backbone_layers_14_ffn_norm_weight_to_fp16, y = var_2353_cast_fp16)[name = string("input_209_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_14_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187193152)))];
            tensor<fp16, [1, 375, 1792]> linear_82_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w1_weight_to_fp16, x = input_209_cast_fp16)[name = string("linear_82_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_14_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189945728)))];
            tensor<fp16, [1, 375, 1792]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w3_weight_to_fp16, x = input_209_cast_fp16)[name = string("linear_83_cast_fp16")];
            fp16 var_2362_to_fp16 = const()[name = string("op_2362_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_2363_cast_fp16 = mul(x = linear_82_cast_fp16, y = var_2362_to_fp16)[name = string("op_2363_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2366_cast_fp16 = tanh(x = var_2363_cast_fp16)[name = string("op_2366_cast_fp16")];
            fp16 var_2367_to_fp16 = const()[name = string("op_2367_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_2368_cast_fp16 = add(x = var_2366_cast_fp16, y = var_2367_to_fp16)[name = string("op_2368_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2369_cast_fp16 = mul(x = var_2363_cast_fp16, y = var_2368_cast_fp16)[name = string("op_2369_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_211_cast_fp16 = mul(x = var_2369_cast_fp16, y = linear_83_cast_fp16)[name = string("input_211_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_14_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192698304)))];
            tensor<fp16, [1, 375, 768]> linear_84_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w2_weight_to_fp16, x = input_211_cast_fp16)[name = string("linear_84_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_167_cast_fp16 = add(x = x_163_cast_fp16, y = linear_84_cast_fp16)[name = string("x_167_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2375_cast_fp16 = abs(x = x_167_cast_fp16)[name = string("op_2375_cast_fp16")];
            tensor<int32, [1]> var_2377_axes_0 = const()[name = string("op_2377_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2377_cast_fp16 = reduce_max(axes = var_2377_axes_0, keep_dims = var_128, x = var_2375_cast_fp16)[name = string("op_2377_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_91_cast_fp16 = maximum(x = var_2377_cast_fp16, y = var_129_to_fp16)[name = string("s_91_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_101_cast_fp16 = real_div(x = x_167_cast_fp16, y = s_91_cast_fp16)[name = string("y_101_cast_fp16")];
            fp16 var_130_promoted_40_to_fp16 = const()[name = string("op_130_promoted_40_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2380_cast_fp16 = pow(x = y_101_cast_fp16, y = var_130_promoted_40_to_fp16)[name = string("op_2380_cast_fp16")];
            tensor<int32, [1]> var_2382_axes_0 = const()[name = string("op_2382_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2382_keep_dims_0 = const()[name = string("op_2382_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2382_cast_fp16 = reduce_mean(axes = var_2382_axes_0, keep_dims = var_2382_keep_dims_0, x = var_2380_cast_fp16)[name = string("op_2382_cast_fp16")];
            fp32 var_2383_epsilon_0 = const()[name = string("op_2383_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2383_cast_fp16 = inverse(epsilon = var_2383_epsilon_0, x = s_91_cast_fp16)[name = string("op_2383_cast_fp16")];
            fp16 var_2384_to_fp16 = const()[name = string("op_2384_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2385_cast_fp16 = mul(x = var_2383_cast_fp16, y = var_2384_to_fp16)[name = string("op_2385_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2386_cast_fp16 = real_div(x = var_2385_cast_fp16, y = s_91_cast_fp16)[name = string("op_2386_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2387_cast_fp16 = add(x = var_2382_cast_fp16, y = var_2386_cast_fp16)[name = string("op_2387_cast_fp16")];
            fp32 var_2388_epsilon_0 = const()[name = string("op_2388_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2388_cast_fp16 = rsqrt(epsilon = var_2388_epsilon_0, x = var_2387_cast_fp16)[name = string("op_2388_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2389_cast_fp16 = mul(x = y_101_cast_fp16, y = var_2388_cast_fp16)[name = string("op_2389_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_15_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195450880)))];
            tensor<fp16, [1, 375, 768]> input_213_cast_fp16 = mul(x = core_backbone_layers_15_operator_norm_weight_to_fp16, y = var_2389_cast_fp16)[name = string("input_213_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195452480)))];
            tensor<fp16, [1, 375, 768]> linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_q_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_85_cast_fp16")];
            tensor<int32, [4]> concat_41 = const()[name = string("concat_41"), val = tensor<int32, [4]>([1, 375, 8, 96])];
            tensor<fp16, [1, 375, 8, 96]> x_169_cast_fp16 = reshape(shape = concat_41, x = linear_85_cast_fp16)[name = string("x_169_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2396_cast_fp16 = abs(x = x_169_cast_fp16)[name = string("op_2396_cast_fp16")];
            tensor<int32, [1]> var_2398_axes_0 = const()[name = string("op_2398_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 8, 1]> var_2398_cast_fp16 = reduce_max(axes = var_2398_axes_0, keep_dims = var_128, x = var_2396_cast_fp16)[name = string("op_2398_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> s_93_cast_fp16 = maximum(x = var_2398_cast_fp16, y = var_129_to_fp16)[name = string("s_93_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> y_103_cast_fp16 = real_div(x = x_169_cast_fp16, y = s_93_cast_fp16)[name = string("y_103_cast_fp16")];
            fp16 var_130_promoted_41_to_fp16 = const()[name = string("op_130_promoted_41_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 8, 96]> var_2401_cast_fp16 = pow(x = y_103_cast_fp16, y = var_130_promoted_41_to_fp16)[name = string("op_2401_cast_fp16")];
            tensor<int32, [1]> var_2403_axes_0 = const()[name = string("op_2403_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2403_keep_dims_0 = const()[name = string("op_2403_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 8, 1]> var_2403_cast_fp16 = reduce_mean(axes = var_2403_axes_0, keep_dims = var_2403_keep_dims_0, x = var_2401_cast_fp16)[name = string("op_2403_cast_fp16")];
            fp32 var_2404_epsilon_0 = const()[name = string("op_2404_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 8, 1]> var_2404_cast_fp16 = inverse(epsilon = var_2404_epsilon_0, x = s_93_cast_fp16)[name = string("op_2404_cast_fp16")];
            fp16 var_2405_to_fp16 = const()[name = string("op_2405_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 8, 1]> var_2406_cast_fp16 = mul(x = var_2404_cast_fp16, y = var_2405_to_fp16)[name = string("op_2406_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_2407_cast_fp16 = real_div(x = var_2406_cast_fp16, y = s_93_cast_fp16)[name = string("op_2407_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_2408_cast_fp16 = add(x = var_2403_cast_fp16, y = var_2407_cast_fp16)[name = string("op_2408_cast_fp16")];
            fp32 var_2409_epsilon_0 = const()[name = string("op_2409_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 8, 1]> var_2409_cast_fp16 = rsqrt(epsilon = var_2409_epsilon_0, x = var_2408_cast_fp16)[name = string("op_2409_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2410_cast_fp16 = mul(x = y_103_cast_fp16, y = var_2409_cast_fp16)[name = string("op_2410_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_15_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196632192)))];
            tensor<fp16, [1, 375, 8, 96]> var_2411_cast_fp16 = mul(x = core_backbone_layers_15_self_attn_q_norm_weight_to_fp16, y = var_2410_cast_fp16)[name = string("op_2411_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196632448)))];
            tensor<fp16, [1, 375, 384]> linear_86_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_k_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_86_cast_fp16")];
            tensor<int32, [4]> concat_42 = const()[name = string("concat_42"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> x_171_cast_fp16 = reshape(shape = concat_42, x = linear_86_cast_fp16)[name = string("x_171_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_2418_cast_fp16 = abs(x = x_171_cast_fp16)[name = string("op_2418_cast_fp16")];
            tensor<int32, [1]> var_2420_axes_0 = const()[name = string("op_2420_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 4, 1]> var_2420_cast_fp16 = reduce_max(axes = var_2420_axes_0, keep_dims = var_128, x = var_2418_cast_fp16)[name = string("op_2420_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> s_95_cast_fp16 = maximum(x = var_2420_cast_fp16, y = var_129_to_fp16)[name = string("s_95_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> y_105_cast_fp16 = real_div(x = x_171_cast_fp16, y = s_95_cast_fp16)[name = string("y_105_cast_fp16")];
            fp16 var_130_promoted_42_to_fp16 = const()[name = string("op_130_promoted_42_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 4, 96]> var_2423_cast_fp16 = pow(x = y_105_cast_fp16, y = var_130_promoted_42_to_fp16)[name = string("op_2423_cast_fp16")];
            tensor<int32, [1]> var_2425_axes_0 = const()[name = string("op_2425_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2425_keep_dims_0 = const()[name = string("op_2425_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 4, 1]> var_2425_cast_fp16 = reduce_mean(axes = var_2425_axes_0, keep_dims = var_2425_keep_dims_0, x = var_2423_cast_fp16)[name = string("op_2425_cast_fp16")];
            fp32 var_2426_epsilon_0 = const()[name = string("op_2426_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 4, 1]> var_2426_cast_fp16 = inverse(epsilon = var_2426_epsilon_0, x = s_95_cast_fp16)[name = string("op_2426_cast_fp16")];
            fp16 var_2427_to_fp16 = const()[name = string("op_2427_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 4, 1]> var_2428_cast_fp16 = mul(x = var_2426_cast_fp16, y = var_2427_to_fp16)[name = string("op_2428_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_2429_cast_fp16 = real_div(x = var_2428_cast_fp16, y = s_95_cast_fp16)[name = string("op_2429_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_2430_cast_fp16 = add(x = var_2425_cast_fp16, y = var_2429_cast_fp16)[name = string("op_2430_cast_fp16")];
            fp32 var_2431_epsilon_0 = const()[name = string("op_2431_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 4, 1]> var_2431_cast_fp16 = rsqrt(epsilon = var_2431_epsilon_0, x = var_2430_cast_fp16)[name = string("op_2431_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_2432_cast_fp16 = mul(x = y_105_cast_fp16, y = var_2431_cast_fp16)[name = string("op_2432_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_15_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197222336)))];
            tensor<fp16, [1, 375, 4, 96]> var_2433_cast_fp16 = mul(x = core_backbone_layers_15_self_attn_k_norm_weight_to_fp16, y = var_2432_cast_fp16)[name = string("op_2433_cast_fp16")];
            tensor<int32, [4]> new_41_perm_0 = const()[name = string("new_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197222592)))];
            tensor<fp16, [1, 375, 384]> linear_87_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_v_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_87_cast_fp16")];
            tensor<int32, [4]> concat_43 = const()[name = string("concat_43"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> var_2438_cast_fp16 = reshape(shape = concat_43, x = linear_87_cast_fp16)[name = string("op_2438_cast_fp16")];
            tensor<int32, [4]> new_43_perm_0 = const()[name = string("new_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_217_begin_0 = const()[name = string("input_217_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])];
            tensor<int32, [5]> input_217_end_0 = const()[name = string("input_217_end_0"), val = tensor<int32, [5]>([6, 1, 4, 31, 96])];
            tensor<bool, [5]> input_217_end_mask_0 = const()[name = string("input_217_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_217_squeeze_mask_0 = const()[name = string("input_217_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_217_cast_fp16 = slice_by_index(begin = input_217_begin_0, end = input_217_end_0, end_mask = input_217_end_mask_0, squeeze_mask = input_217_squeeze_mask_0, x = k)[name = string("input_217_cast_fp16")];
            tensor<int32, [8]> var_2443_pad_0 = const()[name = string("op_2443_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2443_mode_0 = const()[name = string("op_2443_mode_0"), val = string("constant")];
            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_41_cast_fp16 = transpose(perm = new_41_perm_0, x = var_2433_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 4, 406, 96]> var_2443_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = var_2443_mode_0, pad = var_2443_pad_0, x = new_41_cast_fp16)[name = string("op_2443_cast_fp16")];
            tensor<int32, [8]> var_2445_pad_0 = const()[name = string("op_2445_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_2445_mode_0 = const()[name = string("op_2445_mode_0"), val = string("constant")];
            fp16 const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_2445_cast_fp16 = pad(constant_val = const_55_to_fp16, mode = var_2445_mode_0, pad = var_2445_pad_0, x = input_217_cast_fp16)[name = string("op_2445_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> x_175_cast_fp16 = add(x = var_2443_cast_fp16, y = var_2445_cast_fp16)[name = string("x_175_cast_fp16")];
            tensor<int32, [5]> input_221_begin_0 = const()[name = string("input_221_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])];
            tensor<int32, [5]> input_221_end_0 = const()[name = string("input_221_end_0"), val = tensor<int32, [5]>([6, 1, 4, 31, 96])];
            tensor<bool, [5]> input_221_end_mask_0 = const()[name = string("input_221_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_221_squeeze_mask_0 = const()[name = string("input_221_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_221_cast_fp16 = slice_by_index(begin = input_221_begin_0, end = input_221_end_0, end_mask = input_221_end_mask_0, squeeze_mask = input_221_squeeze_mask_0, x = v)[name = string("input_221_cast_fp16")];
            tensor<int32, [8]> var_2450_pad_0 = const()[name = string("op_2450_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2450_mode_0 = const()[name = string("op_2450_mode_0"), val = string("constant")];
            fp16 const_56_to_fp16 = const()[name = string("const_56_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_43_cast_fp16 = transpose(perm = new_43_perm_0, x = var_2438_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 4, 406, 96]> var_2450_cast_fp16 = pad(constant_val = const_56_to_fp16, mode = var_2450_mode_0, pad = var_2450_pad_0, x = new_43_cast_fp16)[name = string("op_2450_cast_fp16")];
            tensor<int32, [8]> var_2452_pad_0 = const()[name = string("op_2452_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_2452_mode_0 = const()[name = string("op_2452_mode_0"), val = string("constant")];
            fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_2452_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = var_2452_mode_0, pad = var_2452_pad_0, x = input_221_cast_fp16)[name = string("op_2452_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> v_23_cast_fp16 = add(x = var_2450_cast_fp16, y = var_2452_cast_fp16)[name = string("v_23_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2454_cast_fp16 = mul(x = var_2411_cast_fp16, y = const_66_to_fp16)[name = string("op_2454_cast_fp16")];
            tensor<int32, [2]> var_2455_split_sizes_0 = const()[name = string("op_2455_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_87 = const()[name = string("const_87"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 48]> var_2455_cast_fp16_0, tensor<fp16, [1, 375, 8, 48]> var_2455_cast_fp16_1 = split(axis = const_87, split_sizes = var_2455_split_sizes_0, x = var_2411_cast_fp16)[name = string("op_2455_cast_fp16")];
            fp16 const_58_promoted_to_fp16 = const()[name = string("const_58_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 375, 8, 48]> var_2457_cast_fp16 = mul(x = var_2455_cast_fp16_1, y = const_58_promoted_to_fp16)[name = string("op_2457_cast_fp16")];
            bool var_2459_interleave_0 = const()[name = string("op_2459_interleave_0"), val = bool(false)];
            int32 const_88 = const()[name = string("const_88"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 96]> var_2459_cast_fp16 = concat(axis = const_88, interleave = var_2459_interleave_0, values = (var_2457_cast_fp16, var_2455_cast_fp16_0))[name = string("op_2459_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2460_cast_fp16 = mul(x = var_2459_cast_fp16, y = const_69_to_fp16)[name = string("op_2460_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> q_11_cast_fp16 = add(x = var_2454_cast_fp16, y = var_2460_cast_fp16)[name = string("q_11_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_2462_cast_fp16 = mul(x = x_175_cast_fp16, y = const_0_to_fp16)[name = string("op_2462_cast_fp16")];
            tensor<int32, [2]> var_2463_split_sizes_0 = const()[name = string("op_2463_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2463_axis_0 = const()[name = string("op_2463_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 406, 48]> var_2463_cast_fp16_0, tensor<fp16, [1, 4, 406, 48]> var_2463_cast_fp16_1 = split(axis = var_2463_axis_0, split_sizes = var_2463_split_sizes_0, x = x_175_cast_fp16)[name = string("op_2463_cast_fp16")];
            fp16 const_59_promoted_to_fp16 = const()[name = string("const_59_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 406, 48]> var_2465_cast_fp16 = mul(x = var_2463_cast_fp16_1, y = const_59_promoted_to_fp16)[name = string("op_2465_cast_fp16")];
            bool var_2467_interleave_0 = const()[name = string("op_2467_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 406, 96]> var_2467_cast_fp16 = concat(axis = var_127, interleave = var_2467_interleave_0, values = (var_2465_cast_fp16, var_2463_cast_fp16_0))[name = string("op_2467_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_2468_cast_fp16 = mul(x = var_2467_cast_fp16, y = const_1_to_fp16)[name = string("op_2468_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> k_23_cast_fp16 = add(x = var_2462_cast_fp16, y = var_2468_cast_fp16)[name = string("k_23_cast_fp16")];
            tensor<int32, [1]> var_2470_axes_0 = const()[name = string("op_2470_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_2470_cast_fp16 = expand_dims(axes = var_2470_axes_0, x = k_23_cast_fp16)[name = string("op_2470_cast_fp16")];
            tensor<int32, [5]> var_2472_reps_0 = const()[name = string("op_2472_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_2472_cast_fp16 = tile(reps = var_2472_reps_0, x = var_2470_cast_fp16)[name = string("op_2472_cast_fp16")];
            tensor<int32, [4]> concat_45 = const()[name = string("concat_45"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> k_25_cast_fp16 = reshape(shape = concat_45, x = var_2472_cast_fp16)[name = string("k_25_cast_fp16")];
            tensor<int32, [1]> var_2474_axes_0 = const()[name = string("op_2474_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_2474_cast_fp16 = expand_dims(axes = var_2474_axes_0, x = v_23_cast_fp16)[name = string("op_2474_cast_fp16")];
            tensor<int32, [5]> var_2476_reps_0 = const()[name = string("op_2476_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_2476_cast_fp16 = tile(reps = var_2476_reps_0, x = var_2474_cast_fp16)[name = string("op_2476_cast_fp16")];
            tensor<int32, [4]> concat_47 = const()[name = string("concat_47"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> v_25_cast_fp16 = reshape(shape = concat_47, x = var_2476_cast_fp16)[name = string("v_25_cast_fp16")];
            tensor<int32, [4]> transpose_10_perm_0 = const()[name = string("transpose_10_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2479_transpose_x_1 = const()[name = string("op_2479_transpose_x_1"), val = bool(false)];
            bool var_2479_transpose_y_1 = const()[name = string("op_2479_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 96]> transpose_10_cast_fp16 = transpose(perm = transpose_10_perm_0, x = q_11_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 8, 375, 406]> var_2479_cast_fp16 = matmul(transpose_x = var_2479_transpose_x_1, transpose_y = var_2479_transpose_y_1, x = transpose_10_cast_fp16, y = k_25_cast_fp16)[name = string("op_2479_cast_fp16")];
            fp16 var_2480_to_fp16 = const()[name = string("op_2480_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 375, 406]> s_97_cast_fp16 = mul(x = var_2479_cast_fp16, y = var_2480_to_fp16)[name = string("s_97_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_177_cast_fp16 = add(x = s_97_cast_fp16, y = mask)[name = string("x_177_cast_fp16")];
            tensor<int32, [1]> var_2484_axes_0 = const()[name = string("op_2484_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 375, 1]> var_2484_cast_fp16 = reduce_max(axes = var_2484_axes_0, keep_dims = var_128, x = x_177_cast_fp16)[name = string("op_2484_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_179_cast_fp16 = sub(x = x_177_cast_fp16, y = var_2484_cast_fp16)[name = string("x_179_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_181_cast_fp16 = exp(x = x_179_cast_fp16)[name = string("x_181_cast_fp16")];
            tensor<int32, [1]> var_2488_axes_0 = const()[name = string("op_2488_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2488_keep_dims_0 = const()[name = string("op_2488_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 1]> var_2488_cast_fp16 = reduce_sum(axes = var_2488_axes_0, keep_dims = var_2488_keep_dims_0, x = x_181_cast_fp16)[name = string("op_2488_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> var_2489_cast_fp16 = real_div(x = x_181_cast_fp16, y = var_2488_cast_fp16)[name = string("op_2489_cast_fp16")];
            bool o_21_transpose_x_0 = const()[name = string("o_21_transpose_x_0"), val = bool(false)];
            bool o_21_transpose_y_0 = const()[name = string("o_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 375, 96]> o_21_cast_fp16 = matmul(transpose_x = o_21_transpose_x_0, transpose_y = o_21_transpose_y_0, x = var_2489_cast_fp16, y = v_25_cast_fp16)[name = string("o_21_cast_fp16")];
            tensor<int32, [4]> o_23_perm_0 = const()[name = string("o_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_48 = const()[name = string("concat_48"), val = tensor<int32, [3]>([1, 375, 768])];
            tensor<fp16, [1, 375, 8, 96]> o_23_cast_fp16 = transpose(perm = o_23_perm_0, x = o_21_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 375, 768]> input_223_cast_fp16 = reshape(shape = concat_48, x = o_23_cast_fp16)[name = string("input_223_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197812480)))];
            tensor<fp16, [1, 375, 768]> linear_88_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_o_proj_weight_to_fp16, x = input_223_cast_fp16)[name = string("linear_88_cast_fp16")];
            tensor<int32, [4]> var_2498_begin_0 = const()[name = string("op_2498_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_2498_end_0 = const()[name = string("op_2498_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_2498_end_mask_0 = const()[name = string("op_2498_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2498_cast_fp16 = slice_by_index(begin = var_2498_begin_0, end = var_2498_end_0, end_mask = var_2498_end_mask_0, x = x_175_cast_fp16)[name = string("op_2498_cast_fp16")];
            tensor<int32, [4]> var_2501_begin_0 = const()[name = string("op_2501_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_2501_end_0 = const()[name = string("op_2501_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_2501_end_mask_0 = const()[name = string("op_2501_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2501_cast_fp16 = slice_by_index(begin = var_2501_begin_0, end = var_2501_end_0, end_mask = var_2501_end_mask_0, x = v_23_cast_fp16)[name = string("op_2501_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_183_cast_fp16 = add(x = x_167_cast_fp16, y = linear_88_cast_fp16)[name = string("x_183_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2504_cast_fp16 = abs(x = x_183_cast_fp16)[name = string("op_2504_cast_fp16")];
            tensor<int32, [1]> var_2506_axes_0 = const()[name = string("op_2506_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2506_cast_fp16 = reduce_max(axes = var_2506_axes_0, keep_dims = var_128, x = var_2504_cast_fp16)[name = string("op_2506_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_99_cast_fp16 = maximum(x = var_2506_cast_fp16, y = var_129_to_fp16)[name = string("s_99_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_107_cast_fp16 = real_div(x = x_183_cast_fp16, y = s_99_cast_fp16)[name = string("y_107_cast_fp16")];
            fp16 var_130_promoted_43_to_fp16 = const()[name = string("op_130_promoted_43_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2509_cast_fp16 = pow(x = y_107_cast_fp16, y = var_130_promoted_43_to_fp16)[name = string("op_2509_cast_fp16")];
            tensor<int32, [1]> var_2511_axes_0 = const()[name = string("op_2511_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2511_keep_dims_0 = const()[name = string("op_2511_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2511_cast_fp16 = reduce_mean(axes = var_2511_axes_0, keep_dims = var_2511_keep_dims_0, x = var_2509_cast_fp16)[name = string("op_2511_cast_fp16")];
            fp32 var_2512_epsilon_0 = const()[name = string("op_2512_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2512_cast_fp16 = inverse(epsilon = var_2512_epsilon_0, x = s_99_cast_fp16)[name = string("op_2512_cast_fp16")];
            fp16 var_2513_to_fp16 = const()[name = string("op_2513_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2514_cast_fp16 = mul(x = var_2512_cast_fp16, y = var_2513_to_fp16)[name = string("op_2514_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2515_cast_fp16 = real_div(x = var_2514_cast_fp16, y = s_99_cast_fp16)[name = string("op_2515_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2516_cast_fp16 = add(x = var_2511_cast_fp16, y = var_2515_cast_fp16)[name = string("op_2516_cast_fp16")];
            fp32 var_2517_epsilon_0 = const()[name = string("op_2517_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2517_cast_fp16 = rsqrt(epsilon = var_2517_epsilon_0, x = var_2516_cast_fp16)[name = string("op_2517_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2518_cast_fp16 = mul(x = y_107_cast_fp16, y = var_2517_cast_fp16)[name = string("op_2518_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_15_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198992192)))];
            tensor<fp16, [1, 375, 768]> input_225_cast_fp16 = mul(x = core_backbone_layers_15_ffn_norm_weight_to_fp16, y = var_2518_cast_fp16)[name = string("input_225_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_15_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198993792)))];
            tensor<fp16, [1, 375, 1792]> linear_89_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w1_weight_to_fp16, x = input_225_cast_fp16)[name = string("linear_89_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_15_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201746368)))];
            tensor<fp16, [1, 375, 1792]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w3_weight_to_fp16, x = input_225_cast_fp16)[name = string("linear_90_cast_fp16")];
            fp16 var_2527_to_fp16 = const()[name = string("op_2527_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_2528_cast_fp16 = mul(x = linear_89_cast_fp16, y = var_2527_to_fp16)[name = string("op_2528_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2531_cast_fp16 = tanh(x = var_2528_cast_fp16)[name = string("op_2531_cast_fp16")];
            fp16 var_2532_to_fp16 = const()[name = string("op_2532_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_2533_cast_fp16 = add(x = var_2531_cast_fp16, y = var_2532_to_fp16)[name = string("op_2533_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2534_cast_fp16 = mul(x = var_2528_cast_fp16, y = var_2533_cast_fp16)[name = string("op_2534_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_227_cast_fp16 = mul(x = var_2534_cast_fp16, y = linear_90_cast_fp16)[name = string("input_227_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_15_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204498944)))];
            tensor<fp16, [1, 375, 768]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w2_weight_to_fp16, x = input_227_cast_fp16)[name = string("linear_91_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_187_cast_fp16 = add(x = x_183_cast_fp16, y = linear_91_cast_fp16)[name = string("x_187_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2540_cast_fp16 = abs(x = x_187_cast_fp16)[name = string("op_2540_cast_fp16")];
            tensor<int32, [1]> var_2542_axes_0 = const()[name = string("op_2542_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2542_cast_fp16 = reduce_max(axes = var_2542_axes_0, keep_dims = var_128, x = var_2540_cast_fp16)[name = string("op_2542_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_101_cast_fp16 = maximum(x = var_2542_cast_fp16, y = var_129_to_fp16)[name = string("s_101_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_109_cast_fp16 = real_div(x = x_187_cast_fp16, y = s_101_cast_fp16)[name = string("y_109_cast_fp16")];
            fp16 var_130_promoted_44_to_fp16 = const()[name = string("op_130_promoted_44_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2545_cast_fp16 = pow(x = y_109_cast_fp16, y = var_130_promoted_44_to_fp16)[name = string("op_2545_cast_fp16")];
            tensor<int32, [1]> var_2547_axes_0 = const()[name = string("op_2547_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2547_keep_dims_0 = const()[name = string("op_2547_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2547_cast_fp16 = reduce_mean(axes = var_2547_axes_0, keep_dims = var_2547_keep_dims_0, x = var_2545_cast_fp16)[name = string("op_2547_cast_fp16")];
            fp32 var_2548_epsilon_0 = const()[name = string("op_2548_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2548_cast_fp16 = inverse(epsilon = var_2548_epsilon_0, x = s_101_cast_fp16)[name = string("op_2548_cast_fp16")];
            fp16 var_2549_to_fp16 = const()[name = string("op_2549_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2550_cast_fp16 = mul(x = var_2548_cast_fp16, y = var_2549_to_fp16)[name = string("op_2550_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2551_cast_fp16 = real_div(x = var_2550_cast_fp16, y = s_101_cast_fp16)[name = string("op_2551_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2552_cast_fp16 = add(x = var_2547_cast_fp16, y = var_2551_cast_fp16)[name = string("op_2552_cast_fp16")];
            fp32 var_2553_epsilon_0 = const()[name = string("op_2553_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2553_cast_fp16 = rsqrt(epsilon = var_2553_epsilon_0, x = var_2552_cast_fp16)[name = string("op_2553_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2554_cast_fp16 = mul(x = y_109_cast_fp16, y = var_2553_cast_fp16)[name = string("op_2554_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_16_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207251520)))];
            tensor<fp16, [1, 375, 768]> input_229_cast_fp16 = mul(x = core_backbone_layers_16_operator_norm_weight_to_fp16, y = var_2554_cast_fp16)[name = string("input_229_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207253120)))];
            tensor<fp16, [1, 375, 768]> linear_92_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_q_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_92_cast_fp16")];
            tensor<int32, [4]> concat_49 = const()[name = string("concat_49"), val = tensor<int32, [4]>([1, 375, 8, 96])];
            tensor<fp16, [1, 375, 8, 96]> x_189_cast_fp16 = reshape(shape = concat_49, x = linear_92_cast_fp16)[name = string("x_189_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2561_cast_fp16 = abs(x = x_189_cast_fp16)[name = string("op_2561_cast_fp16")];
            tensor<int32, [1]> var_2563_axes_0 = const()[name = string("op_2563_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 8, 1]> var_2563_cast_fp16 = reduce_max(axes = var_2563_axes_0, keep_dims = var_128, x = var_2561_cast_fp16)[name = string("op_2563_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> s_103_cast_fp16 = maximum(x = var_2563_cast_fp16, y = var_129_to_fp16)[name = string("s_103_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> y_111_cast_fp16 = real_div(x = x_189_cast_fp16, y = s_103_cast_fp16)[name = string("y_111_cast_fp16")];
            fp16 var_130_promoted_45_to_fp16 = const()[name = string("op_130_promoted_45_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 8, 96]> var_2566_cast_fp16 = pow(x = y_111_cast_fp16, y = var_130_promoted_45_to_fp16)[name = string("op_2566_cast_fp16")];
            tensor<int32, [1]> var_2568_axes_0 = const()[name = string("op_2568_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2568_keep_dims_0 = const()[name = string("op_2568_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 8, 1]> var_2568_cast_fp16 = reduce_mean(axes = var_2568_axes_0, keep_dims = var_2568_keep_dims_0, x = var_2566_cast_fp16)[name = string("op_2568_cast_fp16")];
            fp32 var_2569_epsilon_0 = const()[name = string("op_2569_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 8, 1]> var_2569_cast_fp16 = inverse(epsilon = var_2569_epsilon_0, x = s_103_cast_fp16)[name = string("op_2569_cast_fp16")];
            fp16 var_2570_to_fp16 = const()[name = string("op_2570_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 8, 1]> var_2571_cast_fp16 = mul(x = var_2569_cast_fp16, y = var_2570_to_fp16)[name = string("op_2571_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_2572_cast_fp16 = real_div(x = var_2571_cast_fp16, y = s_103_cast_fp16)[name = string("op_2572_cast_fp16")];
            tensor<fp16, [1, 375, 8, 1]> var_2573_cast_fp16 = add(x = var_2568_cast_fp16, y = var_2572_cast_fp16)[name = string("op_2573_cast_fp16")];
            fp32 var_2574_epsilon_0 = const()[name = string("op_2574_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 8, 1]> var_2574_cast_fp16 = rsqrt(epsilon = var_2574_epsilon_0, x = var_2573_cast_fp16)[name = string("op_2574_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2575_cast_fp16 = mul(x = y_111_cast_fp16, y = var_2574_cast_fp16)[name = string("op_2575_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_16_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208432832)))];
            tensor<fp16, [1, 375, 8, 96]> var_2576_cast_fp16 = mul(x = core_backbone_layers_16_self_attn_q_norm_weight_to_fp16, y = var_2575_cast_fp16)[name = string("op_2576_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208433088)))];
            tensor<fp16, [1, 375, 384]> linear_93_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_k_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_93_cast_fp16")];
            tensor<int32, [4]> concat_50 = const()[name = string("concat_50"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> x_191_cast_fp16 = reshape(shape = concat_50, x = linear_93_cast_fp16)[name = string("x_191_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_2583_cast_fp16 = abs(x = x_191_cast_fp16)[name = string("op_2583_cast_fp16")];
            tensor<int32, [1]> var_2585_axes_0 = const()[name = string("op_2585_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 4, 1]> var_2585_cast_fp16 = reduce_max(axes = var_2585_axes_0, keep_dims = var_128, x = var_2583_cast_fp16)[name = string("op_2585_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> s_105_cast_fp16 = maximum(x = var_2585_cast_fp16, y = var_129_to_fp16)[name = string("s_105_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> y_113_cast_fp16 = real_div(x = x_191_cast_fp16, y = s_105_cast_fp16)[name = string("y_113_cast_fp16")];
            fp16 var_130_promoted_46_to_fp16 = const()[name = string("op_130_promoted_46_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 4, 96]> var_2588_cast_fp16 = pow(x = y_113_cast_fp16, y = var_130_promoted_46_to_fp16)[name = string("op_2588_cast_fp16")];
            tensor<int32, [1]> var_2590_axes_0 = const()[name = string("op_2590_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2590_keep_dims_0 = const()[name = string("op_2590_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 4, 1]> var_2590_cast_fp16 = reduce_mean(axes = var_2590_axes_0, keep_dims = var_2590_keep_dims_0, x = var_2588_cast_fp16)[name = string("op_2590_cast_fp16")];
            fp32 var_2591_epsilon_0 = const()[name = string("op_2591_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 4, 1]> var_2591_cast_fp16 = inverse(epsilon = var_2591_epsilon_0, x = s_105_cast_fp16)[name = string("op_2591_cast_fp16")];
            fp16 var_2592_to_fp16 = const()[name = string("op_2592_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 4, 1]> var_2593_cast_fp16 = mul(x = var_2591_cast_fp16, y = var_2592_to_fp16)[name = string("op_2593_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_2594_cast_fp16 = real_div(x = var_2593_cast_fp16, y = s_105_cast_fp16)[name = string("op_2594_cast_fp16")];
            tensor<fp16, [1, 375, 4, 1]> var_2595_cast_fp16 = add(x = var_2590_cast_fp16, y = var_2594_cast_fp16)[name = string("op_2595_cast_fp16")];
            fp32 var_2596_epsilon_0 = const()[name = string("op_2596_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 4, 1]> var_2596_cast_fp16 = rsqrt(epsilon = var_2596_epsilon_0, x = var_2595_cast_fp16)[name = string("op_2596_cast_fp16")];
            tensor<fp16, [1, 375, 4, 96]> var_2597_cast_fp16 = mul(x = y_113_cast_fp16, y = var_2596_cast_fp16)[name = string("op_2597_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_16_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209022976)))];
            tensor<fp16, [1, 375, 4, 96]> var_2598_cast_fp16 = mul(x = core_backbone_layers_16_self_attn_k_norm_weight_to_fp16, y = var_2597_cast_fp16)[name = string("op_2598_cast_fp16")];
            tensor<int32, [4]> new_45_perm_0 = const()[name = string("new_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209023232)))];
            tensor<fp16, [1, 375, 384]> linear_94_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_v_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_94_cast_fp16")];
            tensor<int32, [4]> concat_51 = const()[name = string("concat_51"), val = tensor<int32, [4]>([1, 375, 4, 96])];
            tensor<fp16, [1, 375, 4, 96]> var_2603_cast_fp16 = reshape(shape = concat_51, x = linear_94_cast_fp16)[name = string("op_2603_cast_fp16")];
            tensor<int32, [4]> new_perm_0 = const()[name = string("new_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_233_begin_0 = const()[name = string("input_233_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])];
            tensor<int32, [5]> input_233_end_0 = const()[name = string("input_233_end_0"), val = tensor<int32, [5]>([7, 1, 4, 31, 96])];
            tensor<bool, [5]> input_233_end_mask_0 = const()[name = string("input_233_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_233_squeeze_mask_0 = const()[name = string("input_233_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_233_cast_fp16 = slice_by_index(begin = input_233_begin_0, end = input_233_end_0, end_mask = input_233_end_mask_0, squeeze_mask = input_233_squeeze_mask_0, x = k)[name = string("input_233_cast_fp16")];
            tensor<int32, [8]> var_2608_pad_0 = const()[name = string("op_2608_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2608_mode_0 = const()[name = string("op_2608_mode_0"), val = string("constant")];
            fp16 const_60_to_fp16 = const()[name = string("const_60_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_45_cast_fp16 = transpose(perm = new_45_perm_0, x = var_2598_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 4, 406, 96]> var_2608_cast_fp16 = pad(constant_val = const_60_to_fp16, mode = var_2608_mode_0, pad = var_2608_pad_0, x = new_45_cast_fp16)[name = string("op_2608_cast_fp16")];
            tensor<int32, [8]> var_2610_pad_0 = const()[name = string("op_2610_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_2610_mode_0 = const()[name = string("op_2610_mode_0"), val = string("constant")];
            fp16 const_61_to_fp16 = const()[name = string("const_61_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_2610_cast_fp16 = pad(constant_val = const_61_to_fp16, mode = var_2610_mode_0, pad = var_2610_pad_0, x = input_233_cast_fp16)[name = string("op_2610_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> x_195_cast_fp16 = add(x = var_2608_cast_fp16, y = var_2610_cast_fp16)[name = string("x_195_cast_fp16")];
            tensor<int32, [5]> input_237_begin_0 = const()[name = string("input_237_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])];
            tensor<int32, [5]> input_237_end_0 = const()[name = string("input_237_end_0"), val = tensor<int32, [5]>([7, 1, 4, 31, 96])];
            tensor<bool, [5]> input_237_end_mask_0 = const()[name = string("input_237_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_237_squeeze_mask_0 = const()[name = string("input_237_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_237_cast_fp16 = slice_by_index(begin = input_237_begin_0, end = input_237_end_0, end_mask = input_237_end_mask_0, squeeze_mask = input_237_squeeze_mask_0, x = v)[name = string("input_237_cast_fp16")];
            tensor<int32, [8]> var_2615_pad_0 = const()[name = string("op_2615_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2615_mode_0 = const()[name = string("op_2615_mode_0"), val = string("constant")];
            fp16 const_62_to_fp16 = const()[name = string("const_62_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 375, 96]> new_cast_fp16 = transpose(perm = new_perm_0, x = var_2603_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 4, 406, 96]> var_2615_cast_fp16 = pad(constant_val = const_62_to_fp16, mode = var_2615_mode_0, pad = var_2615_pad_0, x = new_cast_fp16)[name = string("op_2615_cast_fp16")];
            tensor<int32, [8]> var_2617_pad_0 = const()[name = string("op_2617_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 375, 0, 0])];
            string var_2617_mode_0 = const()[name = string("op_2617_mode_0"), val = string("constant")];
            fp16 const_63_to_fp16 = const()[name = string("const_63_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 406, 96]> var_2617_cast_fp16 = pad(constant_val = const_63_to_fp16, mode = var_2617_mode_0, pad = var_2617_pad_0, x = input_237_cast_fp16)[name = string("op_2617_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> v_27_cast_fp16 = add(x = var_2615_cast_fp16, y = var_2617_cast_fp16)[name = string("v_27_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2619_cast_fp16 = mul(x = var_2576_cast_fp16, y = const_66_to_fp16)[name = string("op_2619_cast_fp16")];
            tensor<int32, [2]> var_2620_split_sizes_0 = const()[name = string("op_2620_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_91 = const()[name = string("const_91"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 48]> var_2620_cast_fp16_0, tensor<fp16, [1, 375, 8, 48]> var_2620_cast_fp16_1 = split(axis = const_91, split_sizes = var_2620_split_sizes_0, x = var_2576_cast_fp16)[name = string("op_2620_cast_fp16")];
            fp16 const_64_promoted_to_fp16 = const()[name = string("const_64_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 375, 8, 48]> var_2622_cast_fp16 = mul(x = var_2620_cast_fp16_1, y = const_64_promoted_to_fp16)[name = string("op_2622_cast_fp16")];
            bool var_2624_interleave_0 = const()[name = string("op_2624_interleave_0"), val = bool(false)];
            int32 const_92 = const()[name = string("const_92"), val = int32(3)];
            tensor<fp16, [1, 375, 8, 96]> var_2624_cast_fp16 = concat(axis = const_92, interleave = var_2624_interleave_0, values = (var_2622_cast_fp16, var_2620_cast_fp16_0))[name = string("op_2624_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> var_2625_cast_fp16 = mul(x = var_2624_cast_fp16, y = const_69_to_fp16)[name = string("op_2625_cast_fp16")];
            tensor<fp16, [1, 375, 8, 96]> q_cast_fp16 = add(x = var_2619_cast_fp16, y = var_2625_cast_fp16)[name = string("q_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_2627_cast_fp16 = mul(x = x_195_cast_fp16, y = const_0_to_fp16)[name = string("op_2627_cast_fp16")];
            tensor<int32, [2]> var_2628_split_sizes_0 = const()[name = string("op_2628_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2628_axis_0 = const()[name = string("op_2628_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 406, 48]> var_2628_cast_fp16_0, tensor<fp16, [1, 4, 406, 48]> var_2628_cast_fp16_1 = split(axis = var_2628_axis_0, split_sizes = var_2628_split_sizes_0, x = x_195_cast_fp16)[name = string("op_2628_cast_fp16")];
            fp16 const_65_promoted_to_fp16 = const()[name = string("const_65_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 406, 48]> var_2630_cast_fp16 = mul(x = var_2628_cast_fp16_1, y = const_65_promoted_to_fp16)[name = string("op_2630_cast_fp16")];
            bool var_2632_interleave_0 = const()[name = string("op_2632_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 406, 96]> var_2632_cast_fp16 = concat(axis = var_127, interleave = var_2632_interleave_0, values = (var_2630_cast_fp16, var_2628_cast_fp16_0))[name = string("op_2632_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> var_2633_cast_fp16 = mul(x = var_2632_cast_fp16, y = const_1_to_fp16)[name = string("op_2633_cast_fp16")];
            tensor<fp16, [1, 4, 406, 96]> k_27_cast_fp16 = add(x = var_2627_cast_fp16, y = var_2633_cast_fp16)[name = string("k_27_cast_fp16")];
            tensor<int32, [1]> var_2635_axes_0 = const()[name = string("op_2635_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_2635_cast_fp16 = expand_dims(axes = var_2635_axes_0, x = k_27_cast_fp16)[name = string("op_2635_cast_fp16")];
            tensor<int32, [5]> var_2637_reps_0 = const()[name = string("op_2637_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_2637_cast_fp16 = tile(reps = var_2637_reps_0, x = var_2635_cast_fp16)[name = string("op_2637_cast_fp16")];
            tensor<int32, [4]> concat_53 = const()[name = string("concat_53"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> k_cast_fp16 = reshape(shape = concat_53, x = var_2637_cast_fp16)[name = string("k_cast_fp16")];
            tensor<int32, [1]> var_2639_axes_0 = const()[name = string("op_2639_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 406, 96]> var_2639_cast_fp16 = expand_dims(axes = var_2639_axes_0, x = v_27_cast_fp16)[name = string("op_2639_cast_fp16")];
            tensor<int32, [5]> var_2641_reps_0 = const()[name = string("op_2641_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 406, 96]> var_2641_cast_fp16 = tile(reps = var_2641_reps_0, x = var_2639_cast_fp16)[name = string("op_2641_cast_fp16")];
            tensor<int32, [4]> concat_55 = const()[name = string("concat_55"), val = tensor<int32, [4]>([1, 8, 406, 96])];
            tensor<fp16, [1, 8, 406, 96]> v_cast_fp16 = reshape(shape = concat_55, x = var_2641_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [4]> transpose_12_perm_0 = const()[name = string("transpose_12_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2644_transpose_x_1 = const()[name = string("op_2644_transpose_x_1"), val = bool(false)];
            bool var_2644_transpose_y_1 = const()[name = string("op_2644_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 96]> transpose_12_cast_fp16 = transpose(perm = transpose_12_perm_0, x = q_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 8, 375, 406]> var_2644_cast_fp16 = matmul(transpose_x = var_2644_transpose_x_1, transpose_y = var_2644_transpose_y_1, x = transpose_12_cast_fp16, y = k_cast_fp16)[name = string("op_2644_cast_fp16")];
            fp16 var_2645_to_fp16 = const()[name = string("op_2645_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 375, 406]> s_107_cast_fp16 = mul(x = var_2644_cast_fp16, y = var_2645_to_fp16)[name = string("s_107_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_197_cast_fp16 = add(x = s_107_cast_fp16, y = mask)[name = string("x_197_cast_fp16")];
            tensor<int32, [1]> var_2649_axes_0 = const()[name = string("op_2649_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 375, 1]> var_2649_cast_fp16 = reduce_max(axes = var_2649_axes_0, keep_dims = var_128, x = x_197_cast_fp16)[name = string("op_2649_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_199_cast_fp16 = sub(x = x_197_cast_fp16, y = var_2649_cast_fp16)[name = string("x_199_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> x_201_cast_fp16 = exp(x = x_199_cast_fp16)[name = string("x_201_cast_fp16")];
            tensor<int32, [1]> var_2653_axes_0 = const()[name = string("op_2653_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2653_keep_dims_0 = const()[name = string("op_2653_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 375, 1]> var_2653_cast_fp16 = reduce_sum(axes = var_2653_axes_0, keep_dims = var_2653_keep_dims_0, x = x_201_cast_fp16)[name = string("op_2653_cast_fp16")];
            tensor<fp16, [1, 8, 375, 406]> var_2654_cast_fp16 = real_div(x = x_201_cast_fp16, y = var_2653_cast_fp16)[name = string("op_2654_cast_fp16")];
            bool o_25_transpose_x_0 = const()[name = string("o_25_transpose_x_0"), val = bool(false)];
            bool o_25_transpose_y_0 = const()[name = string("o_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 375, 96]> o_25_cast_fp16 = matmul(transpose_x = o_25_transpose_x_0, transpose_y = o_25_transpose_y_0, x = var_2654_cast_fp16, y = v_cast_fp16)[name = string("o_25_cast_fp16")];
            tensor<int32, [4]> o_perm_0 = const()[name = string("o_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_56 = const()[name = string("concat_56"), val = tensor<int32, [3]>([1, 375, 768])];
            tensor<fp16, [1, 375, 8, 96]> o_cast_fp16 = transpose(perm = o_perm_0, x = o_25_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 375, 768]> input_239_cast_fp16 = reshape(shape = concat_56, x = o_cast_fp16)[name = string("input_239_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209613120)))];
            tensor<fp16, [1, 375, 768]> linear_95_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_o_proj_weight_to_fp16, x = input_239_cast_fp16)[name = string("linear_95_cast_fp16")];
            tensor<int32, [4]> var_2663_begin_0 = const()[name = string("op_2663_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_2663_end_0 = const()[name = string("op_2663_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_2663_end_mask_0 = const()[name = string("op_2663_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2663_cast_fp16 = slice_by_index(begin = var_2663_begin_0, end = var_2663_end_0, end_mask = var_2663_end_mask_0, x = x_195_cast_fp16)[name = string("op_2663_cast_fp16")];
            tensor<int32, [4]> var_2666_begin_0 = const()[name = string("op_2666_begin_0"), val = tensor<int32, [4]>([0, 0, 375, 0])];
            tensor<int32, [4]> var_2666_end_0 = const()[name = string("op_2666_end_0"), val = tensor<int32, [4]>([1, 4, 406, 96])];
            tensor<bool, [4]> var_2666_end_mask_0 = const()[name = string("op_2666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2666_cast_fp16 = slice_by_index(begin = var_2666_begin_0, end = var_2666_end_0, end_mask = var_2666_end_mask_0, x = v_27_cast_fp16)[name = string("op_2666_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_203_cast_fp16 = add(x = x_187_cast_fp16, y = linear_95_cast_fp16)[name = string("x_203_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2669_cast_fp16 = abs(x = x_203_cast_fp16)[name = string("op_2669_cast_fp16")];
            tensor<int32, [1]> var_2671_axes_0 = const()[name = string("op_2671_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2671_cast_fp16 = reduce_max(axes = var_2671_axes_0, keep_dims = var_128, x = var_2669_cast_fp16)[name = string("op_2671_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_109_cast_fp16 = maximum(x = var_2671_cast_fp16, y = var_129_to_fp16)[name = string("s_109_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_115_cast_fp16 = real_div(x = x_203_cast_fp16, y = s_109_cast_fp16)[name = string("y_115_cast_fp16")];
            fp16 var_130_promoted_47_to_fp16 = const()[name = string("op_130_promoted_47_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2674_cast_fp16 = pow(x = y_115_cast_fp16, y = var_130_promoted_47_to_fp16)[name = string("op_2674_cast_fp16")];
            tensor<int32, [1]> var_2676_axes_0 = const()[name = string("op_2676_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2676_keep_dims_0 = const()[name = string("op_2676_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2676_cast_fp16 = reduce_mean(axes = var_2676_axes_0, keep_dims = var_2676_keep_dims_0, x = var_2674_cast_fp16)[name = string("op_2676_cast_fp16")];
            fp32 var_2677_epsilon_0 = const()[name = string("op_2677_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2677_cast_fp16 = inverse(epsilon = var_2677_epsilon_0, x = s_109_cast_fp16)[name = string("op_2677_cast_fp16")];
            fp16 var_2678_to_fp16 = const()[name = string("op_2678_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2679_cast_fp16 = mul(x = var_2677_cast_fp16, y = var_2678_to_fp16)[name = string("op_2679_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2680_cast_fp16 = real_div(x = var_2679_cast_fp16, y = s_109_cast_fp16)[name = string("op_2680_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2681_cast_fp16 = add(x = var_2676_cast_fp16, y = var_2680_cast_fp16)[name = string("op_2681_cast_fp16")];
            fp32 var_2682_epsilon_0 = const()[name = string("op_2682_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2682_cast_fp16 = rsqrt(epsilon = var_2682_epsilon_0, x = var_2681_cast_fp16)[name = string("op_2682_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2683_cast_fp16 = mul(x = y_115_cast_fp16, y = var_2682_cast_fp16)[name = string("op_2683_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_16_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210792832)))];
            tensor<fp16, [1, 375, 768]> input_241_cast_fp16 = mul(x = core_backbone_layers_16_ffn_norm_weight_to_fp16, y = var_2683_cast_fp16)[name = string("input_241_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_16_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210794432)))];
            tensor<fp16, [1, 375, 1792]> linear_96_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w1_weight_to_fp16, x = input_241_cast_fp16)[name = string("linear_96_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_16_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213547008)))];
            tensor<fp16, [1, 375, 1792]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w3_weight_to_fp16, x = input_241_cast_fp16)[name = string("linear_97_cast_fp16")];
            fp16 var_2692_to_fp16 = const()[name = string("op_2692_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 375, 1792]> var_2693_cast_fp16 = mul(x = linear_96_cast_fp16, y = var_2692_to_fp16)[name = string("op_2693_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2696_cast_fp16 = tanh(x = var_2693_cast_fp16)[name = string("op_2696_cast_fp16")];
            fp16 var_2697_to_fp16 = const()[name = string("op_2697_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 375, 1792]> var_2698_cast_fp16 = add(x = var_2696_cast_fp16, y = var_2697_to_fp16)[name = string("op_2698_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> var_2699_cast_fp16 = mul(x = var_2693_cast_fp16, y = var_2698_cast_fp16)[name = string("op_2699_cast_fp16")];
            tensor<fp16, [1, 375, 1792]> input_243_cast_fp16 = mul(x = var_2699_cast_fp16, y = linear_97_cast_fp16)[name = string("input_243_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_16_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216299584)))];
            tensor<fp16, [1, 375, 768]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w2_weight_to_fp16, x = input_243_cast_fp16)[name = string("linear_98_cast_fp16")];
            tensor<fp16, [1, 375, 768]> x_cast_fp16 = add(x = x_203_cast_fp16, y = linear_98_cast_fp16)[name = string("x_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2705_cast_fp16 = abs(x = x_cast_fp16)[name = string("op_2705_cast_fp16")];
            tensor<int32, [1]> var_2707_axes_0 = const()[name = string("op_2707_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 375, 1]> var_2707_cast_fp16 = reduce_max(axes = var_2707_axes_0, keep_dims = var_128, x = var_2705_cast_fp16)[name = string("op_2707_cast_fp16")];
            tensor<fp16, [1, 375, 1]> s_cast_fp16 = maximum(x = var_2707_cast_fp16, y = var_129_to_fp16)[name = string("s_cast_fp16")];
            tensor<fp16, [1, 375, 768]> y_cast_fp16 = real_div(x = x_cast_fp16, y = s_cast_fp16)[name = string("y_cast_fp16")];
            fp16 var_130_promoted_48_to_fp16 = const()[name = string("op_130_promoted_48_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 375, 768]> var_2710_cast_fp16 = pow(x = y_cast_fp16, y = var_130_promoted_48_to_fp16)[name = string("op_2710_cast_fp16")];
            tensor<int32, [1]> var_2712_axes_0 = const()[name = string("op_2712_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2712_keep_dims_0 = const()[name = string("op_2712_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 375, 1]> var_2712_cast_fp16 = reduce_mean(axes = var_2712_axes_0, keep_dims = var_2712_keep_dims_0, x = var_2710_cast_fp16)[name = string("op_2712_cast_fp16")];
            fp32 var_2713_epsilon_0 = const()[name = string("op_2713_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 375, 1]> var_2713_cast_fp16 = inverse(epsilon = var_2713_epsilon_0, x = s_cast_fp16)[name = string("op_2713_cast_fp16")];
            fp16 var_2714_to_fp16 = const()[name = string("op_2714_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 1]> var_2715_cast_fp16 = mul(x = var_2713_cast_fp16, y = var_2714_to_fp16)[name = string("op_2715_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2716_cast_fp16 = real_div(x = var_2715_cast_fp16, y = s_cast_fp16)[name = string("op_2716_cast_fp16")];
            tensor<fp16, [1, 375, 1]> var_2717_cast_fp16 = add(x = var_2712_cast_fp16, y = var_2716_cast_fp16)[name = string("op_2717_cast_fp16")];
            fp32 var_2718_epsilon_0 = const()[name = string("op_2718_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 375, 1]> var_2718_cast_fp16 = rsqrt(epsilon = var_2718_epsilon_0, x = var_2717_cast_fp16)[name = string("op_2718_cast_fp16")];
            tensor<fp16, [1, 375, 768]> var_2719_cast_fp16 = mul(x = y_cast_fp16, y = var_2718_cast_fp16)[name = string("op_2719_cast_fp16")];
            tensor<fp16, [768]> core_backbone_embedding_norm_weight_to_fp16 = const()[name = string("core_backbone_embedding_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219052160)))];
            tensor<fp16, [1, 375, 768]> input_245_cast_fp16 = mul(x = core_backbone_embedding_norm_weight_to_fp16, y = var_2719_cast_fp16)[name = string("input_245_cast_fp16")];
            int32 var_2722_axis_0 = const()[name = string("op_2722_axis_0"), val = int32(0)];
            tensor<fp16, [10, 1, 768, 2]> conv_out = stack(axis = var_2722_axis_0, values = (var_753_cast_fp16, var_836_cast_fp16, var_1084_cast_fp16, var_1167_cast_fp16, var_1415_cast_fp16, var_1498_cast_fp16, var_1746_cast_fp16, var_1829_cast_fp16, var_2077_cast_fp16, var_2160_cast_fp16))[name = string("op_2722_cast_fp16")];
            int32 var_2724_axis_0 = const()[name = string("op_2724_axis_0"), val = int32(0)];
            tensor<fp16, [7, 1, 4, 31, 96]> k_out = stack(axis = var_2724_axis_0, values = (var_1009_cast_fp16, var_1340_cast_fp16, var_1671_cast_fp16, var_2002_cast_fp16, var_2333_cast_fp16, var_2498_cast_fp16, var_2663_cast_fp16))[name = string("op_2724_cast_fp16")];
            int32 var_2726_axis_0 = const()[name = string("op_2726_axis_0"), val = int32(0)];
            tensor<fp16, [7, 1, 4, 31, 96]> v_out = stack(axis = var_2726_axis_0, values = (var_1012_cast_fp16, var_1343_cast_fp16, var_1674_cast_fp16, var_2005_cast_fp16, var_2336_cast_fp16, var_2501_cast_fp16, var_2666_cast_fp16))[name = string("op_2726_cast_fp16")];
            tensor<int32, [1]> input_247_axes_0 = const()[name = string("input_247_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> projector_layer_norm_weight_to_fp16 = const()[name = string("projector_layer_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219053760)))];
            tensor<fp16, [768]> projector_layer_norm_bias_to_fp16 = const()[name = string("projector_layer_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219055360)))];
            fp16 var_2734_to_fp16 = const()[name = string("op_2734_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 375, 768]> input_247_cast_fp16 = layer_norm(axes = input_247_axes_0, beta = projector_layer_norm_bias_to_fp16, epsilon = var_2734_to_fp16, gamma = projector_layer_norm_weight_to_fp16, x = input_245_cast_fp16)[name = string("input_247_cast_fp16")];
            tensor<fp16, [2048, 768]> projector_linear_1_weight_to_fp16 = const()[name = string("projector_linear_1_weight_to_fp16"), val = tensor<fp16, [2048, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219056960)))];
            tensor<fp16, [2048]> projector_linear_1_bias_to_fp16 = const()[name = string("projector_linear_1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222202752)))];
            tensor<fp16, [1, 375, 2048]> linear_99_cast_fp16 = linear(bias = projector_linear_1_bias_to_fp16, weight = projector_linear_1_weight_to_fp16, x = input_247_cast_fp16)[name = string("linear_99_cast_fp16")];
            string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 375, 2048]> input_cast_fp16 = gelu(mode = input_mode_0, x = linear_99_cast_fp16)[name = string("input_cast_fp16")];
            tensor<fp16, [2048, 2048]> projector_linear_2_weight_to_fp16 = const()[name = string("projector_linear_2_weight_to_fp16"), val = tensor<fp16, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222206912)))];
            tensor<fp16, [2048]> projector_linear_2_bias_to_fp16 = const()[name = string("projector_linear_2_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230595584)))];
            tensor<fp16, [1, 375, 2048]> emb = linear(bias = projector_linear_2_bias_to_fp16, weight = projector_linear_2_weight_to_fp16, x = input_cast_fp16)[name = string("linear_100_cast_fp16")];
        } -> (emb, conv_out, k_out, v_out);
    func n96<ios18>(tensor<fp16, [10, 1, 768, 2]> conv, tensor<fp16, [7, 1, 4, 31, 96]> k, tensor<fp16, [1, 1, 1, 388]> m1, tensor<fp16, [1, 1, 1, 194]> m2, tensor<fp16, [1, 1, 96, 127]> mask, tensor<fp16, [1, 128, 776]> mel, tensor<fp16, [7, 1, 4, 31, 96]> v) {
            tensor<int32, [1]> input_1_axes_0 = const()[name = string("input_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 1, 128, 776]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = mel)[name = string("input_1_cast_fp16")];
            string x_1_pad_type_0 = const()[name = string("x_1_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> x_1_pad_0 = const()[name = string("x_1_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> x_1_strides_0 = const()[name = string("x_1_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> x_1_dilations_0 = const()[name = string("x_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_1_groups_0 = const()[name = string("x_1_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_in_weight_to_fp16 = const()[name = string("downsample_conv_in_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [256]> downsample_conv_in_bias_to_fp16 = const()[name = string("downsample_conv_in_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4736)))];
            tensor<fp16, [1, 256, 64, 388]> x_1_cast_fp16 = conv(bias = downsample_conv_in_bias_to_fp16, dilations = x_1_dilations_0, groups = x_1_groups_0, pad = x_1_pad_0, pad_type = x_1_pad_type_0, strides = x_1_strides_0, weight = downsample_conv_in_weight_to_fp16, x = input_1_cast_fp16)[name = string("x_1_cast_fp16")];
            fp16 var_31_to_fp16 = const()[name = string("op_31_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 64, 388]> var_32_cast_fp16 = mul(x = x_1_cast_fp16, y = var_31_to_fp16)[name = string("op_32_cast_fp16")];
            tensor<fp16, [1, 256, 64, 388]> var_35_cast_fp16 = tanh(x = var_32_cast_fp16)[name = string("op_35_cast_fp16")];
            fp16 var_36_to_fp16 = const()[name = string("op_36_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 64, 388]> var_37_cast_fp16 = add(x = var_35_cast_fp16, y = var_36_to_fp16)[name = string("op_37_cast_fp16")];
            tensor<fp16, [1, 256, 64, 388]> h_1_cast_fp16 = mul(x = var_32_cast_fp16, y = var_37_cast_fp16)[name = string("h_1_cast_fp16")];
            tensor<fp16, [1, 256, 64, 388]> input_3_cast_fp16 = mul(x = h_1_cast_fp16, y = m1)[name = string("input_3_cast_fp16")];
            string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(256)];
            tensor<int32, [2]> input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_mid_weight_to_fp16 = const()[name = string("downsample_conv_mid_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5312)))];
            tensor<fp16, [256]> downsample_conv_mid_bias_to_fp16 = const()[name = string("downsample_conv_mid_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9984)))];
            tensor<fp16, [1, 256, 32, 194]> input_5_cast_fp16 = conv(bias = downsample_conv_mid_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = downsample_conv_mid_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string x_3_pad_type_0 = const()[name = string("x_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_3_strides_0 = const()[name = string("x_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_3_pad_0 = const()[name = string("x_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_3_dilations_0 = const()[name = string("x_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_3_groups_0 = const()[name = string("x_3_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> downsample_lin_mid_weight_to_fp16 = const()[name = string("downsample_lin_mid_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10560)))];
            tensor<fp16, [256]> downsample_lin_mid_bias_to_fp16 = const()[name = string("downsample_lin_mid_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141696)))];
            tensor<fp16, [1, 256, 32, 194]> x_3_cast_fp16 = conv(bias = downsample_lin_mid_bias_to_fp16, dilations = x_3_dilations_0, groups = x_3_groups_0, pad = x_3_pad_0, pad_type = x_3_pad_type_0, strides = x_3_strides_0, weight = downsample_lin_mid_weight_to_fp16, x = input_5_cast_fp16)[name = string("x_3_cast_fp16")];
            fp16 var_54_to_fp16 = const()[name = string("op_54_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 32, 194]> var_55_cast_fp16 = mul(x = x_3_cast_fp16, y = var_54_to_fp16)[name = string("op_55_cast_fp16")];
            tensor<fp16, [1, 256, 32, 194]> var_58_cast_fp16 = tanh(x = var_55_cast_fp16)[name = string("op_58_cast_fp16")];
            fp16 var_59_to_fp16 = const()[name = string("op_59_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 32, 194]> var_60_cast_fp16 = add(x = var_58_cast_fp16, y = var_59_to_fp16)[name = string("op_60_cast_fp16")];
            tensor<fp16, [1, 256, 32, 194]> h_3_cast_fp16 = mul(x = var_55_cast_fp16, y = var_60_cast_fp16)[name = string("h_3_cast_fp16")];
            tensor<fp16, [1, 256, 32, 194]> input_7_cast_fp16 = mul(x = h_3_cast_fp16, y = m2)[name = string("input_7_cast_fp16")];
            string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(256)];
            tensor<int32, [2]> input_9_dilations_0 = const()[name = string("input_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> downsample_conv_end_weight_to_fp16 = const()[name = string("downsample_conv_end_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142272)))];
            tensor<fp16, [256]> downsample_conv_end_bias_to_fp16 = const()[name = string("downsample_conv_end_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146944)))];
            tensor<fp16, [1, 256, 16, 97]> input_9_cast_fp16 = conv(bias = downsample_conv_end_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = downsample_conv_end_weight_to_fp16, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
            string x_5_pad_type_0 = const()[name = string("x_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_5_strides_0 = const()[name = string("x_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_5_pad_0 = const()[name = string("x_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_5_dilations_0 = const()[name = string("x_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_5_groups_0 = const()[name = string("x_5_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> downsample_lin_end_weight_to_fp16 = const()[name = string("downsample_lin_end_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147520)))];
            tensor<fp16, [256]> downsample_lin_end_bias_to_fp16 = const()[name = string("downsample_lin_end_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278656)))];
            tensor<fp16, [1, 256, 16, 97]> x_5_cast_fp16 = conv(bias = downsample_lin_end_bias_to_fp16, dilations = x_5_dilations_0, groups = x_5_groups_0, pad = x_5_pad_0, pad_type = x_5_pad_type_0, strides = x_5_strides_0, weight = downsample_lin_end_weight_to_fp16, x = input_9_cast_fp16)[name = string("x_5_cast_fp16")];
            fp16 var_77_to_fp16 = const()[name = string("op_77_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 256, 16, 97]> var_78_cast_fp16 = mul(x = x_5_cast_fp16, y = var_77_to_fp16)[name = string("op_78_cast_fp16")];
            tensor<fp16, [1, 256, 16, 97]> var_81_cast_fp16 = tanh(x = var_78_cast_fp16)[name = string("op_81_cast_fp16")];
            fp16 var_82_to_fp16 = const()[name = string("op_82_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 256, 16, 97]> var_83_cast_fp16 = add(x = var_81_cast_fp16, y = var_82_to_fp16)[name = string("op_83_cast_fp16")];
            tensor<fp16, [1, 256, 16, 97]> h_cast_fp16 = mul(x = var_78_cast_fp16, y = var_83_cast_fp16)[name = string("h_cast_fp16")];
            tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 4096, 97])];
            tensor<fp16, [1, 4096, 97]> input_11_cast_fp16 = reshape(shape = concat_0, x = h_cast_fp16)[name = string("input_11_cast_fp16")];
            string d_pad_type_0 = const()[name = string("d_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> d_strides_0 = const()[name = string("d_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> d_pad_0 = const()[name = string("d_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> d_dilations_0 = const()[name = string("d_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 d_groups_0 = const()[name = string("d_groups_0"), val = int32(1)];
            tensor<fp16, [768, 4096, 1]> downsample_lin_out_weight_to_fp16 = const()[name = string("downsample_lin_out_weight_to_fp16"), val = tensor<fp16, [768, 4096, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279232)))];
            tensor<fp16, [768]> downsample_lin_out_bias_to_fp16 = const()[name = string("downsample_lin_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6570752)))];
            tensor<fp16, [1, 768, 97]> d_cast_fp16 = conv(bias = downsample_lin_out_bias_to_fp16, dilations = d_dilations_0, groups = d_groups_0, pad = d_pad_0, pad_type = d_pad_type_0, strides = d_strides_0, weight = downsample_lin_out_weight_to_fp16, x = input_11_cast_fp16)[name = string("d_cast_fp16")];
            tensor<int32, [3]> var_107_begin_0 = const()[name = string("op_107_begin_0"), val = tensor<int32, [3]>([0, 0, 1])];
            tensor<int32, [3]> var_107_end_0 = const()[name = string("op_107_end_0"), val = tensor<int32, [3]>([1, 768, 97])];
            tensor<bool, [3]> var_107_end_mask_0 = const()[name = string("op_107_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> var_107_cast_fp16 = slice_by_index(begin = var_107_begin_0, end = var_107_end_0, end_mask = var_107_end_mask_0, x = d_cast_fp16)[name = string("op_107_cast_fp16")];
            tensor<int32, [3]> x_7_perm_0 = const()[name = string("x_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            int32 var_126 = const()[name = string("op_126"), val = int32(-1)];
            bool var_127 = const()[name = string("op_127"), val = bool(true)];
            tensor<fp16, [1, 96, 768]> x_7_cast_fp16 = transpose(perm = x_7_perm_0, x = var_107_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [1, 96, 768]> var_719_cast_fp16 = abs(x = x_7_cast_fp16)[name = string("op_719_cast_fp16")];
            tensor<int32, [1]> var_721_axes_0 = const()[name = string("op_721_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_721_cast_fp16 = reduce_max(axes = var_721_axes_0, keep_dims = var_127, x = var_719_cast_fp16)[name = string("op_721_cast_fp16")];
            fp16 var_128_to_fp16 = const()[name = string("op_128_to_fp16"), val = fp16(0x1.a38p-14)];
            tensor<fp16, [1, 96, 1]> s_1_cast_fp16 = maximum(x = var_721_cast_fp16, y = var_128_to_fp16)[name = string("s_1_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_1_cast_fp16 = real_div(x = x_7_cast_fp16, y = s_1_cast_fp16)[name = string("y_1_cast_fp16")];
            fp16 var_129_promoted_to_fp16 = const()[name = string("op_129_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_724_cast_fp16 = pow(x = y_1_cast_fp16, y = var_129_promoted_to_fp16)[name = string("op_724_cast_fp16")];
            tensor<int32, [1]> var_726_axes_0 = const()[name = string("op_726_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_726_keep_dims_0 = const()[name = string("op_726_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_726_cast_fp16 = reduce_mean(axes = var_726_axes_0, keep_dims = var_726_keep_dims_0, x = var_724_cast_fp16)[name = string("op_726_cast_fp16")];
            fp32 var_727_epsilon_0 = const()[name = string("op_727_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_727_cast_fp16 = inverse(epsilon = var_727_epsilon_0, x = s_1_cast_fp16)[name = string("op_727_cast_fp16")];
            fp16 var_728_to_fp16 = const()[name = string("op_728_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_729_cast_fp16 = mul(x = var_727_cast_fp16, y = var_728_to_fp16)[name = string("op_729_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_730_cast_fp16 = real_div(x = var_729_cast_fp16, y = s_1_cast_fp16)[name = string("op_730_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_731_cast_fp16 = add(x = var_726_cast_fp16, y = var_730_cast_fp16)[name = string("op_731_cast_fp16")];
            fp32 var_732_epsilon_0 = const()[name = string("op_732_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_732_cast_fp16 = rsqrt(epsilon = var_732_epsilon_0, x = var_731_cast_fp16)[name = string("op_732_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_733_cast_fp16 = mul(x = y_1_cast_fp16, y = var_732_cast_fp16)[name = string("op_733_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_0_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6572352)))];
            tensor<fp16, [1, 96, 768]> input_13_cast_fp16 = mul(x = core_backbone_layers_0_operator_norm_weight_to_fp16, y = var_733_cast_fp16)[name = string("input_13_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_0_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6573952)))];
            tensor<fp16, [2304]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [2304]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10112960)))];
            tensor<fp16, [1, 96, 2304]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_0_conv_in_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<int32, [3]> bcx_1_perm_0 = const()[name = string("bcx_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_1_begin_0 = const()[name = string("b_1_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_1_end_0 = const()[name = string("b_1_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_1_end_mask_0 = const()[name = string("b_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_1_cast_fp16 = transpose(perm = bcx_1_perm_0, x = linear_0_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [1, 768, 96]> b_1_cast_fp16 = slice_by_index(begin = b_1_begin_0, end = b_1_end_0, end_mask = b_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("b_1_cast_fp16")];
            tensor<int32, [3]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("c_1_cast_fp16")];
            tensor<int32, [3]> xx_1_begin_0 = const()[name = string("xx_1_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_1_end_0 = const()[name = string("xx_1_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_1_end_mask_0 = const()[name = string("xx_1_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_1_cast_fp16 = slice_by_index(begin = xx_1_begin_0, end = xx_1_end_0, end_mask = xx_1_end_mask_0, x = bcx_1_cast_fp16)[name = string("xx_1_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_1_cast_fp16 = mul(x = b_1_cast_fp16, y = xx_1_cast_fp16)[name = string("new_1_cast_fp16")];
            tensor<int32, [4]> input_17_begin_0 = const()[name = string("input_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_17_end_0 = const()[name = string("input_17_end_0"), val = tensor<int32, [4]>([1, 1, 768, 2])];
            tensor<bool, [4]> input_17_end_mask_0 = const()[name = string("input_17_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_17_squeeze_mask_0 = const()[name = string("input_17_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_17_cast_fp16 = slice_by_index(begin = input_17_begin_0, end = input_17_end_0, end_mask = input_17_end_mask_0, squeeze_mask = input_17_squeeze_mask_0, x = conv)[name = string("input_17_cast_fp16")];
            tensor<int32, [6]> var_748_pad_0 = const()[name = string("op_748_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_748_mode_0 = const()[name = string("op_748_mode_0"), val = string("constant")];
            fp16 const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_748_cast_fp16 = pad(constant_val = const_4_to_fp16, mode = var_748_mode_0, pad = var_748_pad_0, x = new_1_cast_fp16)[name = string("op_748_cast_fp16")];
            tensor<int32, [6]> var_750_pad_0 = const()[name = string("op_750_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_750_mode_0 = const()[name = string("op_750_mode_0"), val = string("constant")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_750_cast_fp16 = pad(constant_val = const_5_to_fp16, mode = var_750_mode_0, pad = var_750_pad_0, x = input_17_cast_fp16)[name = string("op_750_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_1_cast_fp16 = add(x = var_748_cast_fp16, y = var_750_cast_fp16)[name = string("win_1_cast_fp16")];
            tensor<int32, [3]> var_752_begin_0 = const()[name = string("op_752_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_752_end_0 = const()[name = string("op_752_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_752_end_mask_0 = const()[name = string("op_752_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_752_cast_fp16 = slice_by_index(begin = var_752_begin_0, end = var_752_end_0, end_mask = var_752_end_mask_0, x = win_1_cast_fp16)[name = string("op_752_cast_fp16")];
            string var_758_pad_type_0 = const()[name = string("op_758_pad_type_0"), val = string("valid")];
            int32 var_758_groups_0 = const()[name = string("op_758_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_758_strides_0 = const()[name = string("op_758_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_758_pad_0 = const()[name = string("op_758_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_758_dilations_0 = const()[name = string("op_758_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_0_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10117632)))];
            tensor<fp16, [1, 768, 96]> var_758_cast_fp16 = conv(dilations = var_758_dilations_0, groups = var_758_groups_0, pad = var_758_pad_0, pad_type = var_758_pad_type_0, strides = var_758_strides_0, weight = core_backbone_layers_0_conv_conv_weight_to_fp16, x = win_1_cast_fp16)[name = string("op_758_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_760_cast_fp16 = mul(x = c_1_cast_fp16, y = var_758_cast_fp16)[name = string("op_760_cast_fp16")];
            tensor<int32, [3]> input_19_perm_0 = const()[name = string("input_19_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_0_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_0_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10122304)))];
            tensor<fp16, [768]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11302016)))];
            tensor<fp16, [1, 96, 768]> input_19_cast_fp16 = transpose(perm = input_19_perm_0, x = var_760_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [1, 96, 768]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_conv_out_proj_weight_to_fp16, x = input_19_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_9_cast_fp16 = add(x = x_7_cast_fp16, y = linear_1_cast_fp16)[name = string("x_9_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_766_cast_fp16 = abs(x = x_9_cast_fp16)[name = string("op_766_cast_fp16")];
            tensor<int32, [1]> var_768_axes_0 = const()[name = string("op_768_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_768_cast_fp16 = reduce_max(axes = var_768_axes_0, keep_dims = var_127, x = var_766_cast_fp16)[name = string("op_768_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_3_cast_fp16 = maximum(x = var_768_cast_fp16, y = var_128_to_fp16)[name = string("s_3_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_5_cast_fp16 = real_div(x = x_9_cast_fp16, y = s_3_cast_fp16)[name = string("y_5_cast_fp16")];
            fp16 var_129_promoted_1_to_fp16 = const()[name = string("op_129_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_771_cast_fp16 = pow(x = y_5_cast_fp16, y = var_129_promoted_1_to_fp16)[name = string("op_771_cast_fp16")];
            tensor<int32, [1]> var_773_axes_0 = const()[name = string("op_773_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_773_keep_dims_0 = const()[name = string("op_773_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_773_cast_fp16 = reduce_mean(axes = var_773_axes_0, keep_dims = var_773_keep_dims_0, x = var_771_cast_fp16)[name = string("op_773_cast_fp16")];
            fp32 var_774_epsilon_0 = const()[name = string("op_774_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_774_cast_fp16 = inverse(epsilon = var_774_epsilon_0, x = s_3_cast_fp16)[name = string("op_774_cast_fp16")];
            fp16 var_775_to_fp16 = const()[name = string("op_775_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_776_cast_fp16 = mul(x = var_774_cast_fp16, y = var_775_to_fp16)[name = string("op_776_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_777_cast_fp16 = real_div(x = var_776_cast_fp16, y = s_3_cast_fp16)[name = string("op_777_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_778_cast_fp16 = add(x = var_773_cast_fp16, y = var_777_cast_fp16)[name = string("op_778_cast_fp16")];
            fp32 var_779_epsilon_0 = const()[name = string("op_779_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_779_cast_fp16 = rsqrt(epsilon = var_779_epsilon_0, x = var_778_cast_fp16)[name = string("op_779_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_780_cast_fp16 = mul(x = y_5_cast_fp16, y = var_779_cast_fp16)[name = string("op_780_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_0_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_0_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11303616)))];
            tensor<fp16, [1, 96, 768]> input_21_cast_fp16 = mul(x = core_backbone_layers_0_ffn_norm_weight_to_fp16, y = var_780_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_0_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11305216)))];
            tensor<fp16, [1792]> linear_2_bias_0_to_fp16 = const()[name = string("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14057792)))];
            tensor<fp16, [1, 96, 1792]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w1_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_0_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14061440)))];
            tensor<fp16, [1, 96, 1792]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w3_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_3_cast_fp16")];
            fp16 var_789_to_fp16 = const()[name = string("op_789_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_790_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_789_to_fp16)[name = string("op_790_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_793_cast_fp16 = tanh(x = var_790_cast_fp16)[name = string("op_793_cast_fp16")];
            fp16 var_794_to_fp16 = const()[name = string("op_794_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_795_cast_fp16 = add(x = var_793_cast_fp16, y = var_794_to_fp16)[name = string("op_795_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_796_cast_fp16 = mul(x = var_790_cast_fp16, y = var_795_cast_fp16)[name = string("op_796_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_23_cast_fp16 = mul(x = var_796_cast_fp16, y = linear_3_cast_fp16)[name = string("input_23_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_0_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_0_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16814016)))];
            tensor<fp16, [1, 96, 768]> linear_4_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_0_feed_forward_w2_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_13_cast_fp16 = add(x = x_9_cast_fp16, y = linear_4_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_802_cast_fp16 = abs(x = x_13_cast_fp16)[name = string("op_802_cast_fp16")];
            tensor<int32, [1]> var_804_axes_0 = const()[name = string("op_804_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_804_cast_fp16 = reduce_max(axes = var_804_axes_0, keep_dims = var_127, x = var_802_cast_fp16)[name = string("op_804_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_5_cast_fp16 = maximum(x = var_804_cast_fp16, y = var_128_to_fp16)[name = string("s_5_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_7_cast_fp16 = real_div(x = x_13_cast_fp16, y = s_5_cast_fp16)[name = string("y_7_cast_fp16")];
            fp16 var_129_promoted_2_to_fp16 = const()[name = string("op_129_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_807_cast_fp16 = pow(x = y_7_cast_fp16, y = var_129_promoted_2_to_fp16)[name = string("op_807_cast_fp16")];
            tensor<int32, [1]> var_809_axes_0 = const()[name = string("op_809_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_809_keep_dims_0 = const()[name = string("op_809_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_809_cast_fp16 = reduce_mean(axes = var_809_axes_0, keep_dims = var_809_keep_dims_0, x = var_807_cast_fp16)[name = string("op_809_cast_fp16")];
            fp32 var_810_epsilon_0 = const()[name = string("op_810_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_810_cast_fp16 = inverse(epsilon = var_810_epsilon_0, x = s_5_cast_fp16)[name = string("op_810_cast_fp16")];
            fp16 var_811_to_fp16 = const()[name = string("op_811_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_812_cast_fp16 = mul(x = var_810_cast_fp16, y = var_811_to_fp16)[name = string("op_812_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_813_cast_fp16 = real_div(x = var_812_cast_fp16, y = s_5_cast_fp16)[name = string("op_813_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_814_cast_fp16 = add(x = var_809_cast_fp16, y = var_813_cast_fp16)[name = string("op_814_cast_fp16")];
            fp32 var_815_epsilon_0 = const()[name = string("op_815_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_815_cast_fp16 = rsqrt(epsilon = var_815_epsilon_0, x = var_814_cast_fp16)[name = string("op_815_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_816_cast_fp16 = mul(x = y_7_cast_fp16, y = var_815_cast_fp16)[name = string("op_816_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_1_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19566592)))];
            tensor<fp16, [1, 96, 768]> input_25_cast_fp16 = mul(x = core_backbone_layers_1_operator_norm_weight_to_fp16, y = var_816_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_1_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19568192)))];
            tensor<fp16, [1, 96, 2304]> linear_5_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_1_conv_in_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [3]> bcx_3_perm_0 = const()[name = string("bcx_3_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_5_begin_0 = const()[name = string("b_5_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_5_end_0 = const()[name = string("b_5_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_5_end_mask_0 = const()[name = string("b_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_3_cast_fp16 = transpose(perm = bcx_3_perm_0, x = linear_5_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [1, 768, 96]> b_5_cast_fp16 = slice_by_index(begin = b_5_begin_0, end = b_5_end_0, end_mask = b_5_end_mask_0, x = bcx_3_cast_fp16)[name = string("b_5_cast_fp16")];
            tensor<int32, [3]> c_3_begin_0 = const()[name = string("c_3_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_3_end_0 = const()[name = string("c_3_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_3_end_mask_0 = const()[name = string("c_3_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_3_cast_fp16 = slice_by_index(begin = c_3_begin_0, end = c_3_end_0, end_mask = c_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("c_3_cast_fp16")];
            tensor<int32, [3]> xx_3_begin_0 = const()[name = string("xx_3_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_3_end_0 = const()[name = string("xx_3_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_3_end_mask_0 = const()[name = string("xx_3_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_3_cast_fp16 = slice_by_index(begin = xx_3_begin_0, end = xx_3_end_0, end_mask = xx_3_end_mask_0, x = bcx_3_cast_fp16)[name = string("xx_3_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_3_cast_fp16 = mul(x = b_5_cast_fp16, y = xx_3_cast_fp16)[name = string("new_3_cast_fp16")];
            tensor<int32, [4]> input_29_begin_0 = const()[name = string("input_29_begin_0"), val = tensor<int32, [4]>([1, 0, 0, 0])];
            tensor<int32, [4]> input_29_end_0 = const()[name = string("input_29_end_0"), val = tensor<int32, [4]>([2, 1, 768, 2])];
            tensor<bool, [4]> input_29_end_mask_0 = const()[name = string("input_29_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_29_squeeze_mask_0 = const()[name = string("input_29_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_29_cast_fp16 = slice_by_index(begin = input_29_begin_0, end = input_29_end_0, end_mask = input_29_end_mask_0, squeeze_mask = input_29_squeeze_mask_0, x = conv)[name = string("input_29_cast_fp16")];
            tensor<int32, [6]> var_831_pad_0 = const()[name = string("op_831_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_831_mode_0 = const()[name = string("op_831_mode_0"), val = string("constant")];
            fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_831_cast_fp16 = pad(constant_val = const_6_to_fp16, mode = var_831_mode_0, pad = var_831_pad_0, x = new_3_cast_fp16)[name = string("op_831_cast_fp16")];
            tensor<int32, [6]> var_833_pad_0 = const()[name = string("op_833_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_833_mode_0 = const()[name = string("op_833_mode_0"), val = string("constant")];
            fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_833_cast_fp16 = pad(constant_val = const_7_to_fp16, mode = var_833_mode_0, pad = var_833_pad_0, x = input_29_cast_fp16)[name = string("op_833_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_5_cast_fp16 = add(x = var_831_cast_fp16, y = var_833_cast_fp16)[name = string("win_5_cast_fp16")];
            tensor<int32, [3]> var_835_begin_0 = const()[name = string("op_835_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_835_end_0 = const()[name = string("op_835_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_835_end_mask_0 = const()[name = string("op_835_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_835_cast_fp16 = slice_by_index(begin = var_835_begin_0, end = var_835_end_0, end_mask = var_835_end_mask_0, x = win_5_cast_fp16)[name = string("op_835_cast_fp16")];
            string var_841_pad_type_0 = const()[name = string("op_841_pad_type_0"), val = string("valid")];
            int32 var_841_groups_0 = const()[name = string("op_841_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_841_strides_0 = const()[name = string("op_841_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_841_pad_0 = const()[name = string("op_841_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_841_dilations_0 = const()[name = string("op_841_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_1_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23107200)))];
            tensor<fp16, [1, 768, 96]> var_841_cast_fp16 = conv(dilations = var_841_dilations_0, groups = var_841_groups_0, pad = var_841_pad_0, pad_type = var_841_pad_type_0, strides = var_841_strides_0, weight = core_backbone_layers_1_conv_conv_weight_to_fp16, x = win_5_cast_fp16)[name = string("op_841_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_843_cast_fp16 = mul(x = c_3_cast_fp16, y = var_841_cast_fp16)[name = string("op_843_cast_fp16")];
            tensor<int32, [3]> input_31_perm_0 = const()[name = string("input_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_1_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_1_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23111872)))];
            tensor<fp16, [1, 96, 768]> input_31_cast_fp16 = transpose(perm = input_31_perm_0, x = var_843_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [1, 96, 768]> linear_6_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_conv_out_proj_weight_to_fp16, x = input_31_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_15_cast_fp16 = add(x = x_13_cast_fp16, y = linear_6_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_849_cast_fp16 = abs(x = x_15_cast_fp16)[name = string("op_849_cast_fp16")];
            tensor<int32, [1]> var_851_axes_0 = const()[name = string("op_851_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_851_cast_fp16 = reduce_max(axes = var_851_axes_0, keep_dims = var_127, x = var_849_cast_fp16)[name = string("op_851_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_7_cast_fp16 = maximum(x = var_851_cast_fp16, y = var_128_to_fp16)[name = string("s_7_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_11_cast_fp16 = real_div(x = x_15_cast_fp16, y = s_7_cast_fp16)[name = string("y_11_cast_fp16")];
            fp16 var_129_promoted_3_to_fp16 = const()[name = string("op_129_promoted_3_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_854_cast_fp16 = pow(x = y_11_cast_fp16, y = var_129_promoted_3_to_fp16)[name = string("op_854_cast_fp16")];
            tensor<int32, [1]> var_856_axes_0 = const()[name = string("op_856_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_856_keep_dims_0 = const()[name = string("op_856_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_856_cast_fp16 = reduce_mean(axes = var_856_axes_0, keep_dims = var_856_keep_dims_0, x = var_854_cast_fp16)[name = string("op_856_cast_fp16")];
            fp32 var_857_epsilon_0 = const()[name = string("op_857_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_857_cast_fp16 = inverse(epsilon = var_857_epsilon_0, x = s_7_cast_fp16)[name = string("op_857_cast_fp16")];
            fp16 var_858_to_fp16 = const()[name = string("op_858_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_859_cast_fp16 = mul(x = var_857_cast_fp16, y = var_858_to_fp16)[name = string("op_859_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_860_cast_fp16 = real_div(x = var_859_cast_fp16, y = s_7_cast_fp16)[name = string("op_860_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_861_cast_fp16 = add(x = var_856_cast_fp16, y = var_860_cast_fp16)[name = string("op_861_cast_fp16")];
            fp32 var_862_epsilon_0 = const()[name = string("op_862_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_862_cast_fp16 = rsqrt(epsilon = var_862_epsilon_0, x = var_861_cast_fp16)[name = string("op_862_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_863_cast_fp16 = mul(x = y_11_cast_fp16, y = var_862_cast_fp16)[name = string("op_863_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_1_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_1_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24291584)))];
            tensor<fp16, [1, 96, 768]> input_33_cast_fp16 = mul(x = core_backbone_layers_1_ffn_norm_weight_to_fp16, y = var_863_cast_fp16)[name = string("input_33_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_1_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24293184)))];
            tensor<fp16, [1, 96, 1792]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w1_weight_to_fp16, x = input_33_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_1_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27045760)))];
            tensor<fp16, [1, 96, 1792]> linear_8_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w3_weight_to_fp16, x = input_33_cast_fp16)[name = string("linear_8_cast_fp16")];
            fp16 var_872_to_fp16 = const()[name = string("op_872_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_873_cast_fp16 = mul(x = linear_7_cast_fp16, y = var_872_to_fp16)[name = string("op_873_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_876_cast_fp16 = tanh(x = var_873_cast_fp16)[name = string("op_876_cast_fp16")];
            fp16 var_877_to_fp16 = const()[name = string("op_877_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_878_cast_fp16 = add(x = var_876_cast_fp16, y = var_877_to_fp16)[name = string("op_878_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_879_cast_fp16 = mul(x = var_873_cast_fp16, y = var_878_cast_fp16)[name = string("op_879_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_35_cast_fp16 = mul(x = var_879_cast_fp16, y = linear_8_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_1_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_1_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29798336)))];
            tensor<fp16, [1, 96, 768]> linear_9_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_1_feed_forward_w2_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_9_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_19_cast_fp16 = add(x = x_15_cast_fp16, y = linear_9_cast_fp16)[name = string("x_19_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_885_cast_fp16 = abs(x = x_19_cast_fp16)[name = string("op_885_cast_fp16")];
            tensor<int32, [1]> var_887_axes_0 = const()[name = string("op_887_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_887_cast_fp16 = reduce_max(axes = var_887_axes_0, keep_dims = var_127, x = var_885_cast_fp16)[name = string("op_887_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_9_cast_fp16 = maximum(x = var_887_cast_fp16, y = var_128_to_fp16)[name = string("s_9_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_13_cast_fp16 = real_div(x = x_19_cast_fp16, y = s_9_cast_fp16)[name = string("y_13_cast_fp16")];
            fp16 var_129_promoted_4_to_fp16 = const()[name = string("op_129_promoted_4_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_890_cast_fp16 = pow(x = y_13_cast_fp16, y = var_129_promoted_4_to_fp16)[name = string("op_890_cast_fp16")];
            tensor<int32, [1]> var_892_axes_0 = const()[name = string("op_892_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_892_keep_dims_0 = const()[name = string("op_892_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_892_cast_fp16 = reduce_mean(axes = var_892_axes_0, keep_dims = var_892_keep_dims_0, x = var_890_cast_fp16)[name = string("op_892_cast_fp16")];
            fp32 var_893_epsilon_0 = const()[name = string("op_893_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_893_cast_fp16 = inverse(epsilon = var_893_epsilon_0, x = s_9_cast_fp16)[name = string("op_893_cast_fp16")];
            fp16 var_894_to_fp16 = const()[name = string("op_894_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_895_cast_fp16 = mul(x = var_893_cast_fp16, y = var_894_to_fp16)[name = string("op_895_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_896_cast_fp16 = real_div(x = var_895_cast_fp16, y = s_9_cast_fp16)[name = string("op_896_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_897_cast_fp16 = add(x = var_892_cast_fp16, y = var_896_cast_fp16)[name = string("op_897_cast_fp16")];
            fp32 var_898_epsilon_0 = const()[name = string("op_898_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_898_cast_fp16 = rsqrt(epsilon = var_898_epsilon_0, x = var_897_cast_fp16)[name = string("op_898_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_899_cast_fp16 = mul(x = y_13_cast_fp16, y = var_898_cast_fp16)[name = string("op_899_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_2_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32550912)))];
            tensor<fp16, [1, 96, 768]> input_37_cast_fp16 = mul(x = core_backbone_layers_2_operator_norm_weight_to_fp16, y = var_899_cast_fp16)[name = string("input_37_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32552512)))];
            tensor<fp16, [1, 96, 768]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_q_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<int32, [4]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [4]>([1, 96, 8, 96])];
            tensor<fp16, [1, 96, 8, 96]> x_21_cast_fp16 = reshape(shape = concat_1, x = linear_10_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_906_cast_fp16 = abs(x = x_21_cast_fp16)[name = string("op_906_cast_fp16")];
            tensor<int32, [1]> var_908_axes_0 = const()[name = string("op_908_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 8, 1]> var_908_cast_fp16 = reduce_max(axes = var_908_axes_0, keep_dims = var_127, x = var_906_cast_fp16)[name = string("op_908_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> s_11_cast_fp16 = maximum(x = var_908_cast_fp16, y = var_128_to_fp16)[name = string("s_11_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> y_15_cast_fp16 = real_div(x = x_21_cast_fp16, y = s_11_cast_fp16)[name = string("y_15_cast_fp16")];
            fp16 var_129_promoted_5_to_fp16 = const()[name = string("op_129_promoted_5_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 8, 96]> var_911_cast_fp16 = pow(x = y_15_cast_fp16, y = var_129_promoted_5_to_fp16)[name = string("op_911_cast_fp16")];
            tensor<int32, [1]> var_913_axes_0 = const()[name = string("op_913_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_913_keep_dims_0 = const()[name = string("op_913_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 8, 1]> var_913_cast_fp16 = reduce_mean(axes = var_913_axes_0, keep_dims = var_913_keep_dims_0, x = var_911_cast_fp16)[name = string("op_913_cast_fp16")];
            fp32 var_914_epsilon_0 = const()[name = string("op_914_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 8, 1]> var_914_cast_fp16 = inverse(epsilon = var_914_epsilon_0, x = s_11_cast_fp16)[name = string("op_914_cast_fp16")];
            fp16 var_915_to_fp16 = const()[name = string("op_915_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 8, 1]> var_916_cast_fp16 = mul(x = var_914_cast_fp16, y = var_915_to_fp16)[name = string("op_916_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_917_cast_fp16 = real_div(x = var_916_cast_fp16, y = s_11_cast_fp16)[name = string("op_917_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_918_cast_fp16 = add(x = var_913_cast_fp16, y = var_917_cast_fp16)[name = string("op_918_cast_fp16")];
            fp32 var_919_epsilon_0 = const()[name = string("op_919_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 8, 1]> var_919_cast_fp16 = rsqrt(epsilon = var_919_epsilon_0, x = var_918_cast_fp16)[name = string("op_919_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_920_cast_fp16 = mul(x = y_15_cast_fp16, y = var_919_cast_fp16)[name = string("op_920_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_2_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33732224)))];
            tensor<fp16, [1, 96, 8, 96]> var_921_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_q_norm_weight_to_fp16, y = var_920_cast_fp16)[name = string("op_921_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(33732480)))];
            tensor<fp16, [384]> linear_11_bias_0_to_fp16 = const()[name = string("linear_11_bias_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34322368)))];
            tensor<fp16, [1, 96, 384]> linear_11_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_k_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<int32, [4]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> x_23_cast_fp16 = reshape(shape = concat_2, x = linear_11_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_928_cast_fp16 = abs(x = x_23_cast_fp16)[name = string("op_928_cast_fp16")];
            tensor<int32, [1]> var_930_axes_0 = const()[name = string("op_930_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 4, 1]> var_930_cast_fp16 = reduce_max(axes = var_930_axes_0, keep_dims = var_127, x = var_928_cast_fp16)[name = string("op_930_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> s_13_cast_fp16 = maximum(x = var_930_cast_fp16, y = var_128_to_fp16)[name = string("s_13_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> y_17_cast_fp16 = real_div(x = x_23_cast_fp16, y = s_13_cast_fp16)[name = string("y_17_cast_fp16")];
            fp16 var_129_promoted_6_to_fp16 = const()[name = string("op_129_promoted_6_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 4, 96]> var_933_cast_fp16 = pow(x = y_17_cast_fp16, y = var_129_promoted_6_to_fp16)[name = string("op_933_cast_fp16")];
            tensor<int32, [1]> var_935_axes_0 = const()[name = string("op_935_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_935_keep_dims_0 = const()[name = string("op_935_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 4, 1]> var_935_cast_fp16 = reduce_mean(axes = var_935_axes_0, keep_dims = var_935_keep_dims_0, x = var_933_cast_fp16)[name = string("op_935_cast_fp16")];
            fp32 var_936_epsilon_0 = const()[name = string("op_936_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 4, 1]> var_936_cast_fp16 = inverse(epsilon = var_936_epsilon_0, x = s_13_cast_fp16)[name = string("op_936_cast_fp16")];
            fp16 var_937_to_fp16 = const()[name = string("op_937_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 4, 1]> var_938_cast_fp16 = mul(x = var_936_cast_fp16, y = var_937_to_fp16)[name = string("op_938_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_939_cast_fp16 = real_div(x = var_938_cast_fp16, y = s_13_cast_fp16)[name = string("op_939_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_940_cast_fp16 = add(x = var_935_cast_fp16, y = var_939_cast_fp16)[name = string("op_940_cast_fp16")];
            fp32 var_941_epsilon_0 = const()[name = string("op_941_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 4, 1]> var_941_cast_fp16 = rsqrt(epsilon = var_941_epsilon_0, x = var_940_cast_fp16)[name = string("op_941_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_942_cast_fp16 = mul(x = y_17_cast_fp16, y = var_941_cast_fp16)[name = string("op_942_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_2_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34323200)))];
            tensor<fp16, [1, 96, 4, 96]> var_943_cast_fp16 = mul(x = core_backbone_layers_2_self_attn_k_norm_weight_to_fp16, y = var_942_cast_fp16)[name = string("op_943_cast_fp16")];
            tensor<int32, [4]> new_5_perm_0 = const()[name = string("new_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34323456)))];
            tensor<fp16, [1, 96, 384]> linear_12_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_v_proj_weight_to_fp16, x = input_37_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> var_948_cast_fp16 = reshape(shape = concat_3, x = linear_12_cast_fp16)[name = string("op_948_cast_fp16")];
            tensor<int32, [4]> new_7_perm_0 = const()[name = string("new_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_41_begin_0 = const()[name = string("input_41_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_41_end_0 = const()[name = string("input_41_end_0"), val = tensor<int32, [5]>([1, 1, 4, 31, 96])];
            tensor<bool, [5]> input_41_end_mask_0 = const()[name = string("input_41_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_41_squeeze_mask_0 = const()[name = string("input_41_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_41_cast_fp16 = slice_by_index(begin = input_41_begin_0, end = input_41_end_0, end_mask = input_41_end_mask_0, squeeze_mask = input_41_squeeze_mask_0, x = k)[name = string("input_41_cast_fp16")];
            tensor<int32, [8]> var_953_pad_0 = const()[name = string("op_953_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_953_mode_0 = const()[name = string("op_953_mode_0"), val = string("constant")];
            fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_5_cast_fp16 = transpose(perm = new_5_perm_0, x = var_943_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [1, 4, 127, 96]> var_953_cast_fp16 = pad(constant_val = const_8_to_fp16, mode = var_953_mode_0, pad = var_953_pad_0, x = new_5_cast_fp16)[name = string("op_953_cast_fp16")];
            tensor<int32, [8]> var_955_pad_0 = const()[name = string("op_955_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_955_mode_0 = const()[name = string("op_955_mode_0"), val = string("constant")];
            fp16 const_9_to_fp16 = const()[name = string("const_9_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_955_cast_fp16 = pad(constant_val = const_9_to_fp16, mode = var_955_mode_0, pad = var_955_pad_0, x = input_41_cast_fp16)[name = string("op_955_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> x_27_cast_fp16 = add(x = var_953_cast_fp16, y = var_955_cast_fp16)[name = string("x_27_cast_fp16")];
            tensor<int32, [5]> input_45_begin_0 = const()[name = string("input_45_begin_0"), val = tensor<int32, [5]>([0, 0, 0, 0, 0])];
            tensor<int32, [5]> input_45_end_0 = const()[name = string("input_45_end_0"), val = tensor<int32, [5]>([1, 1, 4, 31, 96])];
            tensor<bool, [5]> input_45_end_mask_0 = const()[name = string("input_45_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_45_squeeze_mask_0 = const()[name = string("input_45_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_45_cast_fp16 = slice_by_index(begin = input_45_begin_0, end = input_45_end_0, end_mask = input_45_end_mask_0, squeeze_mask = input_45_squeeze_mask_0, x = v)[name = string("input_45_cast_fp16")];
            tensor<int32, [8]> var_960_pad_0 = const()[name = string("op_960_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_960_mode_0 = const()[name = string("op_960_mode_0"), val = string("constant")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_7_cast_fp16 = transpose(perm = new_7_perm_0, x = var_948_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [1, 4, 127, 96]> var_960_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = var_960_mode_0, pad = var_960_pad_0, x = new_7_cast_fp16)[name = string("op_960_cast_fp16")];
            tensor<int32, [8]> var_962_pad_0 = const()[name = string("op_962_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_962_mode_0 = const()[name = string("op_962_mode_0"), val = string("constant")];
            fp16 const_11_to_fp16 = const()[name = string("const_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_962_cast_fp16 = pad(constant_val = const_11_to_fp16, mode = var_962_mode_0, pad = var_962_pad_0, x = input_45_cast_fp16)[name = string("op_962_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> v_3_cast_fp16 = add(x = var_960_cast_fp16, y = var_962_cast_fp16)[name = string("v_3_cast_fp16")];
            tensor<fp16, [1, 96, 1, 96]> const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = tensor<fp16, [1, 96, 1, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230599744)))];
            tensor<fp16, [1, 96, 8, 96]> var_964_cast_fp16 = mul(x = var_921_cast_fp16, y = const_66_to_fp16)[name = string("op_964_cast_fp16")];
            tensor<int32, [2]> var_965_split_sizes_0 = const()[name = string("op_965_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_67 = const()[name = string("const_67"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 48]> var_965_cast_fp16_0, tensor<fp16, [1, 96, 8, 48]> var_965_cast_fp16_1 = split(axis = const_67, split_sizes = var_965_split_sizes_0, x = var_921_cast_fp16)[name = string("op_965_cast_fp16")];
            fp16 const_12_promoted_to_fp16 = const()[name = string("const_12_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 96, 8, 48]> var_967_cast_fp16 = mul(x = var_965_cast_fp16_1, y = const_12_promoted_to_fp16)[name = string("op_967_cast_fp16")];
            bool var_969_interleave_0 = const()[name = string("op_969_interleave_0"), val = bool(false)];
            int32 const_68 = const()[name = string("const_68"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 96]> var_969_cast_fp16 = concat(axis = const_68, interleave = var_969_interleave_0, values = (var_967_cast_fp16, var_965_cast_fp16_0))[name = string("op_969_cast_fp16")];
            tensor<fp16, [1, 96, 1, 96]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [1, 96, 1, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230618240)))];
            tensor<fp16, [1, 96, 8, 96]> var_970_cast_fp16 = mul(x = var_969_cast_fp16, y = const_69_to_fp16)[name = string("op_970_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> q_1_cast_fp16 = add(x = var_964_cast_fp16, y = var_970_cast_fp16)[name = string("q_1_cast_fp16")];
            tensor<fp16, [1, 1, 127, 96]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [1, 1, 127, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230636736)))];
            tensor<fp16, [1, 4, 127, 96]> var_972_cast_fp16 = mul(x = x_27_cast_fp16, y = const_0_to_fp16)[name = string("op_972_cast_fp16")];
            tensor<int32, [2]> var_973_split_sizes_0 = const()[name = string("op_973_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_973_axis_0 = const()[name = string("op_973_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 127, 48]> var_973_cast_fp16_0, tensor<fp16, [1, 4, 127, 48]> var_973_cast_fp16_1 = split(axis = var_973_axis_0, split_sizes = var_973_split_sizes_0, x = x_27_cast_fp16)[name = string("op_973_cast_fp16")];
            fp16 const_13_promoted_to_fp16 = const()[name = string("const_13_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 127, 48]> var_975_cast_fp16 = mul(x = var_973_cast_fp16_1, y = const_13_promoted_to_fp16)[name = string("op_975_cast_fp16")];
            bool var_977_interleave_0 = const()[name = string("op_977_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 127, 96]> var_977_cast_fp16 = concat(axis = var_126, interleave = var_977_interleave_0, values = (var_975_cast_fp16, var_973_cast_fp16_0))[name = string("op_977_cast_fp16")];
            tensor<fp16, [1, 1, 127, 96]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [1, 1, 127, 96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230661184)))];
            tensor<fp16, [1, 4, 127, 96]> var_978_cast_fp16 = mul(x = var_977_cast_fp16, y = const_1_to_fp16)[name = string("op_978_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> k_3_cast_fp16 = add(x = var_972_cast_fp16, y = var_978_cast_fp16)[name = string("k_3_cast_fp16")];
            tensor<int32, [1]> var_980_axes_0 = const()[name = string("op_980_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_980_cast_fp16 = expand_dims(axes = var_980_axes_0, x = k_3_cast_fp16)[name = string("op_980_cast_fp16")];
            tensor<int32, [5]> var_982_reps_0 = const()[name = string("op_982_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_982_cast_fp16 = tile(reps = var_982_reps_0, x = var_980_cast_fp16)[name = string("op_982_cast_fp16")];
            tensor<int32, [4]> concat_5 = const()[name = string("concat_5"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> k_5_cast_fp16 = reshape(shape = concat_5, x = var_982_cast_fp16)[name = string("k_5_cast_fp16")];
            tensor<int32, [1]> var_984_axes_0 = const()[name = string("op_984_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_984_cast_fp16 = expand_dims(axes = var_984_axes_0, x = v_3_cast_fp16)[name = string("op_984_cast_fp16")];
            tensor<int32, [5]> var_986_reps_0 = const()[name = string("op_986_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_986_cast_fp16 = tile(reps = var_986_reps_0, x = var_984_cast_fp16)[name = string("op_986_cast_fp16")];
            tensor<int32, [4]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> v_5_cast_fp16 = reshape(shape = concat_7, x = var_986_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> transpose_0_perm_0 = const()[name = string("transpose_0_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_989_transpose_x_1 = const()[name = string("op_989_transpose_x_1"), val = bool(false)];
            bool var_989_transpose_y_1 = const()[name = string("op_989_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 96]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = q_1_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [1, 8, 96, 127]> var_989_cast_fp16 = matmul(transpose_x = var_989_transpose_x_1, transpose_y = var_989_transpose_y_1, x = transpose_0_cast_fp16, y = k_5_cast_fp16)[name = string("op_989_cast_fp16")];
            fp16 var_990_to_fp16 = const()[name = string("op_990_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 96, 127]> s_15_cast_fp16 = mul(x = var_989_cast_fp16, y = var_990_to_fp16)[name = string("s_15_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_29_cast_fp16 = add(x = s_15_cast_fp16, y = mask)[name = string("x_29_cast_fp16")];
            tensor<int32, [1]> var_994_axes_0 = const()[name = string("op_994_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 96, 1]> var_994_cast_fp16 = reduce_max(axes = var_994_axes_0, keep_dims = var_127, x = x_29_cast_fp16)[name = string("op_994_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_31_cast_fp16 = sub(x = x_29_cast_fp16, y = var_994_cast_fp16)[name = string("x_31_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_33_cast_fp16 = exp(x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<int32, [1]> var_998_axes_0 = const()[name = string("op_998_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_998_keep_dims_0 = const()[name = string("op_998_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 1]> var_998_cast_fp16 = reduce_sum(axes = var_998_axes_0, keep_dims = var_998_keep_dims_0, x = x_33_cast_fp16)[name = string("op_998_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> var_999_cast_fp16 = real_div(x = x_33_cast_fp16, y = var_998_cast_fp16)[name = string("op_999_cast_fp16")];
            bool o_1_transpose_x_0 = const()[name = string("o_1_transpose_x_0"), val = bool(false)];
            bool o_1_transpose_y_0 = const()[name = string("o_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 96, 96]> o_1_cast_fp16 = matmul(transpose_x = o_1_transpose_x_0, transpose_y = o_1_transpose_y_0, x = var_999_cast_fp16, y = v_5_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<int32, [4]> o_3_perm_0 = const()[name = string("o_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [3]>([1, 96, 768])];
            tensor<fp16, [1, 96, 8, 96]> o_3_cast_fp16 = transpose(perm = o_3_perm_0, x = o_1_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [1, 96, 768]> input_47_cast_fp16 = reshape(shape = concat_8, x = o_3_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34934720)))];
            tensor<fp16, [1, 96, 768]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_self_attn_o_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> var_1008_begin_0 = const()[name = string("op_1008_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_1008_end_0 = const()[name = string("op_1008_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_1008_end_mask_0 = const()[name = string("op_1008_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1008_cast_fp16 = slice_by_index(begin = var_1008_begin_0, end = var_1008_end_0, end_mask = var_1008_end_mask_0, x = x_27_cast_fp16)[name = string("op_1008_cast_fp16")];
            tensor<int32, [4]> var_1011_begin_0 = const()[name = string("op_1011_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_1011_end_0 = const()[name = string("op_1011_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_1011_end_mask_0 = const()[name = string("op_1011_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1011_cast_fp16 = slice_by_index(begin = var_1011_begin_0, end = var_1011_end_0, end_mask = var_1011_end_mask_0, x = v_3_cast_fp16)[name = string("op_1011_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_35_cast_fp16 = add(x = x_19_cast_fp16, y = linear_13_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1014_cast_fp16 = abs(x = x_35_cast_fp16)[name = string("op_1014_cast_fp16")];
            tensor<int32, [1]> var_1016_axes_0 = const()[name = string("op_1016_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1016_cast_fp16 = reduce_max(axes = var_1016_axes_0, keep_dims = var_127, x = var_1014_cast_fp16)[name = string("op_1016_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_17_cast_fp16 = maximum(x = var_1016_cast_fp16, y = var_128_to_fp16)[name = string("s_17_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_19_cast_fp16 = real_div(x = x_35_cast_fp16, y = s_17_cast_fp16)[name = string("y_19_cast_fp16")];
            fp16 var_129_promoted_7_to_fp16 = const()[name = string("op_129_promoted_7_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1019_cast_fp16 = pow(x = y_19_cast_fp16, y = var_129_promoted_7_to_fp16)[name = string("op_1019_cast_fp16")];
            tensor<int32, [1]> var_1021_axes_0 = const()[name = string("op_1021_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1021_keep_dims_0 = const()[name = string("op_1021_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1021_cast_fp16 = reduce_mean(axes = var_1021_axes_0, keep_dims = var_1021_keep_dims_0, x = var_1019_cast_fp16)[name = string("op_1021_cast_fp16")];
            fp32 var_1022_epsilon_0 = const()[name = string("op_1022_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1022_cast_fp16 = inverse(epsilon = var_1022_epsilon_0, x = s_17_cast_fp16)[name = string("op_1022_cast_fp16")];
            fp16 var_1023_to_fp16 = const()[name = string("op_1023_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1024_cast_fp16 = mul(x = var_1022_cast_fp16, y = var_1023_to_fp16)[name = string("op_1024_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1025_cast_fp16 = real_div(x = var_1024_cast_fp16, y = s_17_cast_fp16)[name = string("op_1025_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1026_cast_fp16 = add(x = var_1021_cast_fp16, y = var_1025_cast_fp16)[name = string("op_1026_cast_fp16")];
            fp32 var_1027_epsilon_0 = const()[name = string("op_1027_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1027_cast_fp16 = rsqrt(epsilon = var_1027_epsilon_0, x = var_1026_cast_fp16)[name = string("op_1027_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1028_cast_fp16 = mul(x = y_19_cast_fp16, y = var_1027_cast_fp16)[name = string("op_1028_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_2_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_2_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36114432)))];
            tensor<fp16, [1, 96, 768]> input_49_cast_fp16 = mul(x = core_backbone_layers_2_ffn_norm_weight_to_fp16, y = var_1028_cast_fp16)[name = string("input_49_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_2_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36116032)))];
            tensor<fp16, [1, 96, 1792]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w1_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_2_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38868608)))];
            tensor<fp16, [1, 96, 1792]> linear_15_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w3_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_15_cast_fp16")];
            fp16 var_1037_to_fp16 = const()[name = string("op_1037_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1038_cast_fp16 = mul(x = linear_14_cast_fp16, y = var_1037_to_fp16)[name = string("op_1038_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1041_cast_fp16 = tanh(x = var_1038_cast_fp16)[name = string("op_1041_cast_fp16")];
            fp16 var_1042_to_fp16 = const()[name = string("op_1042_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1043_cast_fp16 = add(x = var_1041_cast_fp16, y = var_1042_to_fp16)[name = string("op_1043_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1044_cast_fp16 = mul(x = var_1038_cast_fp16, y = var_1043_cast_fp16)[name = string("op_1044_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_51_cast_fp16 = mul(x = var_1044_cast_fp16, y = linear_15_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_2_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_2_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41621184)))];
            tensor<fp16, [1, 96, 768]> linear_16_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_2_feed_forward_w2_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_39_cast_fp16 = add(x = x_35_cast_fp16, y = linear_16_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1050_cast_fp16 = abs(x = x_39_cast_fp16)[name = string("op_1050_cast_fp16")];
            tensor<int32, [1]> var_1052_axes_0 = const()[name = string("op_1052_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1052_cast_fp16 = reduce_max(axes = var_1052_axes_0, keep_dims = var_127, x = var_1050_cast_fp16)[name = string("op_1052_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_19_cast_fp16 = maximum(x = var_1052_cast_fp16, y = var_128_to_fp16)[name = string("s_19_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_21_cast_fp16 = real_div(x = x_39_cast_fp16, y = s_19_cast_fp16)[name = string("y_21_cast_fp16")];
            fp16 var_129_promoted_8_to_fp16 = const()[name = string("op_129_promoted_8_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1055_cast_fp16 = pow(x = y_21_cast_fp16, y = var_129_promoted_8_to_fp16)[name = string("op_1055_cast_fp16")];
            tensor<int32, [1]> var_1057_axes_0 = const()[name = string("op_1057_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1057_keep_dims_0 = const()[name = string("op_1057_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1057_cast_fp16 = reduce_mean(axes = var_1057_axes_0, keep_dims = var_1057_keep_dims_0, x = var_1055_cast_fp16)[name = string("op_1057_cast_fp16")];
            fp32 var_1058_epsilon_0 = const()[name = string("op_1058_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1058_cast_fp16 = inverse(epsilon = var_1058_epsilon_0, x = s_19_cast_fp16)[name = string("op_1058_cast_fp16")];
            fp16 var_1059_to_fp16 = const()[name = string("op_1059_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1060_cast_fp16 = mul(x = var_1058_cast_fp16, y = var_1059_to_fp16)[name = string("op_1060_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1061_cast_fp16 = real_div(x = var_1060_cast_fp16, y = s_19_cast_fp16)[name = string("op_1061_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1062_cast_fp16 = add(x = var_1057_cast_fp16, y = var_1061_cast_fp16)[name = string("op_1062_cast_fp16")];
            fp32 var_1063_epsilon_0 = const()[name = string("op_1063_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1063_cast_fp16 = rsqrt(epsilon = var_1063_epsilon_0, x = var_1062_cast_fp16)[name = string("op_1063_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1064_cast_fp16 = mul(x = y_21_cast_fp16, y = var_1063_cast_fp16)[name = string("op_1064_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_3_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44373760)))];
            tensor<fp16, [1, 96, 768]> input_53_cast_fp16 = mul(x = core_backbone_layers_3_operator_norm_weight_to_fp16, y = var_1064_cast_fp16)[name = string("input_53_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_3_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44375360)))];
            tensor<fp16, [1, 96, 2304]> linear_17_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_3_conv_in_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<int32, [3]> bcx_5_perm_0 = const()[name = string("bcx_5_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_11_begin_0 = const()[name = string("b_11_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_11_end_0 = const()[name = string("b_11_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_11_end_mask_0 = const()[name = string("b_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_5_cast_fp16 = transpose(perm = bcx_5_perm_0, x = linear_17_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [1, 768, 96]> b_11_cast_fp16 = slice_by_index(begin = b_11_begin_0, end = b_11_end_0, end_mask = b_11_end_mask_0, x = bcx_5_cast_fp16)[name = string("b_11_cast_fp16")];
            tensor<int32, [3]> c_5_begin_0 = const()[name = string("c_5_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_5_end_0 = const()[name = string("c_5_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_5_end_mask_0 = const()[name = string("c_5_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_5_cast_fp16 = slice_by_index(begin = c_5_begin_0, end = c_5_end_0, end_mask = c_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("c_5_cast_fp16")];
            tensor<int32, [3]> xx_5_begin_0 = const()[name = string("xx_5_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_5_end_0 = const()[name = string("xx_5_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_5_end_mask_0 = const()[name = string("xx_5_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_5_cast_fp16 = slice_by_index(begin = xx_5_begin_0, end = xx_5_end_0, end_mask = xx_5_end_mask_0, x = bcx_5_cast_fp16)[name = string("xx_5_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_9_cast_fp16 = mul(x = b_11_cast_fp16, y = xx_5_cast_fp16)[name = string("new_9_cast_fp16")];
            tensor<int32, [4]> input_57_begin_0 = const()[name = string("input_57_begin_0"), val = tensor<int32, [4]>([2, 0, 0, 0])];
            tensor<int32, [4]> input_57_end_0 = const()[name = string("input_57_end_0"), val = tensor<int32, [4]>([3, 1, 768, 2])];
            tensor<bool, [4]> input_57_end_mask_0 = const()[name = string("input_57_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_57_squeeze_mask_0 = const()[name = string("input_57_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_57_cast_fp16 = slice_by_index(begin = input_57_begin_0, end = input_57_end_0, end_mask = input_57_end_mask_0, squeeze_mask = input_57_squeeze_mask_0, x = conv)[name = string("input_57_cast_fp16")];
            tensor<int32, [6]> var_1079_pad_0 = const()[name = string("op_1079_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1079_mode_0 = const()[name = string("op_1079_mode_0"), val = string("constant")];
            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1079_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = var_1079_mode_0, pad = var_1079_pad_0, x = new_9_cast_fp16)[name = string("op_1079_cast_fp16")];
            tensor<int32, [6]> var_1081_pad_0 = const()[name = string("op_1081_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_1081_mode_0 = const()[name = string("op_1081_mode_0"), val = string("constant")];
            fp16 const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1081_cast_fp16 = pad(constant_val = const_15_to_fp16, mode = var_1081_mode_0, pad = var_1081_pad_0, x = input_57_cast_fp16)[name = string("op_1081_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_9_cast_fp16 = add(x = var_1079_cast_fp16, y = var_1081_cast_fp16)[name = string("win_9_cast_fp16")];
            tensor<int32, [3]> var_1083_begin_0 = const()[name = string("op_1083_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_1083_end_0 = const()[name = string("op_1083_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_1083_end_mask_0 = const()[name = string("op_1083_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1083_cast_fp16 = slice_by_index(begin = var_1083_begin_0, end = var_1083_end_0, end_mask = var_1083_end_mask_0, x = win_9_cast_fp16)[name = string("op_1083_cast_fp16")];
            string var_1089_pad_type_0 = const()[name = string("op_1089_pad_type_0"), val = string("valid")];
            int32 var_1089_groups_0 = const()[name = string("op_1089_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1089_strides_0 = const()[name = string("op_1089_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1089_pad_0 = const()[name = string("op_1089_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1089_dilations_0 = const()[name = string("op_1089_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_3_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47914368)))];
            tensor<fp16, [1, 768, 96]> var_1089_cast_fp16 = conv(dilations = var_1089_dilations_0, groups = var_1089_groups_0, pad = var_1089_pad_0, pad_type = var_1089_pad_type_0, strides = var_1089_strides_0, weight = core_backbone_layers_3_conv_conv_weight_to_fp16, x = win_9_cast_fp16)[name = string("op_1089_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_1091_cast_fp16 = mul(x = c_5_cast_fp16, y = var_1089_cast_fp16)[name = string("op_1091_cast_fp16")];
            tensor<int32, [3]> input_59_perm_0 = const()[name = string("input_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_3_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_3_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47919040)))];
            tensor<fp16, [1, 96, 768]> input_59_cast_fp16 = transpose(perm = input_59_perm_0, x = var_1091_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [1, 96, 768]> linear_18_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_conv_out_proj_weight_to_fp16, x = input_59_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_41_cast_fp16 = add(x = x_39_cast_fp16, y = linear_18_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1097_cast_fp16 = abs(x = x_41_cast_fp16)[name = string("op_1097_cast_fp16")];
            tensor<int32, [1]> var_1099_axes_0 = const()[name = string("op_1099_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1099_cast_fp16 = reduce_max(axes = var_1099_axes_0, keep_dims = var_127, x = var_1097_cast_fp16)[name = string("op_1099_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_21_cast_fp16 = maximum(x = var_1099_cast_fp16, y = var_128_to_fp16)[name = string("s_21_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_25_cast_fp16 = real_div(x = x_41_cast_fp16, y = s_21_cast_fp16)[name = string("y_25_cast_fp16")];
            fp16 var_129_promoted_9_to_fp16 = const()[name = string("op_129_promoted_9_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1102_cast_fp16 = pow(x = y_25_cast_fp16, y = var_129_promoted_9_to_fp16)[name = string("op_1102_cast_fp16")];
            tensor<int32, [1]> var_1104_axes_0 = const()[name = string("op_1104_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1104_keep_dims_0 = const()[name = string("op_1104_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1104_cast_fp16 = reduce_mean(axes = var_1104_axes_0, keep_dims = var_1104_keep_dims_0, x = var_1102_cast_fp16)[name = string("op_1104_cast_fp16")];
            fp32 var_1105_epsilon_0 = const()[name = string("op_1105_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1105_cast_fp16 = inverse(epsilon = var_1105_epsilon_0, x = s_21_cast_fp16)[name = string("op_1105_cast_fp16")];
            fp16 var_1106_to_fp16 = const()[name = string("op_1106_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1107_cast_fp16 = mul(x = var_1105_cast_fp16, y = var_1106_to_fp16)[name = string("op_1107_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1108_cast_fp16 = real_div(x = var_1107_cast_fp16, y = s_21_cast_fp16)[name = string("op_1108_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1109_cast_fp16 = add(x = var_1104_cast_fp16, y = var_1108_cast_fp16)[name = string("op_1109_cast_fp16")];
            fp32 var_1110_epsilon_0 = const()[name = string("op_1110_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1110_cast_fp16 = rsqrt(epsilon = var_1110_epsilon_0, x = var_1109_cast_fp16)[name = string("op_1110_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1111_cast_fp16 = mul(x = y_25_cast_fp16, y = var_1110_cast_fp16)[name = string("op_1111_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_3_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_3_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49098752)))];
            tensor<fp16, [1, 96, 768]> input_61_cast_fp16 = mul(x = core_backbone_layers_3_ffn_norm_weight_to_fp16, y = var_1111_cast_fp16)[name = string("input_61_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_3_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49100352)))];
            tensor<fp16, [1, 96, 1792]> linear_19_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w1_weight_to_fp16, x = input_61_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_3_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51852928)))];
            tensor<fp16, [1, 96, 1792]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w3_weight_to_fp16, x = input_61_cast_fp16)[name = string("linear_20_cast_fp16")];
            fp16 var_1120_to_fp16 = const()[name = string("op_1120_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1121_cast_fp16 = mul(x = linear_19_cast_fp16, y = var_1120_to_fp16)[name = string("op_1121_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1124_cast_fp16 = tanh(x = var_1121_cast_fp16)[name = string("op_1124_cast_fp16")];
            fp16 var_1125_to_fp16 = const()[name = string("op_1125_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1126_cast_fp16 = add(x = var_1124_cast_fp16, y = var_1125_to_fp16)[name = string("op_1126_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1127_cast_fp16 = mul(x = var_1121_cast_fp16, y = var_1126_cast_fp16)[name = string("op_1127_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_63_cast_fp16 = mul(x = var_1127_cast_fp16, y = linear_20_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_3_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_3_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54605504)))];
            tensor<fp16, [1, 96, 768]> linear_21_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_3_feed_forward_w2_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_45_cast_fp16 = add(x = x_41_cast_fp16, y = linear_21_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1133_cast_fp16 = abs(x = x_45_cast_fp16)[name = string("op_1133_cast_fp16")];
            tensor<int32, [1]> var_1135_axes_0 = const()[name = string("op_1135_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1135_cast_fp16 = reduce_max(axes = var_1135_axes_0, keep_dims = var_127, x = var_1133_cast_fp16)[name = string("op_1135_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_23_cast_fp16 = maximum(x = var_1135_cast_fp16, y = var_128_to_fp16)[name = string("s_23_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_27_cast_fp16 = real_div(x = x_45_cast_fp16, y = s_23_cast_fp16)[name = string("y_27_cast_fp16")];
            fp16 var_129_promoted_10_to_fp16 = const()[name = string("op_129_promoted_10_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1138_cast_fp16 = pow(x = y_27_cast_fp16, y = var_129_promoted_10_to_fp16)[name = string("op_1138_cast_fp16")];
            tensor<int32, [1]> var_1140_axes_0 = const()[name = string("op_1140_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1140_keep_dims_0 = const()[name = string("op_1140_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1140_cast_fp16 = reduce_mean(axes = var_1140_axes_0, keep_dims = var_1140_keep_dims_0, x = var_1138_cast_fp16)[name = string("op_1140_cast_fp16")];
            fp32 var_1141_epsilon_0 = const()[name = string("op_1141_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1141_cast_fp16 = inverse(epsilon = var_1141_epsilon_0, x = s_23_cast_fp16)[name = string("op_1141_cast_fp16")];
            fp16 var_1142_to_fp16 = const()[name = string("op_1142_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1143_cast_fp16 = mul(x = var_1141_cast_fp16, y = var_1142_to_fp16)[name = string("op_1143_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1144_cast_fp16 = real_div(x = var_1143_cast_fp16, y = s_23_cast_fp16)[name = string("op_1144_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1145_cast_fp16 = add(x = var_1140_cast_fp16, y = var_1144_cast_fp16)[name = string("op_1145_cast_fp16")];
            fp32 var_1146_epsilon_0 = const()[name = string("op_1146_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1146_cast_fp16 = rsqrt(epsilon = var_1146_epsilon_0, x = var_1145_cast_fp16)[name = string("op_1146_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1147_cast_fp16 = mul(x = y_27_cast_fp16, y = var_1146_cast_fp16)[name = string("op_1147_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_4_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57358080)))];
            tensor<fp16, [1, 96, 768]> input_65_cast_fp16 = mul(x = core_backbone_layers_4_operator_norm_weight_to_fp16, y = var_1147_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_4_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57359680)))];
            tensor<fp16, [1, 96, 2304]> linear_22_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_4_conv_in_proj_weight_to_fp16, x = input_65_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [3]> bcx_7_perm_0 = const()[name = string("bcx_7_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_15_begin_0 = const()[name = string("b_15_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_15_end_0 = const()[name = string("b_15_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_15_end_mask_0 = const()[name = string("b_15_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_7_cast_fp16 = transpose(perm = bcx_7_perm_0, x = linear_22_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [1, 768, 96]> b_15_cast_fp16 = slice_by_index(begin = b_15_begin_0, end = b_15_end_0, end_mask = b_15_end_mask_0, x = bcx_7_cast_fp16)[name = string("b_15_cast_fp16")];
            tensor<int32, [3]> c_7_begin_0 = const()[name = string("c_7_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_7_end_0 = const()[name = string("c_7_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_7_end_mask_0 = const()[name = string("c_7_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_7_cast_fp16 = slice_by_index(begin = c_7_begin_0, end = c_7_end_0, end_mask = c_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("c_7_cast_fp16")];
            tensor<int32, [3]> xx_7_begin_0 = const()[name = string("xx_7_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_7_end_0 = const()[name = string("xx_7_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_7_end_mask_0 = const()[name = string("xx_7_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_7_cast_fp16 = slice_by_index(begin = xx_7_begin_0, end = xx_7_end_0, end_mask = xx_7_end_mask_0, x = bcx_7_cast_fp16)[name = string("xx_7_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_11_cast_fp16 = mul(x = b_15_cast_fp16, y = xx_7_cast_fp16)[name = string("new_11_cast_fp16")];
            tensor<int32, [4]> input_69_begin_0 = const()[name = string("input_69_begin_0"), val = tensor<int32, [4]>([3, 0, 0, 0])];
            tensor<int32, [4]> input_69_end_0 = const()[name = string("input_69_end_0"), val = tensor<int32, [4]>([4, 1, 768, 2])];
            tensor<bool, [4]> input_69_end_mask_0 = const()[name = string("input_69_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_69_squeeze_mask_0 = const()[name = string("input_69_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_69_cast_fp16 = slice_by_index(begin = input_69_begin_0, end = input_69_end_0, end_mask = input_69_end_mask_0, squeeze_mask = input_69_squeeze_mask_0, x = conv)[name = string("input_69_cast_fp16")];
            tensor<int32, [6]> var_1162_pad_0 = const()[name = string("op_1162_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1162_mode_0 = const()[name = string("op_1162_mode_0"), val = string("constant")];
            fp16 const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1162_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = var_1162_mode_0, pad = var_1162_pad_0, x = new_11_cast_fp16)[name = string("op_1162_cast_fp16")];
            tensor<int32, [6]> var_1164_pad_0 = const()[name = string("op_1164_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_1164_mode_0 = const()[name = string("op_1164_mode_0"), val = string("constant")];
            fp16 const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1164_cast_fp16 = pad(constant_val = const_17_to_fp16, mode = var_1164_mode_0, pad = var_1164_pad_0, x = input_69_cast_fp16)[name = string("op_1164_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_13_cast_fp16 = add(x = var_1162_cast_fp16, y = var_1164_cast_fp16)[name = string("win_13_cast_fp16")];
            tensor<int32, [3]> var_1166_begin_0 = const()[name = string("op_1166_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_1166_end_0 = const()[name = string("op_1166_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_1166_end_mask_0 = const()[name = string("op_1166_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1166_cast_fp16 = slice_by_index(begin = var_1166_begin_0, end = var_1166_end_0, end_mask = var_1166_end_mask_0, x = win_13_cast_fp16)[name = string("op_1166_cast_fp16")];
            string var_1172_pad_type_0 = const()[name = string("op_1172_pad_type_0"), val = string("valid")];
            int32 var_1172_groups_0 = const()[name = string("op_1172_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1172_strides_0 = const()[name = string("op_1172_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1172_pad_0 = const()[name = string("op_1172_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1172_dilations_0 = const()[name = string("op_1172_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_4_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60898688)))];
            tensor<fp16, [1, 768, 96]> var_1172_cast_fp16 = conv(dilations = var_1172_dilations_0, groups = var_1172_groups_0, pad = var_1172_pad_0, pad_type = var_1172_pad_type_0, strides = var_1172_strides_0, weight = core_backbone_layers_4_conv_conv_weight_to_fp16, x = win_13_cast_fp16)[name = string("op_1172_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_1174_cast_fp16 = mul(x = c_7_cast_fp16, y = var_1172_cast_fp16)[name = string("op_1174_cast_fp16")];
            tensor<int32, [3]> input_71_perm_0 = const()[name = string("input_71_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_4_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_4_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60903360)))];
            tensor<fp16, [1, 96, 768]> input_71_cast_fp16 = transpose(perm = input_71_perm_0, x = var_1174_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [1, 96, 768]> linear_23_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_4_conv_out_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_47_cast_fp16 = add(x = x_45_cast_fp16, y = linear_23_cast_fp16)[name = string("x_47_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1180_cast_fp16 = abs(x = x_47_cast_fp16)[name = string("op_1180_cast_fp16")];
            tensor<int32, [1]> var_1182_axes_0 = const()[name = string("op_1182_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1182_cast_fp16 = reduce_max(axes = var_1182_axes_0, keep_dims = var_127, x = var_1180_cast_fp16)[name = string("op_1182_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_25_cast_fp16 = maximum(x = var_1182_cast_fp16, y = var_128_to_fp16)[name = string("s_25_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_31_cast_fp16 = real_div(x = x_47_cast_fp16, y = s_25_cast_fp16)[name = string("y_31_cast_fp16")];
            fp16 var_129_promoted_11_to_fp16 = const()[name = string("op_129_promoted_11_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1185_cast_fp16 = pow(x = y_31_cast_fp16, y = var_129_promoted_11_to_fp16)[name = string("op_1185_cast_fp16")];
            tensor<int32, [1]> var_1187_axes_0 = const()[name = string("op_1187_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1187_keep_dims_0 = const()[name = string("op_1187_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1187_cast_fp16 = reduce_mean(axes = var_1187_axes_0, keep_dims = var_1187_keep_dims_0, x = var_1185_cast_fp16)[name = string("op_1187_cast_fp16")];
            fp32 var_1188_epsilon_0 = const()[name = string("op_1188_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1188_cast_fp16 = inverse(epsilon = var_1188_epsilon_0, x = s_25_cast_fp16)[name = string("op_1188_cast_fp16")];
            fp16 var_1189_to_fp16 = const()[name = string("op_1189_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1190_cast_fp16 = mul(x = var_1188_cast_fp16, y = var_1189_to_fp16)[name = string("op_1190_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1191_cast_fp16 = real_div(x = var_1190_cast_fp16, y = s_25_cast_fp16)[name = string("op_1191_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1192_cast_fp16 = add(x = var_1187_cast_fp16, y = var_1191_cast_fp16)[name = string("op_1192_cast_fp16")];
            fp32 var_1193_epsilon_0 = const()[name = string("op_1193_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1193_cast_fp16 = rsqrt(epsilon = var_1193_epsilon_0, x = var_1192_cast_fp16)[name = string("op_1193_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1194_cast_fp16 = mul(x = y_31_cast_fp16, y = var_1193_cast_fp16)[name = string("op_1194_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_4_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_4_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62083072)))];
            tensor<fp16, [1, 96, 768]> input_73_cast_fp16 = mul(x = core_backbone_layers_4_ffn_norm_weight_to_fp16, y = var_1194_cast_fp16)[name = string("input_73_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_4_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62084672)))];
            tensor<fp16, [1, 96, 1792]> linear_24_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w1_weight_to_fp16, x = input_73_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_4_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64837248)))];
            tensor<fp16, [1, 96, 1792]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w3_weight_to_fp16, x = input_73_cast_fp16)[name = string("linear_25_cast_fp16")];
            fp16 var_1203_to_fp16 = const()[name = string("op_1203_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1204_cast_fp16 = mul(x = linear_24_cast_fp16, y = var_1203_to_fp16)[name = string("op_1204_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1207_cast_fp16 = tanh(x = var_1204_cast_fp16)[name = string("op_1207_cast_fp16")];
            fp16 var_1208_to_fp16 = const()[name = string("op_1208_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1209_cast_fp16 = add(x = var_1207_cast_fp16, y = var_1208_to_fp16)[name = string("op_1209_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1210_cast_fp16 = mul(x = var_1204_cast_fp16, y = var_1209_cast_fp16)[name = string("op_1210_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_75_cast_fp16 = mul(x = var_1210_cast_fp16, y = linear_25_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_4_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_4_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67589824)))];
            tensor<fp16, [1, 96, 768]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_4_feed_forward_w2_weight_to_fp16, x = input_75_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_51_cast_fp16 = add(x = x_47_cast_fp16, y = linear_26_cast_fp16)[name = string("x_51_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1216_cast_fp16 = abs(x = x_51_cast_fp16)[name = string("op_1216_cast_fp16")];
            tensor<int32, [1]> var_1218_axes_0 = const()[name = string("op_1218_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1218_cast_fp16 = reduce_max(axes = var_1218_axes_0, keep_dims = var_127, x = var_1216_cast_fp16)[name = string("op_1218_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_27_cast_fp16 = maximum(x = var_1218_cast_fp16, y = var_128_to_fp16)[name = string("s_27_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_33_cast_fp16 = real_div(x = x_51_cast_fp16, y = s_27_cast_fp16)[name = string("y_33_cast_fp16")];
            fp16 var_129_promoted_12_to_fp16 = const()[name = string("op_129_promoted_12_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1221_cast_fp16 = pow(x = y_33_cast_fp16, y = var_129_promoted_12_to_fp16)[name = string("op_1221_cast_fp16")];
            tensor<int32, [1]> var_1223_axes_0 = const()[name = string("op_1223_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1223_keep_dims_0 = const()[name = string("op_1223_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1223_cast_fp16 = reduce_mean(axes = var_1223_axes_0, keep_dims = var_1223_keep_dims_0, x = var_1221_cast_fp16)[name = string("op_1223_cast_fp16")];
            fp32 var_1224_epsilon_0 = const()[name = string("op_1224_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1224_cast_fp16 = inverse(epsilon = var_1224_epsilon_0, x = s_27_cast_fp16)[name = string("op_1224_cast_fp16")];
            fp16 var_1225_to_fp16 = const()[name = string("op_1225_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1226_cast_fp16 = mul(x = var_1224_cast_fp16, y = var_1225_to_fp16)[name = string("op_1226_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1227_cast_fp16 = real_div(x = var_1226_cast_fp16, y = s_27_cast_fp16)[name = string("op_1227_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1228_cast_fp16 = add(x = var_1223_cast_fp16, y = var_1227_cast_fp16)[name = string("op_1228_cast_fp16")];
            fp32 var_1229_epsilon_0 = const()[name = string("op_1229_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1229_cast_fp16 = rsqrt(epsilon = var_1229_epsilon_0, x = var_1228_cast_fp16)[name = string("op_1229_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1230_cast_fp16 = mul(x = y_33_cast_fp16, y = var_1229_cast_fp16)[name = string("op_1230_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_5_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70342400)))];
            tensor<fp16, [1, 96, 768]> input_77_cast_fp16 = mul(x = core_backbone_layers_5_operator_norm_weight_to_fp16, y = var_1230_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70344000)))];
            tensor<fp16, [1, 96, 768]> linear_27_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_q_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_27_cast_fp16")];
            tensor<int32, [4]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [4]>([1, 96, 8, 96])];
            tensor<fp16, [1, 96, 8, 96]> x_53_cast_fp16 = reshape(shape = concat_9, x = linear_27_cast_fp16)[name = string("x_53_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1237_cast_fp16 = abs(x = x_53_cast_fp16)[name = string("op_1237_cast_fp16")];
            tensor<int32, [1]> var_1239_axes_0 = const()[name = string("op_1239_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 8, 1]> var_1239_cast_fp16 = reduce_max(axes = var_1239_axes_0, keep_dims = var_127, x = var_1237_cast_fp16)[name = string("op_1239_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> s_29_cast_fp16 = maximum(x = var_1239_cast_fp16, y = var_128_to_fp16)[name = string("s_29_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> y_35_cast_fp16 = real_div(x = x_53_cast_fp16, y = s_29_cast_fp16)[name = string("y_35_cast_fp16")];
            fp16 var_129_promoted_13_to_fp16 = const()[name = string("op_129_promoted_13_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 8, 96]> var_1242_cast_fp16 = pow(x = y_35_cast_fp16, y = var_129_promoted_13_to_fp16)[name = string("op_1242_cast_fp16")];
            tensor<int32, [1]> var_1244_axes_0 = const()[name = string("op_1244_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1244_keep_dims_0 = const()[name = string("op_1244_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 8, 1]> var_1244_cast_fp16 = reduce_mean(axes = var_1244_axes_0, keep_dims = var_1244_keep_dims_0, x = var_1242_cast_fp16)[name = string("op_1244_cast_fp16")];
            fp32 var_1245_epsilon_0 = const()[name = string("op_1245_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 8, 1]> var_1245_cast_fp16 = inverse(epsilon = var_1245_epsilon_0, x = s_29_cast_fp16)[name = string("op_1245_cast_fp16")];
            fp16 var_1246_to_fp16 = const()[name = string("op_1246_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 8, 1]> var_1247_cast_fp16 = mul(x = var_1245_cast_fp16, y = var_1246_to_fp16)[name = string("op_1247_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_1248_cast_fp16 = real_div(x = var_1247_cast_fp16, y = s_29_cast_fp16)[name = string("op_1248_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_1249_cast_fp16 = add(x = var_1244_cast_fp16, y = var_1248_cast_fp16)[name = string("op_1249_cast_fp16")];
            fp32 var_1250_epsilon_0 = const()[name = string("op_1250_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 8, 1]> var_1250_cast_fp16 = rsqrt(epsilon = var_1250_epsilon_0, x = var_1249_cast_fp16)[name = string("op_1250_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1251_cast_fp16 = mul(x = y_35_cast_fp16, y = var_1250_cast_fp16)[name = string("op_1251_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_5_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71523712)))];
            tensor<fp16, [1, 96, 8, 96]> var_1252_cast_fp16 = mul(x = core_backbone_layers_5_self_attn_q_norm_weight_to_fp16, y = var_1251_cast_fp16)[name = string("op_1252_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71523968)))];
            tensor<fp16, [1, 96, 384]> linear_28_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_k_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [4]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> x_55_cast_fp16 = reshape(shape = concat_10, x = linear_28_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_1259_cast_fp16 = abs(x = x_55_cast_fp16)[name = string("op_1259_cast_fp16")];
            tensor<int32, [1]> var_1261_axes_0 = const()[name = string("op_1261_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 4, 1]> var_1261_cast_fp16 = reduce_max(axes = var_1261_axes_0, keep_dims = var_127, x = var_1259_cast_fp16)[name = string("op_1261_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> s_31_cast_fp16 = maximum(x = var_1261_cast_fp16, y = var_128_to_fp16)[name = string("s_31_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> y_37_cast_fp16 = real_div(x = x_55_cast_fp16, y = s_31_cast_fp16)[name = string("y_37_cast_fp16")];
            fp16 var_129_promoted_14_to_fp16 = const()[name = string("op_129_promoted_14_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 4, 96]> var_1264_cast_fp16 = pow(x = y_37_cast_fp16, y = var_129_promoted_14_to_fp16)[name = string("op_1264_cast_fp16")];
            tensor<int32, [1]> var_1266_axes_0 = const()[name = string("op_1266_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1266_keep_dims_0 = const()[name = string("op_1266_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 4, 1]> var_1266_cast_fp16 = reduce_mean(axes = var_1266_axes_0, keep_dims = var_1266_keep_dims_0, x = var_1264_cast_fp16)[name = string("op_1266_cast_fp16")];
            fp32 var_1267_epsilon_0 = const()[name = string("op_1267_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 4, 1]> var_1267_cast_fp16 = inverse(epsilon = var_1267_epsilon_0, x = s_31_cast_fp16)[name = string("op_1267_cast_fp16")];
            fp16 var_1268_to_fp16 = const()[name = string("op_1268_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 4, 1]> var_1269_cast_fp16 = mul(x = var_1267_cast_fp16, y = var_1268_to_fp16)[name = string("op_1269_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_1270_cast_fp16 = real_div(x = var_1269_cast_fp16, y = s_31_cast_fp16)[name = string("op_1270_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_1271_cast_fp16 = add(x = var_1266_cast_fp16, y = var_1270_cast_fp16)[name = string("op_1271_cast_fp16")];
            fp32 var_1272_epsilon_0 = const()[name = string("op_1272_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 4, 1]> var_1272_cast_fp16 = rsqrt(epsilon = var_1272_epsilon_0, x = var_1271_cast_fp16)[name = string("op_1272_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_1273_cast_fp16 = mul(x = y_37_cast_fp16, y = var_1272_cast_fp16)[name = string("op_1273_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_5_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72113856)))];
            tensor<fp16, [1, 96, 4, 96]> var_1274_cast_fp16 = mul(x = core_backbone_layers_5_self_attn_k_norm_weight_to_fp16, y = var_1273_cast_fp16)[name = string("op_1274_cast_fp16")];
            tensor<int32, [4]> new_13_perm_0 = const()[name = string("new_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72114112)))];
            tensor<fp16, [1, 96, 384]> linear_29_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_v_proj_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<int32, [4]> concat_11 = const()[name = string("concat_11"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> var_1279_cast_fp16 = reshape(shape = concat_11, x = linear_29_cast_fp16)[name = string("op_1279_cast_fp16")];
            tensor<int32, [4]> new_15_perm_0 = const()[name = string("new_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_81_begin_0 = const()[name = string("input_81_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_81_end_0 = const()[name = string("input_81_end_0"), val = tensor<int32, [5]>([2, 1, 4, 31, 96])];
            tensor<bool, [5]> input_81_end_mask_0 = const()[name = string("input_81_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_81_squeeze_mask_0 = const()[name = string("input_81_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_81_cast_fp16 = slice_by_index(begin = input_81_begin_0, end = input_81_end_0, end_mask = input_81_end_mask_0, squeeze_mask = input_81_squeeze_mask_0, x = k)[name = string("input_81_cast_fp16")];
            tensor<int32, [8]> var_1284_pad_0 = const()[name = string("op_1284_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1284_mode_0 = const()[name = string("op_1284_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_13_cast_fp16 = transpose(perm = new_13_perm_0, x = var_1274_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [1, 4, 127, 96]> var_1284_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = var_1284_mode_0, pad = var_1284_pad_0, x = new_13_cast_fp16)[name = string("op_1284_cast_fp16")];
            tensor<int32, [8]> var_1286_pad_0 = const()[name = string("op_1286_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_1286_mode_0 = const()[name = string("op_1286_mode_0"), val = string("constant")];
            fp16 const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_1286_cast_fp16 = pad(constant_val = const_19_to_fp16, mode = var_1286_mode_0, pad = var_1286_pad_0, x = input_81_cast_fp16)[name = string("op_1286_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> x_59_cast_fp16 = add(x = var_1284_cast_fp16, y = var_1286_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<int32, [5]> input_85_begin_0 = const()[name = string("input_85_begin_0"), val = tensor<int32, [5]>([1, 0, 0, 0, 0])];
            tensor<int32, [5]> input_85_end_0 = const()[name = string("input_85_end_0"), val = tensor<int32, [5]>([2, 1, 4, 31, 96])];
            tensor<bool, [5]> input_85_end_mask_0 = const()[name = string("input_85_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_85_squeeze_mask_0 = const()[name = string("input_85_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_85_cast_fp16 = slice_by_index(begin = input_85_begin_0, end = input_85_end_0, end_mask = input_85_end_mask_0, squeeze_mask = input_85_squeeze_mask_0, x = v)[name = string("input_85_cast_fp16")];
            tensor<int32, [8]> var_1291_pad_0 = const()[name = string("op_1291_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1291_mode_0 = const()[name = string("op_1291_mode_0"), val = string("constant")];
            fp16 const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_15_cast_fp16 = transpose(perm = new_15_perm_0, x = var_1279_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [1, 4, 127, 96]> var_1291_cast_fp16 = pad(constant_val = const_20_to_fp16, mode = var_1291_mode_0, pad = var_1291_pad_0, x = new_15_cast_fp16)[name = string("op_1291_cast_fp16")];
            tensor<int32, [8]> var_1293_pad_0 = const()[name = string("op_1293_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_1293_mode_0 = const()[name = string("op_1293_mode_0"), val = string("constant")];
            fp16 const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_1293_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = var_1293_mode_0, pad = var_1293_pad_0, x = input_85_cast_fp16)[name = string("op_1293_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> v_7_cast_fp16 = add(x = var_1291_cast_fp16, y = var_1293_cast_fp16)[name = string("v_7_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1295_cast_fp16 = mul(x = var_1252_cast_fp16, y = const_66_to_fp16)[name = string("op_1295_cast_fp16")];
            tensor<int32, [2]> var_1296_split_sizes_0 = const()[name = string("op_1296_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_71 = const()[name = string("const_71"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 48]> var_1296_cast_fp16_0, tensor<fp16, [1, 96, 8, 48]> var_1296_cast_fp16_1 = split(axis = const_71, split_sizes = var_1296_split_sizes_0, x = var_1252_cast_fp16)[name = string("op_1296_cast_fp16")];
            fp16 const_22_promoted_to_fp16 = const()[name = string("const_22_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 96, 8, 48]> var_1298_cast_fp16 = mul(x = var_1296_cast_fp16_1, y = const_22_promoted_to_fp16)[name = string("op_1298_cast_fp16")];
            bool var_1300_interleave_0 = const()[name = string("op_1300_interleave_0"), val = bool(false)];
            int32 const_72 = const()[name = string("const_72"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 96]> var_1300_cast_fp16 = concat(axis = const_72, interleave = var_1300_interleave_0, values = (var_1298_cast_fp16, var_1296_cast_fp16_0))[name = string("op_1300_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1301_cast_fp16 = mul(x = var_1300_cast_fp16, y = const_69_to_fp16)[name = string("op_1301_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> q_3_cast_fp16 = add(x = var_1295_cast_fp16, y = var_1301_cast_fp16)[name = string("q_3_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_1303_cast_fp16 = mul(x = x_59_cast_fp16, y = const_0_to_fp16)[name = string("op_1303_cast_fp16")];
            tensor<int32, [2]> var_1304_split_sizes_0 = const()[name = string("op_1304_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1304_axis_0 = const()[name = string("op_1304_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 127, 48]> var_1304_cast_fp16_0, tensor<fp16, [1, 4, 127, 48]> var_1304_cast_fp16_1 = split(axis = var_1304_axis_0, split_sizes = var_1304_split_sizes_0, x = x_59_cast_fp16)[name = string("op_1304_cast_fp16")];
            fp16 const_23_promoted_to_fp16 = const()[name = string("const_23_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 127, 48]> var_1306_cast_fp16 = mul(x = var_1304_cast_fp16_1, y = const_23_promoted_to_fp16)[name = string("op_1306_cast_fp16")];
            bool var_1308_interleave_0 = const()[name = string("op_1308_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 127, 96]> var_1308_cast_fp16 = concat(axis = var_126, interleave = var_1308_interleave_0, values = (var_1306_cast_fp16, var_1304_cast_fp16_0))[name = string("op_1308_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_1309_cast_fp16 = mul(x = var_1308_cast_fp16, y = const_1_to_fp16)[name = string("op_1309_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> k_7_cast_fp16 = add(x = var_1303_cast_fp16, y = var_1309_cast_fp16)[name = string("k_7_cast_fp16")];
            tensor<int32, [1]> var_1311_axes_0 = const()[name = string("op_1311_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_1311_cast_fp16 = expand_dims(axes = var_1311_axes_0, x = k_7_cast_fp16)[name = string("op_1311_cast_fp16")];
            tensor<int32, [5]> var_1313_reps_0 = const()[name = string("op_1313_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_1313_cast_fp16 = tile(reps = var_1313_reps_0, x = var_1311_cast_fp16)[name = string("op_1313_cast_fp16")];
            tensor<int32, [4]> concat_13 = const()[name = string("concat_13"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> k_9_cast_fp16 = reshape(shape = concat_13, x = var_1313_cast_fp16)[name = string("k_9_cast_fp16")];
            tensor<int32, [1]> var_1315_axes_0 = const()[name = string("op_1315_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_1315_cast_fp16 = expand_dims(axes = var_1315_axes_0, x = v_7_cast_fp16)[name = string("op_1315_cast_fp16")];
            tensor<int32, [5]> var_1317_reps_0 = const()[name = string("op_1317_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_1317_cast_fp16 = tile(reps = var_1317_reps_0, x = var_1315_cast_fp16)[name = string("op_1317_cast_fp16")];
            tensor<int32, [4]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> v_9_cast_fp16 = reshape(shape = concat_15, x = var_1317_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> transpose_2_perm_0 = const()[name = string("transpose_2_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1320_transpose_x_1 = const()[name = string("op_1320_transpose_x_1"), val = bool(false)];
            bool var_1320_transpose_y_1 = const()[name = string("op_1320_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 96]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = q_3_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [1, 8, 96, 127]> var_1320_cast_fp16 = matmul(transpose_x = var_1320_transpose_x_1, transpose_y = var_1320_transpose_y_1, x = transpose_2_cast_fp16, y = k_9_cast_fp16)[name = string("op_1320_cast_fp16")];
            fp16 var_1321_to_fp16 = const()[name = string("op_1321_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 96, 127]> s_33_cast_fp16 = mul(x = var_1320_cast_fp16, y = var_1321_to_fp16)[name = string("s_33_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_61_cast_fp16 = add(x = s_33_cast_fp16, y = mask)[name = string("x_61_cast_fp16")];
            tensor<int32, [1]> var_1325_axes_0 = const()[name = string("op_1325_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 96, 1]> var_1325_cast_fp16 = reduce_max(axes = var_1325_axes_0, keep_dims = var_127, x = x_61_cast_fp16)[name = string("op_1325_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_63_cast_fp16 = sub(x = x_61_cast_fp16, y = var_1325_cast_fp16)[name = string("x_63_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_65_cast_fp16 = exp(x = x_63_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [1]> var_1329_axes_0 = const()[name = string("op_1329_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1329_keep_dims_0 = const()[name = string("op_1329_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 1]> var_1329_cast_fp16 = reduce_sum(axes = var_1329_axes_0, keep_dims = var_1329_keep_dims_0, x = x_65_cast_fp16)[name = string("op_1329_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> var_1330_cast_fp16 = real_div(x = x_65_cast_fp16, y = var_1329_cast_fp16)[name = string("op_1330_cast_fp16")];
            bool o_5_transpose_x_0 = const()[name = string("o_5_transpose_x_0"), val = bool(false)];
            bool o_5_transpose_y_0 = const()[name = string("o_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 96, 96]> o_5_cast_fp16 = matmul(transpose_x = o_5_transpose_x_0, transpose_y = o_5_transpose_y_0, x = var_1330_cast_fp16, y = v_9_cast_fp16)[name = string("o_5_cast_fp16")];
            tensor<int32, [4]> o_7_perm_0 = const()[name = string("o_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [3]>([1, 96, 768])];
            tensor<fp16, [1, 96, 8, 96]> o_7_cast_fp16 = transpose(perm = o_7_perm_0, x = o_5_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [1, 96, 768]> input_87_cast_fp16 = reshape(shape = concat_16, x = o_7_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72704000)))];
            tensor<fp16, [1, 96, 768]> linear_30_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_self_attn_o_proj_weight_to_fp16, x = input_87_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> var_1339_begin_0 = const()[name = string("op_1339_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_1339_end_0 = const()[name = string("op_1339_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_1339_end_mask_0 = const()[name = string("op_1339_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1339_cast_fp16 = slice_by_index(begin = var_1339_begin_0, end = var_1339_end_0, end_mask = var_1339_end_mask_0, x = x_59_cast_fp16)[name = string("op_1339_cast_fp16")];
            tensor<int32, [4]> var_1342_begin_0 = const()[name = string("op_1342_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_1342_end_0 = const()[name = string("op_1342_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_1342_end_mask_0 = const()[name = string("op_1342_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1342_cast_fp16 = slice_by_index(begin = var_1342_begin_0, end = var_1342_end_0, end_mask = var_1342_end_mask_0, x = v_7_cast_fp16)[name = string("op_1342_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_67_cast_fp16 = add(x = x_51_cast_fp16, y = linear_30_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1345_cast_fp16 = abs(x = x_67_cast_fp16)[name = string("op_1345_cast_fp16")];
            tensor<int32, [1]> var_1347_axes_0 = const()[name = string("op_1347_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1347_cast_fp16 = reduce_max(axes = var_1347_axes_0, keep_dims = var_127, x = var_1345_cast_fp16)[name = string("op_1347_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_35_cast_fp16 = maximum(x = var_1347_cast_fp16, y = var_128_to_fp16)[name = string("s_35_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_39_cast_fp16 = real_div(x = x_67_cast_fp16, y = s_35_cast_fp16)[name = string("y_39_cast_fp16")];
            fp16 var_129_promoted_15_to_fp16 = const()[name = string("op_129_promoted_15_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1350_cast_fp16 = pow(x = y_39_cast_fp16, y = var_129_promoted_15_to_fp16)[name = string("op_1350_cast_fp16")];
            tensor<int32, [1]> var_1352_axes_0 = const()[name = string("op_1352_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1352_keep_dims_0 = const()[name = string("op_1352_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1352_cast_fp16 = reduce_mean(axes = var_1352_axes_0, keep_dims = var_1352_keep_dims_0, x = var_1350_cast_fp16)[name = string("op_1352_cast_fp16")];
            fp32 var_1353_epsilon_0 = const()[name = string("op_1353_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1353_cast_fp16 = inverse(epsilon = var_1353_epsilon_0, x = s_35_cast_fp16)[name = string("op_1353_cast_fp16")];
            fp16 var_1354_to_fp16 = const()[name = string("op_1354_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1355_cast_fp16 = mul(x = var_1353_cast_fp16, y = var_1354_to_fp16)[name = string("op_1355_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1356_cast_fp16 = real_div(x = var_1355_cast_fp16, y = s_35_cast_fp16)[name = string("op_1356_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1357_cast_fp16 = add(x = var_1352_cast_fp16, y = var_1356_cast_fp16)[name = string("op_1357_cast_fp16")];
            fp32 var_1358_epsilon_0 = const()[name = string("op_1358_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1358_cast_fp16 = rsqrt(epsilon = var_1358_epsilon_0, x = var_1357_cast_fp16)[name = string("op_1358_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1359_cast_fp16 = mul(x = y_39_cast_fp16, y = var_1358_cast_fp16)[name = string("op_1359_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_5_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_5_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73883712)))];
            tensor<fp16, [1, 96, 768]> input_89_cast_fp16 = mul(x = core_backbone_layers_5_ffn_norm_weight_to_fp16, y = var_1359_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_5_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73885312)))];
            tensor<fp16, [1, 96, 1792]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w1_weight_to_fp16, x = input_89_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_5_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76637888)))];
            tensor<fp16, [1, 96, 1792]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w3_weight_to_fp16, x = input_89_cast_fp16)[name = string("linear_32_cast_fp16")];
            fp16 var_1368_to_fp16 = const()[name = string("op_1368_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1369_cast_fp16 = mul(x = linear_31_cast_fp16, y = var_1368_to_fp16)[name = string("op_1369_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1372_cast_fp16 = tanh(x = var_1369_cast_fp16)[name = string("op_1372_cast_fp16")];
            fp16 var_1373_to_fp16 = const()[name = string("op_1373_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1374_cast_fp16 = add(x = var_1372_cast_fp16, y = var_1373_to_fp16)[name = string("op_1374_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1375_cast_fp16 = mul(x = var_1369_cast_fp16, y = var_1374_cast_fp16)[name = string("op_1375_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_91_cast_fp16 = mul(x = var_1375_cast_fp16, y = linear_32_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_5_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_5_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79390464)))];
            tensor<fp16, [1, 96, 768]> linear_33_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_5_feed_forward_w2_weight_to_fp16, x = input_91_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_71_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1381_cast_fp16 = abs(x = x_71_cast_fp16)[name = string("op_1381_cast_fp16")];
            tensor<int32, [1]> var_1383_axes_0 = const()[name = string("op_1383_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1383_cast_fp16 = reduce_max(axes = var_1383_axes_0, keep_dims = var_127, x = var_1381_cast_fp16)[name = string("op_1383_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_37_cast_fp16 = maximum(x = var_1383_cast_fp16, y = var_128_to_fp16)[name = string("s_37_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_41_cast_fp16 = real_div(x = x_71_cast_fp16, y = s_37_cast_fp16)[name = string("y_41_cast_fp16")];
            fp16 var_129_promoted_16_to_fp16 = const()[name = string("op_129_promoted_16_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1386_cast_fp16 = pow(x = y_41_cast_fp16, y = var_129_promoted_16_to_fp16)[name = string("op_1386_cast_fp16")];
            tensor<int32, [1]> var_1388_axes_0 = const()[name = string("op_1388_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1388_keep_dims_0 = const()[name = string("op_1388_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1388_cast_fp16 = reduce_mean(axes = var_1388_axes_0, keep_dims = var_1388_keep_dims_0, x = var_1386_cast_fp16)[name = string("op_1388_cast_fp16")];
            fp32 var_1389_epsilon_0 = const()[name = string("op_1389_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1389_cast_fp16 = inverse(epsilon = var_1389_epsilon_0, x = s_37_cast_fp16)[name = string("op_1389_cast_fp16")];
            fp16 var_1390_to_fp16 = const()[name = string("op_1390_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1391_cast_fp16 = mul(x = var_1389_cast_fp16, y = var_1390_to_fp16)[name = string("op_1391_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1392_cast_fp16 = real_div(x = var_1391_cast_fp16, y = s_37_cast_fp16)[name = string("op_1392_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1393_cast_fp16 = add(x = var_1388_cast_fp16, y = var_1392_cast_fp16)[name = string("op_1393_cast_fp16")];
            fp32 var_1394_epsilon_0 = const()[name = string("op_1394_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1394_cast_fp16 = rsqrt(epsilon = var_1394_epsilon_0, x = var_1393_cast_fp16)[name = string("op_1394_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1395_cast_fp16 = mul(x = y_41_cast_fp16, y = var_1394_cast_fp16)[name = string("op_1395_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_6_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82143040)))];
            tensor<fp16, [1, 96, 768]> input_93_cast_fp16 = mul(x = core_backbone_layers_6_operator_norm_weight_to_fp16, y = var_1395_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_6_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82144640)))];
            tensor<fp16, [1, 96, 2304]> linear_34_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_6_conv_in_proj_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<int32, [3]> bcx_9_perm_0 = const()[name = string("bcx_9_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_21_begin_0 = const()[name = string("b_21_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_21_end_0 = const()[name = string("b_21_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_21_end_mask_0 = const()[name = string("b_21_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_9_cast_fp16 = transpose(perm = bcx_9_perm_0, x = linear_34_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [1, 768, 96]> b_21_cast_fp16 = slice_by_index(begin = b_21_begin_0, end = b_21_end_0, end_mask = b_21_end_mask_0, x = bcx_9_cast_fp16)[name = string("b_21_cast_fp16")];
            tensor<int32, [3]> c_9_begin_0 = const()[name = string("c_9_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_9_end_0 = const()[name = string("c_9_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_9_end_mask_0 = const()[name = string("c_9_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_9_cast_fp16 = slice_by_index(begin = c_9_begin_0, end = c_9_end_0, end_mask = c_9_end_mask_0, x = bcx_9_cast_fp16)[name = string("c_9_cast_fp16")];
            tensor<int32, [3]> xx_9_begin_0 = const()[name = string("xx_9_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_9_end_0 = const()[name = string("xx_9_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_9_end_mask_0 = const()[name = string("xx_9_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_9_cast_fp16 = slice_by_index(begin = xx_9_begin_0, end = xx_9_end_0, end_mask = xx_9_end_mask_0, x = bcx_9_cast_fp16)[name = string("xx_9_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_17_cast_fp16 = mul(x = b_21_cast_fp16, y = xx_9_cast_fp16)[name = string("new_17_cast_fp16")];
            tensor<int32, [4]> input_97_begin_0 = const()[name = string("input_97_begin_0"), val = tensor<int32, [4]>([4, 0, 0, 0])];
            tensor<int32, [4]> input_97_end_0 = const()[name = string("input_97_end_0"), val = tensor<int32, [4]>([5, 1, 768, 2])];
            tensor<bool, [4]> input_97_end_mask_0 = const()[name = string("input_97_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_97_squeeze_mask_0 = const()[name = string("input_97_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_97_cast_fp16 = slice_by_index(begin = input_97_begin_0, end = input_97_end_0, end_mask = input_97_end_mask_0, squeeze_mask = input_97_squeeze_mask_0, x = conv)[name = string("input_97_cast_fp16")];
            tensor<int32, [6]> var_1410_pad_0 = const()[name = string("op_1410_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1410_mode_0 = const()[name = string("op_1410_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1410_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_1410_mode_0, pad = var_1410_pad_0, x = new_17_cast_fp16)[name = string("op_1410_cast_fp16")];
            tensor<int32, [6]> var_1412_pad_0 = const()[name = string("op_1412_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_1412_mode_0 = const()[name = string("op_1412_mode_0"), val = string("constant")];
            fp16 const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1412_cast_fp16 = pad(constant_val = const_25_to_fp16, mode = var_1412_mode_0, pad = var_1412_pad_0, x = input_97_cast_fp16)[name = string("op_1412_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_17_cast_fp16 = add(x = var_1410_cast_fp16, y = var_1412_cast_fp16)[name = string("win_17_cast_fp16")];
            tensor<int32, [3]> var_1414_begin_0 = const()[name = string("op_1414_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_1414_end_0 = const()[name = string("op_1414_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_1414_end_mask_0 = const()[name = string("op_1414_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1414_cast_fp16 = slice_by_index(begin = var_1414_begin_0, end = var_1414_end_0, end_mask = var_1414_end_mask_0, x = win_17_cast_fp16)[name = string("op_1414_cast_fp16")];
            string var_1420_pad_type_0 = const()[name = string("op_1420_pad_type_0"), val = string("valid")];
            int32 var_1420_groups_0 = const()[name = string("op_1420_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1420_strides_0 = const()[name = string("op_1420_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1420_pad_0 = const()[name = string("op_1420_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1420_dilations_0 = const()[name = string("op_1420_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_6_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85683648)))];
            tensor<fp16, [1, 768, 96]> var_1420_cast_fp16 = conv(dilations = var_1420_dilations_0, groups = var_1420_groups_0, pad = var_1420_pad_0, pad_type = var_1420_pad_type_0, strides = var_1420_strides_0, weight = core_backbone_layers_6_conv_conv_weight_to_fp16, x = win_17_cast_fp16)[name = string("op_1420_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_1422_cast_fp16 = mul(x = c_9_cast_fp16, y = var_1420_cast_fp16)[name = string("op_1422_cast_fp16")];
            tensor<int32, [3]> input_99_perm_0 = const()[name = string("input_99_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_6_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_6_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85688320)))];
            tensor<fp16, [1, 96, 768]> input_99_cast_fp16 = transpose(perm = input_99_perm_0, x = var_1422_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [1, 96, 768]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_6_conv_out_proj_weight_to_fp16, x = input_99_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_73_cast_fp16 = add(x = x_71_cast_fp16, y = linear_35_cast_fp16)[name = string("x_73_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1428_cast_fp16 = abs(x = x_73_cast_fp16)[name = string("op_1428_cast_fp16")];
            tensor<int32, [1]> var_1430_axes_0 = const()[name = string("op_1430_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1430_cast_fp16 = reduce_max(axes = var_1430_axes_0, keep_dims = var_127, x = var_1428_cast_fp16)[name = string("op_1430_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_39_cast_fp16 = maximum(x = var_1430_cast_fp16, y = var_128_to_fp16)[name = string("s_39_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_45_cast_fp16 = real_div(x = x_73_cast_fp16, y = s_39_cast_fp16)[name = string("y_45_cast_fp16")];
            fp16 var_129_promoted_17_to_fp16 = const()[name = string("op_129_promoted_17_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1433_cast_fp16 = pow(x = y_45_cast_fp16, y = var_129_promoted_17_to_fp16)[name = string("op_1433_cast_fp16")];
            tensor<int32, [1]> var_1435_axes_0 = const()[name = string("op_1435_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1435_keep_dims_0 = const()[name = string("op_1435_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1435_cast_fp16 = reduce_mean(axes = var_1435_axes_0, keep_dims = var_1435_keep_dims_0, x = var_1433_cast_fp16)[name = string("op_1435_cast_fp16")];
            fp32 var_1436_epsilon_0 = const()[name = string("op_1436_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1436_cast_fp16 = inverse(epsilon = var_1436_epsilon_0, x = s_39_cast_fp16)[name = string("op_1436_cast_fp16")];
            fp16 var_1437_to_fp16 = const()[name = string("op_1437_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1438_cast_fp16 = mul(x = var_1436_cast_fp16, y = var_1437_to_fp16)[name = string("op_1438_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1439_cast_fp16 = real_div(x = var_1438_cast_fp16, y = s_39_cast_fp16)[name = string("op_1439_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1440_cast_fp16 = add(x = var_1435_cast_fp16, y = var_1439_cast_fp16)[name = string("op_1440_cast_fp16")];
            fp32 var_1441_epsilon_0 = const()[name = string("op_1441_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1441_cast_fp16 = rsqrt(epsilon = var_1441_epsilon_0, x = var_1440_cast_fp16)[name = string("op_1441_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1442_cast_fp16 = mul(x = y_45_cast_fp16, y = var_1441_cast_fp16)[name = string("op_1442_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_6_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_6_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86868032)))];
            tensor<fp16, [1, 96, 768]> input_101_cast_fp16 = mul(x = core_backbone_layers_6_ffn_norm_weight_to_fp16, y = var_1442_cast_fp16)[name = string("input_101_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_6_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86869632)))];
            tensor<fp16, [1, 96, 1792]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w1_weight_to_fp16, x = input_101_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_6_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89622208)))];
            tensor<fp16, [1, 96, 1792]> linear_37_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w3_weight_to_fp16, x = input_101_cast_fp16)[name = string("linear_37_cast_fp16")];
            fp16 var_1451_to_fp16 = const()[name = string("op_1451_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1452_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_1451_to_fp16)[name = string("op_1452_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1455_cast_fp16 = tanh(x = var_1452_cast_fp16)[name = string("op_1455_cast_fp16")];
            fp16 var_1456_to_fp16 = const()[name = string("op_1456_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1457_cast_fp16 = add(x = var_1455_cast_fp16, y = var_1456_to_fp16)[name = string("op_1457_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1458_cast_fp16 = mul(x = var_1452_cast_fp16, y = var_1457_cast_fp16)[name = string("op_1458_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_103_cast_fp16 = mul(x = var_1458_cast_fp16, y = linear_37_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_6_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_6_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92374784)))];
            tensor<fp16, [1, 96, 768]> linear_38_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_6_feed_forward_w2_weight_to_fp16, x = input_103_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_77_cast_fp16 = add(x = x_73_cast_fp16, y = linear_38_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1464_cast_fp16 = abs(x = x_77_cast_fp16)[name = string("op_1464_cast_fp16")];
            tensor<int32, [1]> var_1466_axes_0 = const()[name = string("op_1466_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1466_cast_fp16 = reduce_max(axes = var_1466_axes_0, keep_dims = var_127, x = var_1464_cast_fp16)[name = string("op_1466_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_41_cast_fp16 = maximum(x = var_1466_cast_fp16, y = var_128_to_fp16)[name = string("s_41_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_47_cast_fp16 = real_div(x = x_77_cast_fp16, y = s_41_cast_fp16)[name = string("y_47_cast_fp16")];
            fp16 var_129_promoted_18_to_fp16 = const()[name = string("op_129_promoted_18_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1469_cast_fp16 = pow(x = y_47_cast_fp16, y = var_129_promoted_18_to_fp16)[name = string("op_1469_cast_fp16")];
            tensor<int32, [1]> var_1471_axes_0 = const()[name = string("op_1471_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1471_keep_dims_0 = const()[name = string("op_1471_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1471_cast_fp16 = reduce_mean(axes = var_1471_axes_0, keep_dims = var_1471_keep_dims_0, x = var_1469_cast_fp16)[name = string("op_1471_cast_fp16")];
            fp32 var_1472_epsilon_0 = const()[name = string("op_1472_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1472_cast_fp16 = inverse(epsilon = var_1472_epsilon_0, x = s_41_cast_fp16)[name = string("op_1472_cast_fp16")];
            fp16 var_1473_to_fp16 = const()[name = string("op_1473_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1474_cast_fp16 = mul(x = var_1472_cast_fp16, y = var_1473_to_fp16)[name = string("op_1474_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1475_cast_fp16 = real_div(x = var_1474_cast_fp16, y = s_41_cast_fp16)[name = string("op_1475_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1476_cast_fp16 = add(x = var_1471_cast_fp16, y = var_1475_cast_fp16)[name = string("op_1476_cast_fp16")];
            fp32 var_1477_epsilon_0 = const()[name = string("op_1477_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1477_cast_fp16 = rsqrt(epsilon = var_1477_epsilon_0, x = var_1476_cast_fp16)[name = string("op_1477_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1478_cast_fp16 = mul(x = y_47_cast_fp16, y = var_1477_cast_fp16)[name = string("op_1478_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_7_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95127360)))];
            tensor<fp16, [1, 96, 768]> input_105_cast_fp16 = mul(x = core_backbone_layers_7_operator_norm_weight_to_fp16, y = var_1478_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_7_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95128960)))];
            tensor<fp16, [1, 96, 2304]> linear_39_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_7_conv_in_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [3]> bcx_11_perm_0 = const()[name = string("bcx_11_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_25_begin_0 = const()[name = string("b_25_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_25_end_0 = const()[name = string("b_25_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_25_end_mask_0 = const()[name = string("b_25_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_11_cast_fp16 = transpose(perm = bcx_11_perm_0, x = linear_39_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [1, 768, 96]> b_25_cast_fp16 = slice_by_index(begin = b_25_begin_0, end = b_25_end_0, end_mask = b_25_end_mask_0, x = bcx_11_cast_fp16)[name = string("b_25_cast_fp16")];
            tensor<int32, [3]> c_11_begin_0 = const()[name = string("c_11_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_11_end_0 = const()[name = string("c_11_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_11_end_mask_0 = const()[name = string("c_11_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_11_cast_fp16 = slice_by_index(begin = c_11_begin_0, end = c_11_end_0, end_mask = c_11_end_mask_0, x = bcx_11_cast_fp16)[name = string("c_11_cast_fp16")];
            tensor<int32, [3]> xx_11_begin_0 = const()[name = string("xx_11_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_11_end_0 = const()[name = string("xx_11_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_11_end_mask_0 = const()[name = string("xx_11_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_11_cast_fp16 = slice_by_index(begin = xx_11_begin_0, end = xx_11_end_0, end_mask = xx_11_end_mask_0, x = bcx_11_cast_fp16)[name = string("xx_11_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_19_cast_fp16 = mul(x = b_25_cast_fp16, y = xx_11_cast_fp16)[name = string("new_19_cast_fp16")];
            tensor<int32, [4]> input_109_begin_0 = const()[name = string("input_109_begin_0"), val = tensor<int32, [4]>([5, 0, 0, 0])];
            tensor<int32, [4]> input_109_end_0 = const()[name = string("input_109_end_0"), val = tensor<int32, [4]>([6, 1, 768, 2])];
            tensor<bool, [4]> input_109_end_mask_0 = const()[name = string("input_109_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_109_squeeze_mask_0 = const()[name = string("input_109_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_109_cast_fp16 = slice_by_index(begin = input_109_begin_0, end = input_109_end_0, end_mask = input_109_end_mask_0, squeeze_mask = input_109_squeeze_mask_0, x = conv)[name = string("input_109_cast_fp16")];
            tensor<int32, [6]> var_1493_pad_0 = const()[name = string("op_1493_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1493_mode_0 = const()[name = string("op_1493_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1493_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = var_1493_mode_0, pad = var_1493_pad_0, x = new_19_cast_fp16)[name = string("op_1493_cast_fp16")];
            tensor<int32, [6]> var_1495_pad_0 = const()[name = string("op_1495_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_1495_mode_0 = const()[name = string("op_1495_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1495_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_1495_mode_0, pad = var_1495_pad_0, x = input_109_cast_fp16)[name = string("op_1495_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_21_cast_fp16 = add(x = var_1493_cast_fp16, y = var_1495_cast_fp16)[name = string("win_21_cast_fp16")];
            tensor<int32, [3]> var_1497_begin_0 = const()[name = string("op_1497_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_1497_end_0 = const()[name = string("op_1497_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_1497_end_mask_0 = const()[name = string("op_1497_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1497_cast_fp16 = slice_by_index(begin = var_1497_begin_0, end = var_1497_end_0, end_mask = var_1497_end_mask_0, x = win_21_cast_fp16)[name = string("op_1497_cast_fp16")];
            string var_1503_pad_type_0 = const()[name = string("op_1503_pad_type_0"), val = string("valid")];
            int32 var_1503_groups_0 = const()[name = string("op_1503_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1503_strides_0 = const()[name = string("op_1503_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1503_pad_0 = const()[name = string("op_1503_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1503_dilations_0 = const()[name = string("op_1503_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_7_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98667968)))];
            tensor<fp16, [1, 768, 96]> var_1503_cast_fp16 = conv(dilations = var_1503_dilations_0, groups = var_1503_groups_0, pad = var_1503_pad_0, pad_type = var_1503_pad_type_0, strides = var_1503_strides_0, weight = core_backbone_layers_7_conv_conv_weight_to_fp16, x = win_21_cast_fp16)[name = string("op_1503_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_1505_cast_fp16 = mul(x = c_11_cast_fp16, y = var_1503_cast_fp16)[name = string("op_1505_cast_fp16")];
            tensor<int32, [3]> input_111_perm_0 = const()[name = string("input_111_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_7_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_7_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98672640)))];
            tensor<fp16, [1, 96, 768]> input_111_cast_fp16 = transpose(perm = input_111_perm_0, x = var_1505_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [1, 96, 768]> linear_40_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_conv_out_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_79_cast_fp16 = add(x = x_77_cast_fp16, y = linear_40_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1511_cast_fp16 = abs(x = x_79_cast_fp16)[name = string("op_1511_cast_fp16")];
            tensor<int32, [1]> var_1513_axes_0 = const()[name = string("op_1513_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1513_cast_fp16 = reduce_max(axes = var_1513_axes_0, keep_dims = var_127, x = var_1511_cast_fp16)[name = string("op_1513_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_43_cast_fp16 = maximum(x = var_1513_cast_fp16, y = var_128_to_fp16)[name = string("s_43_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_51_cast_fp16 = real_div(x = x_79_cast_fp16, y = s_43_cast_fp16)[name = string("y_51_cast_fp16")];
            fp16 var_129_promoted_19_to_fp16 = const()[name = string("op_129_promoted_19_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1516_cast_fp16 = pow(x = y_51_cast_fp16, y = var_129_promoted_19_to_fp16)[name = string("op_1516_cast_fp16")];
            tensor<int32, [1]> var_1518_axes_0 = const()[name = string("op_1518_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1518_keep_dims_0 = const()[name = string("op_1518_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1518_cast_fp16 = reduce_mean(axes = var_1518_axes_0, keep_dims = var_1518_keep_dims_0, x = var_1516_cast_fp16)[name = string("op_1518_cast_fp16")];
            fp32 var_1519_epsilon_0 = const()[name = string("op_1519_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1519_cast_fp16 = inverse(epsilon = var_1519_epsilon_0, x = s_43_cast_fp16)[name = string("op_1519_cast_fp16")];
            fp16 var_1520_to_fp16 = const()[name = string("op_1520_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1521_cast_fp16 = mul(x = var_1519_cast_fp16, y = var_1520_to_fp16)[name = string("op_1521_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1522_cast_fp16 = real_div(x = var_1521_cast_fp16, y = s_43_cast_fp16)[name = string("op_1522_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1523_cast_fp16 = add(x = var_1518_cast_fp16, y = var_1522_cast_fp16)[name = string("op_1523_cast_fp16")];
            fp32 var_1524_epsilon_0 = const()[name = string("op_1524_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1524_cast_fp16 = rsqrt(epsilon = var_1524_epsilon_0, x = var_1523_cast_fp16)[name = string("op_1524_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1525_cast_fp16 = mul(x = y_51_cast_fp16, y = var_1524_cast_fp16)[name = string("op_1525_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_7_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_7_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99852352)))];
            tensor<fp16, [1, 96, 768]> input_113_cast_fp16 = mul(x = core_backbone_layers_7_ffn_norm_weight_to_fp16, y = var_1525_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_7_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99853952)))];
            tensor<fp16, [1, 96, 1792]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w1_weight_to_fp16, x = input_113_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_7_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102606528)))];
            tensor<fp16, [1, 96, 1792]> linear_42_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w3_weight_to_fp16, x = input_113_cast_fp16)[name = string("linear_42_cast_fp16")];
            fp16 var_1534_to_fp16 = const()[name = string("op_1534_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1535_cast_fp16 = mul(x = linear_41_cast_fp16, y = var_1534_to_fp16)[name = string("op_1535_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1538_cast_fp16 = tanh(x = var_1535_cast_fp16)[name = string("op_1538_cast_fp16")];
            fp16 var_1539_to_fp16 = const()[name = string("op_1539_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1540_cast_fp16 = add(x = var_1538_cast_fp16, y = var_1539_to_fp16)[name = string("op_1540_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1541_cast_fp16 = mul(x = var_1535_cast_fp16, y = var_1540_cast_fp16)[name = string("op_1541_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_115_cast_fp16 = mul(x = var_1541_cast_fp16, y = linear_42_cast_fp16)[name = string("input_115_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_7_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_7_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105359104)))];
            tensor<fp16, [1, 96, 768]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_7_feed_forward_w2_weight_to_fp16, x = input_115_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_83_cast_fp16 = add(x = x_79_cast_fp16, y = linear_43_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1547_cast_fp16 = abs(x = x_83_cast_fp16)[name = string("op_1547_cast_fp16")];
            tensor<int32, [1]> var_1549_axes_0 = const()[name = string("op_1549_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1549_cast_fp16 = reduce_max(axes = var_1549_axes_0, keep_dims = var_127, x = var_1547_cast_fp16)[name = string("op_1549_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_45_cast_fp16 = maximum(x = var_1549_cast_fp16, y = var_128_to_fp16)[name = string("s_45_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_53_cast_fp16 = real_div(x = x_83_cast_fp16, y = s_45_cast_fp16)[name = string("y_53_cast_fp16")];
            fp16 var_129_promoted_20_to_fp16 = const()[name = string("op_129_promoted_20_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1552_cast_fp16 = pow(x = y_53_cast_fp16, y = var_129_promoted_20_to_fp16)[name = string("op_1552_cast_fp16")];
            tensor<int32, [1]> var_1554_axes_0 = const()[name = string("op_1554_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1554_keep_dims_0 = const()[name = string("op_1554_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1554_cast_fp16 = reduce_mean(axes = var_1554_axes_0, keep_dims = var_1554_keep_dims_0, x = var_1552_cast_fp16)[name = string("op_1554_cast_fp16")];
            fp32 var_1555_epsilon_0 = const()[name = string("op_1555_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1555_cast_fp16 = inverse(epsilon = var_1555_epsilon_0, x = s_45_cast_fp16)[name = string("op_1555_cast_fp16")];
            fp16 var_1556_to_fp16 = const()[name = string("op_1556_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1557_cast_fp16 = mul(x = var_1555_cast_fp16, y = var_1556_to_fp16)[name = string("op_1557_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1558_cast_fp16 = real_div(x = var_1557_cast_fp16, y = s_45_cast_fp16)[name = string("op_1558_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1559_cast_fp16 = add(x = var_1554_cast_fp16, y = var_1558_cast_fp16)[name = string("op_1559_cast_fp16")];
            fp32 var_1560_epsilon_0 = const()[name = string("op_1560_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1560_cast_fp16 = rsqrt(epsilon = var_1560_epsilon_0, x = var_1559_cast_fp16)[name = string("op_1560_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1561_cast_fp16 = mul(x = y_53_cast_fp16, y = var_1560_cast_fp16)[name = string("op_1561_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_8_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108111680)))];
            tensor<fp16, [1, 96, 768]> input_117_cast_fp16 = mul(x = core_backbone_layers_8_operator_norm_weight_to_fp16, y = var_1561_cast_fp16)[name = string("input_117_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108113280)))];
            tensor<fp16, [1, 96, 768]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_q_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<int32, [4]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [4]>([1, 96, 8, 96])];
            tensor<fp16, [1, 96, 8, 96]> x_85_cast_fp16 = reshape(shape = concat_17, x = linear_44_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1568_cast_fp16 = abs(x = x_85_cast_fp16)[name = string("op_1568_cast_fp16")];
            tensor<int32, [1]> var_1570_axes_0 = const()[name = string("op_1570_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 8, 1]> var_1570_cast_fp16 = reduce_max(axes = var_1570_axes_0, keep_dims = var_127, x = var_1568_cast_fp16)[name = string("op_1570_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> s_47_cast_fp16 = maximum(x = var_1570_cast_fp16, y = var_128_to_fp16)[name = string("s_47_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> y_55_cast_fp16 = real_div(x = x_85_cast_fp16, y = s_47_cast_fp16)[name = string("y_55_cast_fp16")];
            fp16 var_129_promoted_21_to_fp16 = const()[name = string("op_129_promoted_21_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 8, 96]> var_1573_cast_fp16 = pow(x = y_55_cast_fp16, y = var_129_promoted_21_to_fp16)[name = string("op_1573_cast_fp16")];
            tensor<int32, [1]> var_1575_axes_0 = const()[name = string("op_1575_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1575_keep_dims_0 = const()[name = string("op_1575_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 8, 1]> var_1575_cast_fp16 = reduce_mean(axes = var_1575_axes_0, keep_dims = var_1575_keep_dims_0, x = var_1573_cast_fp16)[name = string("op_1575_cast_fp16")];
            fp32 var_1576_epsilon_0 = const()[name = string("op_1576_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 8, 1]> var_1576_cast_fp16 = inverse(epsilon = var_1576_epsilon_0, x = s_47_cast_fp16)[name = string("op_1576_cast_fp16")];
            fp16 var_1577_to_fp16 = const()[name = string("op_1577_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 8, 1]> var_1578_cast_fp16 = mul(x = var_1576_cast_fp16, y = var_1577_to_fp16)[name = string("op_1578_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_1579_cast_fp16 = real_div(x = var_1578_cast_fp16, y = s_47_cast_fp16)[name = string("op_1579_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_1580_cast_fp16 = add(x = var_1575_cast_fp16, y = var_1579_cast_fp16)[name = string("op_1580_cast_fp16")];
            fp32 var_1581_epsilon_0 = const()[name = string("op_1581_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 8, 1]> var_1581_cast_fp16 = rsqrt(epsilon = var_1581_epsilon_0, x = var_1580_cast_fp16)[name = string("op_1581_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1582_cast_fp16 = mul(x = y_55_cast_fp16, y = var_1581_cast_fp16)[name = string("op_1582_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_8_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109292992)))];
            tensor<fp16, [1, 96, 8, 96]> var_1583_cast_fp16 = mul(x = core_backbone_layers_8_self_attn_q_norm_weight_to_fp16, y = var_1582_cast_fp16)[name = string("op_1583_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109293248)))];
            tensor<fp16, [1, 96, 384]> linear_45_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_k_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<int32, [4]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> x_87_cast_fp16 = reshape(shape = concat_18, x = linear_45_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_1590_cast_fp16 = abs(x = x_87_cast_fp16)[name = string("op_1590_cast_fp16")];
            tensor<int32, [1]> var_1592_axes_0 = const()[name = string("op_1592_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 4, 1]> var_1592_cast_fp16 = reduce_max(axes = var_1592_axes_0, keep_dims = var_127, x = var_1590_cast_fp16)[name = string("op_1592_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> s_49_cast_fp16 = maximum(x = var_1592_cast_fp16, y = var_128_to_fp16)[name = string("s_49_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> y_57_cast_fp16 = real_div(x = x_87_cast_fp16, y = s_49_cast_fp16)[name = string("y_57_cast_fp16")];
            fp16 var_129_promoted_22_to_fp16 = const()[name = string("op_129_promoted_22_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 4, 96]> var_1595_cast_fp16 = pow(x = y_57_cast_fp16, y = var_129_promoted_22_to_fp16)[name = string("op_1595_cast_fp16")];
            tensor<int32, [1]> var_1597_axes_0 = const()[name = string("op_1597_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1597_keep_dims_0 = const()[name = string("op_1597_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 4, 1]> var_1597_cast_fp16 = reduce_mean(axes = var_1597_axes_0, keep_dims = var_1597_keep_dims_0, x = var_1595_cast_fp16)[name = string("op_1597_cast_fp16")];
            fp32 var_1598_epsilon_0 = const()[name = string("op_1598_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 4, 1]> var_1598_cast_fp16 = inverse(epsilon = var_1598_epsilon_0, x = s_49_cast_fp16)[name = string("op_1598_cast_fp16")];
            fp16 var_1599_to_fp16 = const()[name = string("op_1599_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 4, 1]> var_1600_cast_fp16 = mul(x = var_1598_cast_fp16, y = var_1599_to_fp16)[name = string("op_1600_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_1601_cast_fp16 = real_div(x = var_1600_cast_fp16, y = s_49_cast_fp16)[name = string("op_1601_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_1602_cast_fp16 = add(x = var_1597_cast_fp16, y = var_1601_cast_fp16)[name = string("op_1602_cast_fp16")];
            fp32 var_1603_epsilon_0 = const()[name = string("op_1603_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 4, 1]> var_1603_cast_fp16 = rsqrt(epsilon = var_1603_epsilon_0, x = var_1602_cast_fp16)[name = string("op_1603_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_1604_cast_fp16 = mul(x = y_57_cast_fp16, y = var_1603_cast_fp16)[name = string("op_1604_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_8_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109883136)))];
            tensor<fp16, [1, 96, 4, 96]> var_1605_cast_fp16 = mul(x = core_backbone_layers_8_self_attn_k_norm_weight_to_fp16, y = var_1604_cast_fp16)[name = string("op_1605_cast_fp16")];
            tensor<int32, [4]> new_21_perm_0 = const()[name = string("new_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109883392)))];
            tensor<fp16, [1, 96, 384]> linear_46_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_v_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<int32, [4]> concat_19 = const()[name = string("concat_19"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> var_1610_cast_fp16 = reshape(shape = concat_19, x = linear_46_cast_fp16)[name = string("op_1610_cast_fp16")];
            tensor<int32, [4]> new_23_perm_0 = const()[name = string("new_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_121_begin_0 = const()[name = string("input_121_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_121_end_0 = const()[name = string("input_121_end_0"), val = tensor<int32, [5]>([3, 1, 4, 31, 96])];
            tensor<bool, [5]> input_121_end_mask_0 = const()[name = string("input_121_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_121_squeeze_mask_0 = const()[name = string("input_121_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_121_cast_fp16 = slice_by_index(begin = input_121_begin_0, end = input_121_end_0, end_mask = input_121_end_mask_0, squeeze_mask = input_121_squeeze_mask_0, x = k)[name = string("input_121_cast_fp16")];
            tensor<int32, [8]> var_1615_pad_0 = const()[name = string("op_1615_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1615_mode_0 = const()[name = string("op_1615_mode_0"), val = string("constant")];
            fp16 const_28_to_fp16 = const()[name = string("const_28_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_21_cast_fp16 = transpose(perm = new_21_perm_0, x = var_1605_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [1, 4, 127, 96]> var_1615_cast_fp16 = pad(constant_val = const_28_to_fp16, mode = var_1615_mode_0, pad = var_1615_pad_0, x = new_21_cast_fp16)[name = string("op_1615_cast_fp16")];
            tensor<int32, [8]> var_1617_pad_0 = const()[name = string("op_1617_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_1617_mode_0 = const()[name = string("op_1617_mode_0"), val = string("constant")];
            fp16 const_29_to_fp16 = const()[name = string("const_29_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_1617_cast_fp16 = pad(constant_val = const_29_to_fp16, mode = var_1617_mode_0, pad = var_1617_pad_0, x = input_121_cast_fp16)[name = string("op_1617_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> x_91_cast_fp16 = add(x = var_1615_cast_fp16, y = var_1617_cast_fp16)[name = string("x_91_cast_fp16")];
            tensor<int32, [5]> input_125_begin_0 = const()[name = string("input_125_begin_0"), val = tensor<int32, [5]>([2, 0, 0, 0, 0])];
            tensor<int32, [5]> input_125_end_0 = const()[name = string("input_125_end_0"), val = tensor<int32, [5]>([3, 1, 4, 31, 96])];
            tensor<bool, [5]> input_125_end_mask_0 = const()[name = string("input_125_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_125_squeeze_mask_0 = const()[name = string("input_125_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_125_cast_fp16 = slice_by_index(begin = input_125_begin_0, end = input_125_end_0, end_mask = input_125_end_mask_0, squeeze_mask = input_125_squeeze_mask_0, x = v)[name = string("input_125_cast_fp16")];
            tensor<int32, [8]> var_1622_pad_0 = const()[name = string("op_1622_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1622_mode_0 = const()[name = string("op_1622_mode_0"), val = string("constant")];
            fp16 const_30_to_fp16 = const()[name = string("const_30_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_23_cast_fp16 = transpose(perm = new_23_perm_0, x = var_1610_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [1, 4, 127, 96]> var_1622_cast_fp16 = pad(constant_val = const_30_to_fp16, mode = var_1622_mode_0, pad = var_1622_pad_0, x = new_23_cast_fp16)[name = string("op_1622_cast_fp16")];
            tensor<int32, [8]> var_1624_pad_0 = const()[name = string("op_1624_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_1624_mode_0 = const()[name = string("op_1624_mode_0"), val = string("constant")];
            fp16 const_31_to_fp16 = const()[name = string("const_31_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_1624_cast_fp16 = pad(constant_val = const_31_to_fp16, mode = var_1624_mode_0, pad = var_1624_pad_0, x = input_125_cast_fp16)[name = string("op_1624_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> v_11_cast_fp16 = add(x = var_1622_cast_fp16, y = var_1624_cast_fp16)[name = string("v_11_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1626_cast_fp16 = mul(x = var_1583_cast_fp16, y = const_66_to_fp16)[name = string("op_1626_cast_fp16")];
            tensor<int32, [2]> var_1627_split_sizes_0 = const()[name = string("op_1627_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_75 = const()[name = string("const_75"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 48]> var_1627_cast_fp16_0, tensor<fp16, [1, 96, 8, 48]> var_1627_cast_fp16_1 = split(axis = const_75, split_sizes = var_1627_split_sizes_0, x = var_1583_cast_fp16)[name = string("op_1627_cast_fp16")];
            fp16 const_32_promoted_to_fp16 = const()[name = string("const_32_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 96, 8, 48]> var_1629_cast_fp16 = mul(x = var_1627_cast_fp16_1, y = const_32_promoted_to_fp16)[name = string("op_1629_cast_fp16")];
            bool var_1631_interleave_0 = const()[name = string("op_1631_interleave_0"), val = bool(false)];
            int32 const_76 = const()[name = string("const_76"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 96]> var_1631_cast_fp16 = concat(axis = const_76, interleave = var_1631_interleave_0, values = (var_1629_cast_fp16, var_1627_cast_fp16_0))[name = string("op_1631_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1632_cast_fp16 = mul(x = var_1631_cast_fp16, y = const_69_to_fp16)[name = string("op_1632_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> q_5_cast_fp16 = add(x = var_1626_cast_fp16, y = var_1632_cast_fp16)[name = string("q_5_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_1634_cast_fp16 = mul(x = x_91_cast_fp16, y = const_0_to_fp16)[name = string("op_1634_cast_fp16")];
            tensor<int32, [2]> var_1635_split_sizes_0 = const()[name = string("op_1635_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1635_axis_0 = const()[name = string("op_1635_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 127, 48]> var_1635_cast_fp16_0, tensor<fp16, [1, 4, 127, 48]> var_1635_cast_fp16_1 = split(axis = var_1635_axis_0, split_sizes = var_1635_split_sizes_0, x = x_91_cast_fp16)[name = string("op_1635_cast_fp16")];
            fp16 const_33_promoted_to_fp16 = const()[name = string("const_33_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 127, 48]> var_1637_cast_fp16 = mul(x = var_1635_cast_fp16_1, y = const_33_promoted_to_fp16)[name = string("op_1637_cast_fp16")];
            bool var_1639_interleave_0 = const()[name = string("op_1639_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 127, 96]> var_1639_cast_fp16 = concat(axis = var_126, interleave = var_1639_interleave_0, values = (var_1637_cast_fp16, var_1635_cast_fp16_0))[name = string("op_1639_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_1640_cast_fp16 = mul(x = var_1639_cast_fp16, y = const_1_to_fp16)[name = string("op_1640_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> k_11_cast_fp16 = add(x = var_1634_cast_fp16, y = var_1640_cast_fp16)[name = string("k_11_cast_fp16")];
            tensor<int32, [1]> var_1642_axes_0 = const()[name = string("op_1642_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_1642_cast_fp16 = expand_dims(axes = var_1642_axes_0, x = k_11_cast_fp16)[name = string("op_1642_cast_fp16")];
            tensor<int32, [5]> var_1644_reps_0 = const()[name = string("op_1644_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_1644_cast_fp16 = tile(reps = var_1644_reps_0, x = var_1642_cast_fp16)[name = string("op_1644_cast_fp16")];
            tensor<int32, [4]> concat_21 = const()[name = string("concat_21"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> k_13_cast_fp16 = reshape(shape = concat_21, x = var_1644_cast_fp16)[name = string("k_13_cast_fp16")];
            tensor<int32, [1]> var_1646_axes_0 = const()[name = string("op_1646_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_1646_cast_fp16 = expand_dims(axes = var_1646_axes_0, x = v_11_cast_fp16)[name = string("op_1646_cast_fp16")];
            tensor<int32, [5]> var_1648_reps_0 = const()[name = string("op_1648_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_1648_cast_fp16 = tile(reps = var_1648_reps_0, x = var_1646_cast_fp16)[name = string("op_1648_cast_fp16")];
            tensor<int32, [4]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> v_13_cast_fp16 = reshape(shape = concat_23, x = var_1648_cast_fp16)[name = string("v_13_cast_fp16")];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = string("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1651_transpose_x_1 = const()[name = string("op_1651_transpose_x_1"), val = bool(false)];
            bool var_1651_transpose_y_1 = const()[name = string("op_1651_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 96]> transpose_4_cast_fp16 = transpose(perm = transpose_4_perm_0, x = q_5_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [1, 8, 96, 127]> var_1651_cast_fp16 = matmul(transpose_x = var_1651_transpose_x_1, transpose_y = var_1651_transpose_y_1, x = transpose_4_cast_fp16, y = k_13_cast_fp16)[name = string("op_1651_cast_fp16")];
            fp16 var_1652_to_fp16 = const()[name = string("op_1652_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 96, 127]> s_51_cast_fp16 = mul(x = var_1651_cast_fp16, y = var_1652_to_fp16)[name = string("s_51_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_93_cast_fp16 = add(x = s_51_cast_fp16, y = mask)[name = string("x_93_cast_fp16")];
            tensor<int32, [1]> var_1656_axes_0 = const()[name = string("op_1656_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 96, 1]> var_1656_cast_fp16 = reduce_max(axes = var_1656_axes_0, keep_dims = var_127, x = x_93_cast_fp16)[name = string("op_1656_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_95_cast_fp16 = sub(x = x_93_cast_fp16, y = var_1656_cast_fp16)[name = string("x_95_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_97_cast_fp16 = exp(x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<int32, [1]> var_1660_axes_0 = const()[name = string("op_1660_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1660_keep_dims_0 = const()[name = string("op_1660_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 1]> var_1660_cast_fp16 = reduce_sum(axes = var_1660_axes_0, keep_dims = var_1660_keep_dims_0, x = x_97_cast_fp16)[name = string("op_1660_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> var_1661_cast_fp16 = real_div(x = x_97_cast_fp16, y = var_1660_cast_fp16)[name = string("op_1661_cast_fp16")];
            bool o_9_transpose_x_0 = const()[name = string("o_9_transpose_x_0"), val = bool(false)];
            bool o_9_transpose_y_0 = const()[name = string("o_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 96, 96]> o_9_cast_fp16 = matmul(transpose_x = o_9_transpose_x_0, transpose_y = o_9_transpose_y_0, x = var_1661_cast_fp16, y = v_13_cast_fp16)[name = string("o_9_cast_fp16")];
            tensor<int32, [4]> o_11_perm_0 = const()[name = string("o_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_24 = const()[name = string("concat_24"), val = tensor<int32, [3]>([1, 96, 768])];
            tensor<fp16, [1, 96, 8, 96]> o_11_cast_fp16 = transpose(perm = o_11_perm_0, x = o_9_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [1, 96, 768]> input_127_cast_fp16 = reshape(shape = concat_24, x = o_11_cast_fp16)[name = string("input_127_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110473280)))];
            tensor<fp16, [1, 96, 768]> linear_47_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_self_attn_o_proj_weight_to_fp16, x = input_127_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<int32, [4]> var_1670_begin_0 = const()[name = string("op_1670_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_1670_end_0 = const()[name = string("op_1670_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_1670_end_mask_0 = const()[name = string("op_1670_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1670_cast_fp16 = slice_by_index(begin = var_1670_begin_0, end = var_1670_end_0, end_mask = var_1670_end_mask_0, x = x_91_cast_fp16)[name = string("op_1670_cast_fp16")];
            tensor<int32, [4]> var_1673_begin_0 = const()[name = string("op_1673_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_1673_end_0 = const()[name = string("op_1673_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_1673_end_mask_0 = const()[name = string("op_1673_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_1673_cast_fp16 = slice_by_index(begin = var_1673_begin_0, end = var_1673_end_0, end_mask = var_1673_end_mask_0, x = v_11_cast_fp16)[name = string("op_1673_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_99_cast_fp16 = add(x = x_83_cast_fp16, y = linear_47_cast_fp16)[name = string("x_99_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1676_cast_fp16 = abs(x = x_99_cast_fp16)[name = string("op_1676_cast_fp16")];
            tensor<int32, [1]> var_1678_axes_0 = const()[name = string("op_1678_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1678_cast_fp16 = reduce_max(axes = var_1678_axes_0, keep_dims = var_127, x = var_1676_cast_fp16)[name = string("op_1678_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_53_cast_fp16 = maximum(x = var_1678_cast_fp16, y = var_128_to_fp16)[name = string("s_53_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_59_cast_fp16 = real_div(x = x_99_cast_fp16, y = s_53_cast_fp16)[name = string("y_59_cast_fp16")];
            fp16 var_129_promoted_23_to_fp16 = const()[name = string("op_129_promoted_23_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1681_cast_fp16 = pow(x = y_59_cast_fp16, y = var_129_promoted_23_to_fp16)[name = string("op_1681_cast_fp16")];
            tensor<int32, [1]> var_1683_axes_0 = const()[name = string("op_1683_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1683_keep_dims_0 = const()[name = string("op_1683_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1683_cast_fp16 = reduce_mean(axes = var_1683_axes_0, keep_dims = var_1683_keep_dims_0, x = var_1681_cast_fp16)[name = string("op_1683_cast_fp16")];
            fp32 var_1684_epsilon_0 = const()[name = string("op_1684_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1684_cast_fp16 = inverse(epsilon = var_1684_epsilon_0, x = s_53_cast_fp16)[name = string("op_1684_cast_fp16")];
            fp16 var_1685_to_fp16 = const()[name = string("op_1685_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1686_cast_fp16 = mul(x = var_1684_cast_fp16, y = var_1685_to_fp16)[name = string("op_1686_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1687_cast_fp16 = real_div(x = var_1686_cast_fp16, y = s_53_cast_fp16)[name = string("op_1687_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1688_cast_fp16 = add(x = var_1683_cast_fp16, y = var_1687_cast_fp16)[name = string("op_1688_cast_fp16")];
            fp32 var_1689_epsilon_0 = const()[name = string("op_1689_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1689_cast_fp16 = rsqrt(epsilon = var_1689_epsilon_0, x = var_1688_cast_fp16)[name = string("op_1689_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1690_cast_fp16 = mul(x = y_59_cast_fp16, y = var_1689_cast_fp16)[name = string("op_1690_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_8_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_8_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111652992)))];
            tensor<fp16, [1, 96, 768]> input_129_cast_fp16 = mul(x = core_backbone_layers_8_ffn_norm_weight_to_fp16, y = var_1690_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_8_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111654592)))];
            tensor<fp16, [1, 96, 1792]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w1_weight_to_fp16, x = input_129_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_8_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114407168)))];
            tensor<fp16, [1, 96, 1792]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w3_weight_to_fp16, x = input_129_cast_fp16)[name = string("linear_49_cast_fp16")];
            fp16 var_1699_to_fp16 = const()[name = string("op_1699_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1700_cast_fp16 = mul(x = linear_48_cast_fp16, y = var_1699_to_fp16)[name = string("op_1700_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1703_cast_fp16 = tanh(x = var_1700_cast_fp16)[name = string("op_1703_cast_fp16")];
            fp16 var_1704_to_fp16 = const()[name = string("op_1704_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1705_cast_fp16 = add(x = var_1703_cast_fp16, y = var_1704_to_fp16)[name = string("op_1705_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1706_cast_fp16 = mul(x = var_1700_cast_fp16, y = var_1705_cast_fp16)[name = string("op_1706_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_131_cast_fp16 = mul(x = var_1706_cast_fp16, y = linear_49_cast_fp16)[name = string("input_131_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_8_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_8_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117159744)))];
            tensor<fp16, [1, 96, 768]> linear_50_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_8_feed_forward_w2_weight_to_fp16, x = input_131_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_103_cast_fp16 = add(x = x_99_cast_fp16, y = linear_50_cast_fp16)[name = string("x_103_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1712_cast_fp16 = abs(x = x_103_cast_fp16)[name = string("op_1712_cast_fp16")];
            tensor<int32, [1]> var_1714_axes_0 = const()[name = string("op_1714_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1714_cast_fp16 = reduce_max(axes = var_1714_axes_0, keep_dims = var_127, x = var_1712_cast_fp16)[name = string("op_1714_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_55_cast_fp16 = maximum(x = var_1714_cast_fp16, y = var_128_to_fp16)[name = string("s_55_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_61_cast_fp16 = real_div(x = x_103_cast_fp16, y = s_55_cast_fp16)[name = string("y_61_cast_fp16")];
            fp16 var_129_promoted_24_to_fp16 = const()[name = string("op_129_promoted_24_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1717_cast_fp16 = pow(x = y_61_cast_fp16, y = var_129_promoted_24_to_fp16)[name = string("op_1717_cast_fp16")];
            tensor<int32, [1]> var_1719_axes_0 = const()[name = string("op_1719_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1719_keep_dims_0 = const()[name = string("op_1719_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1719_cast_fp16 = reduce_mean(axes = var_1719_axes_0, keep_dims = var_1719_keep_dims_0, x = var_1717_cast_fp16)[name = string("op_1719_cast_fp16")];
            fp32 var_1720_epsilon_0 = const()[name = string("op_1720_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1720_cast_fp16 = inverse(epsilon = var_1720_epsilon_0, x = s_55_cast_fp16)[name = string("op_1720_cast_fp16")];
            fp16 var_1721_to_fp16 = const()[name = string("op_1721_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1722_cast_fp16 = mul(x = var_1720_cast_fp16, y = var_1721_to_fp16)[name = string("op_1722_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1723_cast_fp16 = real_div(x = var_1722_cast_fp16, y = s_55_cast_fp16)[name = string("op_1723_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1724_cast_fp16 = add(x = var_1719_cast_fp16, y = var_1723_cast_fp16)[name = string("op_1724_cast_fp16")];
            fp32 var_1725_epsilon_0 = const()[name = string("op_1725_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1725_cast_fp16 = rsqrt(epsilon = var_1725_epsilon_0, x = var_1724_cast_fp16)[name = string("op_1725_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1726_cast_fp16 = mul(x = y_61_cast_fp16, y = var_1725_cast_fp16)[name = string("op_1726_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_9_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_9_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119912320)))];
            tensor<fp16, [1, 96, 768]> input_133_cast_fp16 = mul(x = core_backbone_layers_9_operator_norm_weight_to_fp16, y = var_1726_cast_fp16)[name = string("input_133_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_9_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119913920)))];
            tensor<fp16, [1, 96, 2304]> linear_51_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_9_conv_in_proj_weight_to_fp16, x = input_133_cast_fp16)[name = string("linear_51_cast_fp16")];
            tensor<int32, [3]> bcx_13_perm_0 = const()[name = string("bcx_13_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_31_begin_0 = const()[name = string("b_31_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_31_end_0 = const()[name = string("b_31_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_31_end_mask_0 = const()[name = string("b_31_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_13_cast_fp16 = transpose(perm = bcx_13_perm_0, x = linear_51_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 768, 96]> b_31_cast_fp16 = slice_by_index(begin = b_31_begin_0, end = b_31_end_0, end_mask = b_31_end_mask_0, x = bcx_13_cast_fp16)[name = string("b_31_cast_fp16")];
            tensor<int32, [3]> c_13_begin_0 = const()[name = string("c_13_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_13_end_0 = const()[name = string("c_13_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_13_end_mask_0 = const()[name = string("c_13_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_13_cast_fp16 = slice_by_index(begin = c_13_begin_0, end = c_13_end_0, end_mask = c_13_end_mask_0, x = bcx_13_cast_fp16)[name = string("c_13_cast_fp16")];
            tensor<int32, [3]> xx_13_begin_0 = const()[name = string("xx_13_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_13_end_0 = const()[name = string("xx_13_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_13_end_mask_0 = const()[name = string("xx_13_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_13_cast_fp16 = slice_by_index(begin = xx_13_begin_0, end = xx_13_end_0, end_mask = xx_13_end_mask_0, x = bcx_13_cast_fp16)[name = string("xx_13_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_25_cast_fp16 = mul(x = b_31_cast_fp16, y = xx_13_cast_fp16)[name = string("new_25_cast_fp16")];
            tensor<int32, [4]> input_137_begin_0 = const()[name = string("input_137_begin_0"), val = tensor<int32, [4]>([6, 0, 0, 0])];
            tensor<int32, [4]> input_137_end_0 = const()[name = string("input_137_end_0"), val = tensor<int32, [4]>([7, 1, 768, 2])];
            tensor<bool, [4]> input_137_end_mask_0 = const()[name = string("input_137_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_137_squeeze_mask_0 = const()[name = string("input_137_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_137_cast_fp16 = slice_by_index(begin = input_137_begin_0, end = input_137_end_0, end_mask = input_137_end_mask_0, squeeze_mask = input_137_squeeze_mask_0, x = conv)[name = string("input_137_cast_fp16")];
            tensor<int32, [6]> var_1741_pad_0 = const()[name = string("op_1741_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1741_mode_0 = const()[name = string("op_1741_mode_0"), val = string("constant")];
            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1741_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = var_1741_mode_0, pad = var_1741_pad_0, x = new_25_cast_fp16)[name = string("op_1741_cast_fp16")];
            tensor<int32, [6]> var_1743_pad_0 = const()[name = string("op_1743_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_1743_mode_0 = const()[name = string("op_1743_mode_0"), val = string("constant")];
            fp16 const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1743_cast_fp16 = pad(constant_val = const_35_to_fp16, mode = var_1743_mode_0, pad = var_1743_pad_0, x = input_137_cast_fp16)[name = string("op_1743_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_25_cast_fp16 = add(x = var_1741_cast_fp16, y = var_1743_cast_fp16)[name = string("win_25_cast_fp16")];
            tensor<int32, [3]> var_1745_begin_0 = const()[name = string("op_1745_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_1745_end_0 = const()[name = string("op_1745_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_1745_end_mask_0 = const()[name = string("op_1745_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1745_cast_fp16 = slice_by_index(begin = var_1745_begin_0, end = var_1745_end_0, end_mask = var_1745_end_mask_0, x = win_25_cast_fp16)[name = string("op_1745_cast_fp16")];
            string var_1751_pad_type_0 = const()[name = string("op_1751_pad_type_0"), val = string("valid")];
            int32 var_1751_groups_0 = const()[name = string("op_1751_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1751_strides_0 = const()[name = string("op_1751_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1751_pad_0 = const()[name = string("op_1751_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1751_dilations_0 = const()[name = string("op_1751_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_9_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123452928)))];
            tensor<fp16, [1, 768, 96]> var_1751_cast_fp16 = conv(dilations = var_1751_dilations_0, groups = var_1751_groups_0, pad = var_1751_pad_0, pad_type = var_1751_pad_type_0, strides = var_1751_strides_0, weight = core_backbone_layers_9_conv_conv_weight_to_fp16, x = win_25_cast_fp16)[name = string("op_1751_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_1753_cast_fp16 = mul(x = c_13_cast_fp16, y = var_1751_cast_fp16)[name = string("op_1753_cast_fp16")];
            tensor<int32, [3]> input_139_perm_0 = const()[name = string("input_139_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_9_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_9_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123457600)))];
            tensor<fp16, [1, 96, 768]> input_139_cast_fp16 = transpose(perm = input_139_perm_0, x = var_1753_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 96, 768]> linear_52_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_9_conv_out_proj_weight_to_fp16, x = input_139_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_105_cast_fp16 = add(x = x_103_cast_fp16, y = linear_52_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1759_cast_fp16 = abs(x = x_105_cast_fp16)[name = string("op_1759_cast_fp16")];
            tensor<int32, [1]> var_1761_axes_0 = const()[name = string("op_1761_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1761_cast_fp16 = reduce_max(axes = var_1761_axes_0, keep_dims = var_127, x = var_1759_cast_fp16)[name = string("op_1761_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_57_cast_fp16 = maximum(x = var_1761_cast_fp16, y = var_128_to_fp16)[name = string("s_57_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_65_cast_fp16 = real_div(x = x_105_cast_fp16, y = s_57_cast_fp16)[name = string("y_65_cast_fp16")];
            fp16 var_129_promoted_25_to_fp16 = const()[name = string("op_129_promoted_25_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1764_cast_fp16 = pow(x = y_65_cast_fp16, y = var_129_promoted_25_to_fp16)[name = string("op_1764_cast_fp16")];
            tensor<int32, [1]> var_1766_axes_0 = const()[name = string("op_1766_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1766_keep_dims_0 = const()[name = string("op_1766_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1766_cast_fp16 = reduce_mean(axes = var_1766_axes_0, keep_dims = var_1766_keep_dims_0, x = var_1764_cast_fp16)[name = string("op_1766_cast_fp16")];
            fp32 var_1767_epsilon_0 = const()[name = string("op_1767_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1767_cast_fp16 = inverse(epsilon = var_1767_epsilon_0, x = s_57_cast_fp16)[name = string("op_1767_cast_fp16")];
            fp16 var_1768_to_fp16 = const()[name = string("op_1768_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1769_cast_fp16 = mul(x = var_1767_cast_fp16, y = var_1768_to_fp16)[name = string("op_1769_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1770_cast_fp16 = real_div(x = var_1769_cast_fp16, y = s_57_cast_fp16)[name = string("op_1770_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1771_cast_fp16 = add(x = var_1766_cast_fp16, y = var_1770_cast_fp16)[name = string("op_1771_cast_fp16")];
            fp32 var_1772_epsilon_0 = const()[name = string("op_1772_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1772_cast_fp16 = rsqrt(epsilon = var_1772_epsilon_0, x = var_1771_cast_fp16)[name = string("op_1772_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1773_cast_fp16 = mul(x = y_65_cast_fp16, y = var_1772_cast_fp16)[name = string("op_1773_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_9_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_9_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124637312)))];
            tensor<fp16, [1, 96, 768]> input_141_cast_fp16 = mul(x = core_backbone_layers_9_ffn_norm_weight_to_fp16, y = var_1773_cast_fp16)[name = string("input_141_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_9_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124638912)))];
            tensor<fp16, [1, 96, 1792]> linear_53_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w1_weight_to_fp16, x = input_141_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_9_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127391488)))];
            tensor<fp16, [1, 96, 1792]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w3_weight_to_fp16, x = input_141_cast_fp16)[name = string("linear_54_cast_fp16")];
            fp16 var_1782_to_fp16 = const()[name = string("op_1782_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1783_cast_fp16 = mul(x = linear_53_cast_fp16, y = var_1782_to_fp16)[name = string("op_1783_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1786_cast_fp16 = tanh(x = var_1783_cast_fp16)[name = string("op_1786_cast_fp16")];
            fp16 var_1787_to_fp16 = const()[name = string("op_1787_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1788_cast_fp16 = add(x = var_1786_cast_fp16, y = var_1787_to_fp16)[name = string("op_1788_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1789_cast_fp16 = mul(x = var_1783_cast_fp16, y = var_1788_cast_fp16)[name = string("op_1789_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_143_cast_fp16 = mul(x = var_1789_cast_fp16, y = linear_54_cast_fp16)[name = string("input_143_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_9_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_9_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130144064)))];
            tensor<fp16, [1, 96, 768]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_9_feed_forward_w2_weight_to_fp16, x = input_143_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_109_cast_fp16 = add(x = x_105_cast_fp16, y = linear_55_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1795_cast_fp16 = abs(x = x_109_cast_fp16)[name = string("op_1795_cast_fp16")];
            tensor<int32, [1]> var_1797_axes_0 = const()[name = string("op_1797_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1797_cast_fp16 = reduce_max(axes = var_1797_axes_0, keep_dims = var_127, x = var_1795_cast_fp16)[name = string("op_1797_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_59_cast_fp16 = maximum(x = var_1797_cast_fp16, y = var_128_to_fp16)[name = string("s_59_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_67_cast_fp16 = real_div(x = x_109_cast_fp16, y = s_59_cast_fp16)[name = string("y_67_cast_fp16")];
            fp16 var_129_promoted_26_to_fp16 = const()[name = string("op_129_promoted_26_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1800_cast_fp16 = pow(x = y_67_cast_fp16, y = var_129_promoted_26_to_fp16)[name = string("op_1800_cast_fp16")];
            tensor<int32, [1]> var_1802_axes_0 = const()[name = string("op_1802_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1802_keep_dims_0 = const()[name = string("op_1802_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1802_cast_fp16 = reduce_mean(axes = var_1802_axes_0, keep_dims = var_1802_keep_dims_0, x = var_1800_cast_fp16)[name = string("op_1802_cast_fp16")];
            fp32 var_1803_epsilon_0 = const()[name = string("op_1803_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1803_cast_fp16 = inverse(epsilon = var_1803_epsilon_0, x = s_59_cast_fp16)[name = string("op_1803_cast_fp16")];
            fp16 var_1804_to_fp16 = const()[name = string("op_1804_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1805_cast_fp16 = mul(x = var_1803_cast_fp16, y = var_1804_to_fp16)[name = string("op_1805_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1806_cast_fp16 = real_div(x = var_1805_cast_fp16, y = s_59_cast_fp16)[name = string("op_1806_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1807_cast_fp16 = add(x = var_1802_cast_fp16, y = var_1806_cast_fp16)[name = string("op_1807_cast_fp16")];
            fp32 var_1808_epsilon_0 = const()[name = string("op_1808_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1808_cast_fp16 = rsqrt(epsilon = var_1808_epsilon_0, x = var_1807_cast_fp16)[name = string("op_1808_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1809_cast_fp16 = mul(x = y_67_cast_fp16, y = var_1808_cast_fp16)[name = string("op_1809_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_10_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_10_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132896640)))];
            tensor<fp16, [1, 96, 768]> input_145_cast_fp16 = mul(x = core_backbone_layers_10_operator_norm_weight_to_fp16, y = var_1809_cast_fp16)[name = string("input_145_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_10_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132898240)))];
            tensor<fp16, [1, 96, 2304]> linear_56_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_10_conv_in_proj_weight_to_fp16, x = input_145_cast_fp16)[name = string("linear_56_cast_fp16")];
            tensor<int32, [3]> bcx_15_perm_0 = const()[name = string("bcx_15_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_35_begin_0 = const()[name = string("b_35_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_35_end_0 = const()[name = string("b_35_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_35_end_mask_0 = const()[name = string("b_35_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_15_cast_fp16 = transpose(perm = bcx_15_perm_0, x = linear_56_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [1, 768, 96]> b_35_cast_fp16 = slice_by_index(begin = b_35_begin_0, end = b_35_end_0, end_mask = b_35_end_mask_0, x = bcx_15_cast_fp16)[name = string("b_35_cast_fp16")];
            tensor<int32, [3]> c_15_begin_0 = const()[name = string("c_15_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_15_end_0 = const()[name = string("c_15_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_15_end_mask_0 = const()[name = string("c_15_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_15_cast_fp16 = slice_by_index(begin = c_15_begin_0, end = c_15_end_0, end_mask = c_15_end_mask_0, x = bcx_15_cast_fp16)[name = string("c_15_cast_fp16")];
            tensor<int32, [3]> xx_15_begin_0 = const()[name = string("xx_15_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_15_end_0 = const()[name = string("xx_15_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_15_end_mask_0 = const()[name = string("xx_15_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_15_cast_fp16 = slice_by_index(begin = xx_15_begin_0, end = xx_15_end_0, end_mask = xx_15_end_mask_0, x = bcx_15_cast_fp16)[name = string("xx_15_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_27_cast_fp16 = mul(x = b_35_cast_fp16, y = xx_15_cast_fp16)[name = string("new_27_cast_fp16")];
            tensor<int32, [4]> input_149_begin_0 = const()[name = string("input_149_begin_0"), val = tensor<int32, [4]>([7, 0, 0, 0])];
            tensor<int32, [4]> input_149_end_0 = const()[name = string("input_149_end_0"), val = tensor<int32, [4]>([8, 1, 768, 2])];
            tensor<bool, [4]> input_149_end_mask_0 = const()[name = string("input_149_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_149_squeeze_mask_0 = const()[name = string("input_149_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_149_cast_fp16 = slice_by_index(begin = input_149_begin_0, end = input_149_end_0, end_mask = input_149_end_mask_0, squeeze_mask = input_149_squeeze_mask_0, x = conv)[name = string("input_149_cast_fp16")];
            tensor<int32, [6]> var_1824_pad_0 = const()[name = string("op_1824_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_1824_mode_0 = const()[name = string("op_1824_mode_0"), val = string("constant")];
            fp16 const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1824_cast_fp16 = pad(constant_val = const_36_to_fp16, mode = var_1824_mode_0, pad = var_1824_pad_0, x = new_27_cast_fp16)[name = string("op_1824_cast_fp16")];
            tensor<int32, [6]> var_1826_pad_0 = const()[name = string("op_1826_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_1826_mode_0 = const()[name = string("op_1826_mode_0"), val = string("constant")];
            fp16 const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_1826_cast_fp16 = pad(constant_val = const_37_to_fp16, mode = var_1826_mode_0, pad = var_1826_pad_0, x = input_149_cast_fp16)[name = string("op_1826_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_29_cast_fp16 = add(x = var_1824_cast_fp16, y = var_1826_cast_fp16)[name = string("win_29_cast_fp16")];
            tensor<int32, [3]> var_1828_begin_0 = const()[name = string("op_1828_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_1828_end_0 = const()[name = string("op_1828_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_1828_end_mask_0 = const()[name = string("op_1828_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_1828_cast_fp16 = slice_by_index(begin = var_1828_begin_0, end = var_1828_end_0, end_mask = var_1828_end_mask_0, x = win_29_cast_fp16)[name = string("op_1828_cast_fp16")];
            string var_1834_pad_type_0 = const()[name = string("op_1834_pad_type_0"), val = string("valid")];
            int32 var_1834_groups_0 = const()[name = string("op_1834_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_1834_strides_0 = const()[name = string("op_1834_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_1834_pad_0 = const()[name = string("op_1834_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_1834_dilations_0 = const()[name = string("op_1834_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_10_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136437248)))];
            tensor<fp16, [1, 768, 96]> var_1834_cast_fp16 = conv(dilations = var_1834_dilations_0, groups = var_1834_groups_0, pad = var_1834_pad_0, pad_type = var_1834_pad_type_0, strides = var_1834_strides_0, weight = core_backbone_layers_10_conv_conv_weight_to_fp16, x = win_29_cast_fp16)[name = string("op_1834_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_1836_cast_fp16 = mul(x = c_15_cast_fp16, y = var_1834_cast_fp16)[name = string("op_1836_cast_fp16")];
            tensor<int32, [3]> input_151_perm_0 = const()[name = string("input_151_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_10_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_10_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136441920)))];
            tensor<fp16, [1, 96, 768]> input_151_cast_fp16 = transpose(perm = input_151_perm_0, x = var_1836_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 96, 768]> linear_57_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_10_conv_out_proj_weight_to_fp16, x = input_151_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_111_cast_fp16 = add(x = x_109_cast_fp16, y = linear_57_cast_fp16)[name = string("x_111_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1842_cast_fp16 = abs(x = x_111_cast_fp16)[name = string("op_1842_cast_fp16")];
            tensor<int32, [1]> var_1844_axes_0 = const()[name = string("op_1844_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1844_cast_fp16 = reduce_max(axes = var_1844_axes_0, keep_dims = var_127, x = var_1842_cast_fp16)[name = string("op_1844_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_61_cast_fp16 = maximum(x = var_1844_cast_fp16, y = var_128_to_fp16)[name = string("s_61_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_71_cast_fp16 = real_div(x = x_111_cast_fp16, y = s_61_cast_fp16)[name = string("y_71_cast_fp16")];
            fp16 var_129_promoted_27_to_fp16 = const()[name = string("op_129_promoted_27_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1847_cast_fp16 = pow(x = y_71_cast_fp16, y = var_129_promoted_27_to_fp16)[name = string("op_1847_cast_fp16")];
            tensor<int32, [1]> var_1849_axes_0 = const()[name = string("op_1849_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1849_keep_dims_0 = const()[name = string("op_1849_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1849_cast_fp16 = reduce_mean(axes = var_1849_axes_0, keep_dims = var_1849_keep_dims_0, x = var_1847_cast_fp16)[name = string("op_1849_cast_fp16")];
            fp32 var_1850_epsilon_0 = const()[name = string("op_1850_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1850_cast_fp16 = inverse(epsilon = var_1850_epsilon_0, x = s_61_cast_fp16)[name = string("op_1850_cast_fp16")];
            fp16 var_1851_to_fp16 = const()[name = string("op_1851_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1852_cast_fp16 = mul(x = var_1850_cast_fp16, y = var_1851_to_fp16)[name = string("op_1852_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1853_cast_fp16 = real_div(x = var_1852_cast_fp16, y = s_61_cast_fp16)[name = string("op_1853_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1854_cast_fp16 = add(x = var_1849_cast_fp16, y = var_1853_cast_fp16)[name = string("op_1854_cast_fp16")];
            fp32 var_1855_epsilon_0 = const()[name = string("op_1855_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1855_cast_fp16 = rsqrt(epsilon = var_1855_epsilon_0, x = var_1854_cast_fp16)[name = string("op_1855_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1856_cast_fp16 = mul(x = y_71_cast_fp16, y = var_1855_cast_fp16)[name = string("op_1856_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_10_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_10_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137621632)))];
            tensor<fp16, [1, 96, 768]> input_153_cast_fp16 = mul(x = core_backbone_layers_10_ffn_norm_weight_to_fp16, y = var_1856_cast_fp16)[name = string("input_153_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_10_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137623232)))];
            tensor<fp16, [1, 96, 1792]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w1_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_58_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_10_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140375808)))];
            tensor<fp16, [1, 96, 1792]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w3_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_59_cast_fp16")];
            fp16 var_1865_to_fp16 = const()[name = string("op_1865_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_1866_cast_fp16 = mul(x = linear_58_cast_fp16, y = var_1865_to_fp16)[name = string("op_1866_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1869_cast_fp16 = tanh(x = var_1866_cast_fp16)[name = string("op_1869_cast_fp16")];
            fp16 var_1870_to_fp16 = const()[name = string("op_1870_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_1871_cast_fp16 = add(x = var_1869_cast_fp16, y = var_1870_to_fp16)[name = string("op_1871_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_1872_cast_fp16 = mul(x = var_1866_cast_fp16, y = var_1871_cast_fp16)[name = string("op_1872_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_155_cast_fp16 = mul(x = var_1872_cast_fp16, y = linear_59_cast_fp16)[name = string("input_155_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_10_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_10_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143128384)))];
            tensor<fp16, [1, 96, 768]> linear_60_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_10_feed_forward_w2_weight_to_fp16, x = input_155_cast_fp16)[name = string("linear_60_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_115_cast_fp16 = add(x = x_111_cast_fp16, y = linear_60_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1878_cast_fp16 = abs(x = x_115_cast_fp16)[name = string("op_1878_cast_fp16")];
            tensor<int32, [1]> var_1880_axes_0 = const()[name = string("op_1880_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_1880_cast_fp16 = reduce_max(axes = var_1880_axes_0, keep_dims = var_127, x = var_1878_cast_fp16)[name = string("op_1880_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_63_cast_fp16 = maximum(x = var_1880_cast_fp16, y = var_128_to_fp16)[name = string("s_63_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_73_cast_fp16 = real_div(x = x_115_cast_fp16, y = s_63_cast_fp16)[name = string("y_73_cast_fp16")];
            fp16 var_129_promoted_28_to_fp16 = const()[name = string("op_129_promoted_28_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_1883_cast_fp16 = pow(x = y_73_cast_fp16, y = var_129_promoted_28_to_fp16)[name = string("op_1883_cast_fp16")];
            tensor<int32, [1]> var_1885_axes_0 = const()[name = string("op_1885_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1885_keep_dims_0 = const()[name = string("op_1885_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_1885_cast_fp16 = reduce_mean(axes = var_1885_axes_0, keep_dims = var_1885_keep_dims_0, x = var_1883_cast_fp16)[name = string("op_1885_cast_fp16")];
            fp32 var_1886_epsilon_0 = const()[name = string("op_1886_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_1886_cast_fp16 = inverse(epsilon = var_1886_epsilon_0, x = s_63_cast_fp16)[name = string("op_1886_cast_fp16")];
            fp16 var_1887_to_fp16 = const()[name = string("op_1887_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_1888_cast_fp16 = mul(x = var_1886_cast_fp16, y = var_1887_to_fp16)[name = string("op_1888_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1889_cast_fp16 = real_div(x = var_1888_cast_fp16, y = s_63_cast_fp16)[name = string("op_1889_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_1890_cast_fp16 = add(x = var_1885_cast_fp16, y = var_1889_cast_fp16)[name = string("op_1890_cast_fp16")];
            fp32 var_1891_epsilon_0 = const()[name = string("op_1891_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_1891_cast_fp16 = rsqrt(epsilon = var_1891_epsilon_0, x = var_1890_cast_fp16)[name = string("op_1891_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_1892_cast_fp16 = mul(x = y_73_cast_fp16, y = var_1891_cast_fp16)[name = string("op_1892_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_11_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145880960)))];
            tensor<fp16, [1, 96, 768]> input_157_cast_fp16 = mul(x = core_backbone_layers_11_operator_norm_weight_to_fp16, y = var_1892_cast_fp16)[name = string("input_157_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145882560)))];
            tensor<fp16, [1, 96, 768]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_q_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<int32, [4]> concat_25 = const()[name = string("concat_25"), val = tensor<int32, [4]>([1, 96, 8, 96])];
            tensor<fp16, [1, 96, 8, 96]> x_117_cast_fp16 = reshape(shape = concat_25, x = linear_61_cast_fp16)[name = string("x_117_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1899_cast_fp16 = abs(x = x_117_cast_fp16)[name = string("op_1899_cast_fp16")];
            tensor<int32, [1]> var_1901_axes_0 = const()[name = string("op_1901_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 8, 1]> var_1901_cast_fp16 = reduce_max(axes = var_1901_axes_0, keep_dims = var_127, x = var_1899_cast_fp16)[name = string("op_1901_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> s_65_cast_fp16 = maximum(x = var_1901_cast_fp16, y = var_128_to_fp16)[name = string("s_65_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> y_75_cast_fp16 = real_div(x = x_117_cast_fp16, y = s_65_cast_fp16)[name = string("y_75_cast_fp16")];
            fp16 var_129_promoted_29_to_fp16 = const()[name = string("op_129_promoted_29_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 8, 96]> var_1904_cast_fp16 = pow(x = y_75_cast_fp16, y = var_129_promoted_29_to_fp16)[name = string("op_1904_cast_fp16")];
            tensor<int32, [1]> var_1906_axes_0 = const()[name = string("op_1906_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1906_keep_dims_0 = const()[name = string("op_1906_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 8, 1]> var_1906_cast_fp16 = reduce_mean(axes = var_1906_axes_0, keep_dims = var_1906_keep_dims_0, x = var_1904_cast_fp16)[name = string("op_1906_cast_fp16")];
            fp32 var_1907_epsilon_0 = const()[name = string("op_1907_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 8, 1]> var_1907_cast_fp16 = inverse(epsilon = var_1907_epsilon_0, x = s_65_cast_fp16)[name = string("op_1907_cast_fp16")];
            fp16 var_1908_to_fp16 = const()[name = string("op_1908_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 8, 1]> var_1909_cast_fp16 = mul(x = var_1907_cast_fp16, y = var_1908_to_fp16)[name = string("op_1909_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_1910_cast_fp16 = real_div(x = var_1909_cast_fp16, y = s_65_cast_fp16)[name = string("op_1910_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_1911_cast_fp16 = add(x = var_1906_cast_fp16, y = var_1910_cast_fp16)[name = string("op_1911_cast_fp16")];
            fp32 var_1912_epsilon_0 = const()[name = string("op_1912_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 8, 1]> var_1912_cast_fp16 = rsqrt(epsilon = var_1912_epsilon_0, x = var_1911_cast_fp16)[name = string("op_1912_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1913_cast_fp16 = mul(x = y_75_cast_fp16, y = var_1912_cast_fp16)[name = string("op_1913_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_11_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147062272)))];
            tensor<fp16, [1, 96, 8, 96]> var_1914_cast_fp16 = mul(x = core_backbone_layers_11_self_attn_q_norm_weight_to_fp16, y = var_1913_cast_fp16)[name = string("op_1914_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147062528)))];
            tensor<fp16, [1, 96, 384]> linear_62_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_k_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_62_cast_fp16")];
            tensor<int32, [4]> concat_26 = const()[name = string("concat_26"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> x_119_cast_fp16 = reshape(shape = concat_26, x = linear_62_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_1921_cast_fp16 = abs(x = x_119_cast_fp16)[name = string("op_1921_cast_fp16")];
            tensor<int32, [1]> var_1923_axes_0 = const()[name = string("op_1923_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 4, 1]> var_1923_cast_fp16 = reduce_max(axes = var_1923_axes_0, keep_dims = var_127, x = var_1921_cast_fp16)[name = string("op_1923_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> s_67_cast_fp16 = maximum(x = var_1923_cast_fp16, y = var_128_to_fp16)[name = string("s_67_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> y_77_cast_fp16 = real_div(x = x_119_cast_fp16, y = s_67_cast_fp16)[name = string("y_77_cast_fp16")];
            fp16 var_129_promoted_30_to_fp16 = const()[name = string("op_129_promoted_30_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 4, 96]> var_1926_cast_fp16 = pow(x = y_77_cast_fp16, y = var_129_promoted_30_to_fp16)[name = string("op_1926_cast_fp16")];
            tensor<int32, [1]> var_1928_axes_0 = const()[name = string("op_1928_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1928_keep_dims_0 = const()[name = string("op_1928_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 4, 1]> var_1928_cast_fp16 = reduce_mean(axes = var_1928_axes_0, keep_dims = var_1928_keep_dims_0, x = var_1926_cast_fp16)[name = string("op_1928_cast_fp16")];
            fp32 var_1929_epsilon_0 = const()[name = string("op_1929_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 4, 1]> var_1929_cast_fp16 = inverse(epsilon = var_1929_epsilon_0, x = s_67_cast_fp16)[name = string("op_1929_cast_fp16")];
            fp16 var_1930_to_fp16 = const()[name = string("op_1930_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 4, 1]> var_1931_cast_fp16 = mul(x = var_1929_cast_fp16, y = var_1930_to_fp16)[name = string("op_1931_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_1932_cast_fp16 = real_div(x = var_1931_cast_fp16, y = s_67_cast_fp16)[name = string("op_1932_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_1933_cast_fp16 = add(x = var_1928_cast_fp16, y = var_1932_cast_fp16)[name = string("op_1933_cast_fp16")];
            fp32 var_1934_epsilon_0 = const()[name = string("op_1934_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 4, 1]> var_1934_cast_fp16 = rsqrt(epsilon = var_1934_epsilon_0, x = var_1933_cast_fp16)[name = string("op_1934_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_1935_cast_fp16 = mul(x = y_77_cast_fp16, y = var_1934_cast_fp16)[name = string("op_1935_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_11_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147652416)))];
            tensor<fp16, [1, 96, 4, 96]> var_1936_cast_fp16 = mul(x = core_backbone_layers_11_self_attn_k_norm_weight_to_fp16, y = var_1935_cast_fp16)[name = string("op_1936_cast_fp16")];
            tensor<int32, [4]> new_29_perm_0 = const()[name = string("new_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147652672)))];
            tensor<fp16, [1, 96, 384]> linear_63_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_v_proj_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_63_cast_fp16")];
            tensor<int32, [4]> concat_27 = const()[name = string("concat_27"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> var_1941_cast_fp16 = reshape(shape = concat_27, x = linear_63_cast_fp16)[name = string("op_1941_cast_fp16")];
            tensor<int32, [4]> new_31_perm_0 = const()[name = string("new_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_161_begin_0 = const()[name = string("input_161_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])];
            tensor<int32, [5]> input_161_end_0 = const()[name = string("input_161_end_0"), val = tensor<int32, [5]>([4, 1, 4, 31, 96])];
            tensor<bool, [5]> input_161_end_mask_0 = const()[name = string("input_161_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_161_squeeze_mask_0 = const()[name = string("input_161_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_161_cast_fp16 = slice_by_index(begin = input_161_begin_0, end = input_161_end_0, end_mask = input_161_end_mask_0, squeeze_mask = input_161_squeeze_mask_0, x = k)[name = string("input_161_cast_fp16")];
            tensor<int32, [8]> var_1946_pad_0 = const()[name = string("op_1946_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1946_mode_0 = const()[name = string("op_1946_mode_0"), val = string("constant")];
            fp16 const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_29_cast_fp16 = transpose(perm = new_29_perm_0, x = var_1936_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [1, 4, 127, 96]> var_1946_cast_fp16 = pad(constant_val = const_38_to_fp16, mode = var_1946_mode_0, pad = var_1946_pad_0, x = new_29_cast_fp16)[name = string("op_1946_cast_fp16")];
            tensor<int32, [8]> var_1948_pad_0 = const()[name = string("op_1948_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_1948_mode_0 = const()[name = string("op_1948_mode_0"), val = string("constant")];
            fp16 const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_1948_cast_fp16 = pad(constant_val = const_39_to_fp16, mode = var_1948_mode_0, pad = var_1948_pad_0, x = input_161_cast_fp16)[name = string("op_1948_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> x_123_cast_fp16 = add(x = var_1946_cast_fp16, y = var_1948_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<int32, [5]> input_165_begin_0 = const()[name = string("input_165_begin_0"), val = tensor<int32, [5]>([3, 0, 0, 0, 0])];
            tensor<int32, [5]> input_165_end_0 = const()[name = string("input_165_end_0"), val = tensor<int32, [5]>([4, 1, 4, 31, 96])];
            tensor<bool, [5]> input_165_end_mask_0 = const()[name = string("input_165_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_165_squeeze_mask_0 = const()[name = string("input_165_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_165_cast_fp16 = slice_by_index(begin = input_165_begin_0, end = input_165_end_0, end_mask = input_165_end_mask_0, squeeze_mask = input_165_squeeze_mask_0, x = v)[name = string("input_165_cast_fp16")];
            tensor<int32, [8]> var_1953_pad_0 = const()[name = string("op_1953_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_1953_mode_0 = const()[name = string("op_1953_mode_0"), val = string("constant")];
            fp16 const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_31_cast_fp16 = transpose(perm = new_31_perm_0, x = var_1941_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [1, 4, 127, 96]> var_1953_cast_fp16 = pad(constant_val = const_40_to_fp16, mode = var_1953_mode_0, pad = var_1953_pad_0, x = new_31_cast_fp16)[name = string("op_1953_cast_fp16")];
            tensor<int32, [8]> var_1955_pad_0 = const()[name = string("op_1955_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_1955_mode_0 = const()[name = string("op_1955_mode_0"), val = string("constant")];
            fp16 const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_1955_cast_fp16 = pad(constant_val = const_41_to_fp16, mode = var_1955_mode_0, pad = var_1955_pad_0, x = input_165_cast_fp16)[name = string("op_1955_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> v_15_cast_fp16 = add(x = var_1953_cast_fp16, y = var_1955_cast_fp16)[name = string("v_15_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1957_cast_fp16 = mul(x = var_1914_cast_fp16, y = const_66_to_fp16)[name = string("op_1957_cast_fp16")];
            tensor<int32, [2]> var_1958_split_sizes_0 = const()[name = string("op_1958_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_79 = const()[name = string("const_79"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 48]> var_1958_cast_fp16_0, tensor<fp16, [1, 96, 8, 48]> var_1958_cast_fp16_1 = split(axis = const_79, split_sizes = var_1958_split_sizes_0, x = var_1914_cast_fp16)[name = string("op_1958_cast_fp16")];
            fp16 const_42_promoted_to_fp16 = const()[name = string("const_42_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 96, 8, 48]> var_1960_cast_fp16 = mul(x = var_1958_cast_fp16_1, y = const_42_promoted_to_fp16)[name = string("op_1960_cast_fp16")];
            bool var_1962_interleave_0 = const()[name = string("op_1962_interleave_0"), val = bool(false)];
            int32 const_80 = const()[name = string("const_80"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 96]> var_1962_cast_fp16 = concat(axis = const_80, interleave = var_1962_interleave_0, values = (var_1960_cast_fp16, var_1958_cast_fp16_0))[name = string("op_1962_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_1963_cast_fp16 = mul(x = var_1962_cast_fp16, y = const_69_to_fp16)[name = string("op_1963_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> q_7_cast_fp16 = add(x = var_1957_cast_fp16, y = var_1963_cast_fp16)[name = string("q_7_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_1965_cast_fp16 = mul(x = x_123_cast_fp16, y = const_0_to_fp16)[name = string("op_1965_cast_fp16")];
            tensor<int32, [2]> var_1966_split_sizes_0 = const()[name = string("op_1966_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_1966_axis_0 = const()[name = string("op_1966_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 127, 48]> var_1966_cast_fp16_0, tensor<fp16, [1, 4, 127, 48]> var_1966_cast_fp16_1 = split(axis = var_1966_axis_0, split_sizes = var_1966_split_sizes_0, x = x_123_cast_fp16)[name = string("op_1966_cast_fp16")];
            fp16 const_43_promoted_to_fp16 = const()[name = string("const_43_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 127, 48]> var_1968_cast_fp16 = mul(x = var_1966_cast_fp16_1, y = const_43_promoted_to_fp16)[name = string("op_1968_cast_fp16")];
            bool var_1970_interleave_0 = const()[name = string("op_1970_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 127, 96]> var_1970_cast_fp16 = concat(axis = var_126, interleave = var_1970_interleave_0, values = (var_1968_cast_fp16, var_1966_cast_fp16_0))[name = string("op_1970_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_1971_cast_fp16 = mul(x = var_1970_cast_fp16, y = const_1_to_fp16)[name = string("op_1971_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> k_15_cast_fp16 = add(x = var_1965_cast_fp16, y = var_1971_cast_fp16)[name = string("k_15_cast_fp16")];
            tensor<int32, [1]> var_1973_axes_0 = const()[name = string("op_1973_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_1973_cast_fp16 = expand_dims(axes = var_1973_axes_0, x = k_15_cast_fp16)[name = string("op_1973_cast_fp16")];
            tensor<int32, [5]> var_1975_reps_0 = const()[name = string("op_1975_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_1975_cast_fp16 = tile(reps = var_1975_reps_0, x = var_1973_cast_fp16)[name = string("op_1975_cast_fp16")];
            tensor<int32, [4]> concat_29 = const()[name = string("concat_29"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> k_17_cast_fp16 = reshape(shape = concat_29, x = var_1975_cast_fp16)[name = string("k_17_cast_fp16")];
            tensor<int32, [1]> var_1977_axes_0 = const()[name = string("op_1977_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_1977_cast_fp16 = expand_dims(axes = var_1977_axes_0, x = v_15_cast_fp16)[name = string("op_1977_cast_fp16")];
            tensor<int32, [5]> var_1979_reps_0 = const()[name = string("op_1979_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_1979_cast_fp16 = tile(reps = var_1979_reps_0, x = var_1977_cast_fp16)[name = string("op_1979_cast_fp16")];
            tensor<int32, [4]> concat_31 = const()[name = string("concat_31"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> v_17_cast_fp16 = reshape(shape = concat_31, x = var_1979_cast_fp16)[name = string("v_17_cast_fp16")];
            tensor<int32, [4]> transpose_6_perm_0 = const()[name = string("transpose_6_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_1982_transpose_x_1 = const()[name = string("op_1982_transpose_x_1"), val = bool(false)];
            bool var_1982_transpose_y_1 = const()[name = string("op_1982_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 96]> transpose_6_cast_fp16 = transpose(perm = transpose_6_perm_0, x = q_7_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 8, 96, 127]> var_1982_cast_fp16 = matmul(transpose_x = var_1982_transpose_x_1, transpose_y = var_1982_transpose_y_1, x = transpose_6_cast_fp16, y = k_17_cast_fp16)[name = string("op_1982_cast_fp16")];
            fp16 var_1983_to_fp16 = const()[name = string("op_1983_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 96, 127]> s_69_cast_fp16 = mul(x = var_1982_cast_fp16, y = var_1983_to_fp16)[name = string("s_69_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_125_cast_fp16 = add(x = s_69_cast_fp16, y = mask)[name = string("x_125_cast_fp16")];
            tensor<int32, [1]> var_1987_axes_0 = const()[name = string("op_1987_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 96, 1]> var_1987_cast_fp16 = reduce_max(axes = var_1987_axes_0, keep_dims = var_127, x = x_125_cast_fp16)[name = string("op_1987_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_127_cast_fp16 = sub(x = x_125_cast_fp16, y = var_1987_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_129_cast_fp16 = exp(x = x_127_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<int32, [1]> var_1991_axes_0 = const()[name = string("op_1991_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_1991_keep_dims_0 = const()[name = string("op_1991_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 1]> var_1991_cast_fp16 = reduce_sum(axes = var_1991_axes_0, keep_dims = var_1991_keep_dims_0, x = x_129_cast_fp16)[name = string("op_1991_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> var_1992_cast_fp16 = real_div(x = x_129_cast_fp16, y = var_1991_cast_fp16)[name = string("op_1992_cast_fp16")];
            bool o_13_transpose_x_0 = const()[name = string("o_13_transpose_x_0"), val = bool(false)];
            bool o_13_transpose_y_0 = const()[name = string("o_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 96, 96]> o_13_cast_fp16 = matmul(transpose_x = o_13_transpose_x_0, transpose_y = o_13_transpose_y_0, x = var_1992_cast_fp16, y = v_17_cast_fp16)[name = string("o_13_cast_fp16")];
            tensor<int32, [4]> o_15_perm_0 = const()[name = string("o_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_32 = const()[name = string("concat_32"), val = tensor<int32, [3]>([1, 96, 768])];
            tensor<fp16, [1, 96, 8, 96]> o_15_cast_fp16 = transpose(perm = o_15_perm_0, x = o_13_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 96, 768]> input_167_cast_fp16 = reshape(shape = concat_32, x = o_15_cast_fp16)[name = string("input_167_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148242560)))];
            tensor<fp16, [1, 96, 768]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_self_attn_o_proj_weight_to_fp16, x = input_167_cast_fp16)[name = string("linear_64_cast_fp16")];
            tensor<int32, [4]> var_2001_begin_0 = const()[name = string("op_2001_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_2001_end_0 = const()[name = string("op_2001_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_2001_end_mask_0 = const()[name = string("op_2001_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2001_cast_fp16 = slice_by_index(begin = var_2001_begin_0, end = var_2001_end_0, end_mask = var_2001_end_mask_0, x = x_123_cast_fp16)[name = string("op_2001_cast_fp16")];
            tensor<int32, [4]> var_2004_begin_0 = const()[name = string("op_2004_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_2004_end_0 = const()[name = string("op_2004_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_2004_end_mask_0 = const()[name = string("op_2004_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2004_cast_fp16 = slice_by_index(begin = var_2004_begin_0, end = var_2004_end_0, end_mask = var_2004_end_mask_0, x = v_15_cast_fp16)[name = string("op_2004_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_131_cast_fp16 = add(x = x_115_cast_fp16, y = linear_64_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2007_cast_fp16 = abs(x = x_131_cast_fp16)[name = string("op_2007_cast_fp16")];
            tensor<int32, [1]> var_2009_axes_0 = const()[name = string("op_2009_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2009_cast_fp16 = reduce_max(axes = var_2009_axes_0, keep_dims = var_127, x = var_2007_cast_fp16)[name = string("op_2009_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_71_cast_fp16 = maximum(x = var_2009_cast_fp16, y = var_128_to_fp16)[name = string("s_71_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_79_cast_fp16 = real_div(x = x_131_cast_fp16, y = s_71_cast_fp16)[name = string("y_79_cast_fp16")];
            fp16 var_129_promoted_31_to_fp16 = const()[name = string("op_129_promoted_31_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2012_cast_fp16 = pow(x = y_79_cast_fp16, y = var_129_promoted_31_to_fp16)[name = string("op_2012_cast_fp16")];
            tensor<int32, [1]> var_2014_axes_0 = const()[name = string("op_2014_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2014_keep_dims_0 = const()[name = string("op_2014_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2014_cast_fp16 = reduce_mean(axes = var_2014_axes_0, keep_dims = var_2014_keep_dims_0, x = var_2012_cast_fp16)[name = string("op_2014_cast_fp16")];
            fp32 var_2015_epsilon_0 = const()[name = string("op_2015_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2015_cast_fp16 = inverse(epsilon = var_2015_epsilon_0, x = s_71_cast_fp16)[name = string("op_2015_cast_fp16")];
            fp16 var_2016_to_fp16 = const()[name = string("op_2016_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2017_cast_fp16 = mul(x = var_2015_cast_fp16, y = var_2016_to_fp16)[name = string("op_2017_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2018_cast_fp16 = real_div(x = var_2017_cast_fp16, y = s_71_cast_fp16)[name = string("op_2018_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2019_cast_fp16 = add(x = var_2014_cast_fp16, y = var_2018_cast_fp16)[name = string("op_2019_cast_fp16")];
            fp32 var_2020_epsilon_0 = const()[name = string("op_2020_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2020_cast_fp16 = rsqrt(epsilon = var_2020_epsilon_0, x = var_2019_cast_fp16)[name = string("op_2020_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2021_cast_fp16 = mul(x = y_79_cast_fp16, y = var_2020_cast_fp16)[name = string("op_2021_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_11_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_11_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149422272)))];
            tensor<fp16, [1, 96, 768]> input_169_cast_fp16 = mul(x = core_backbone_layers_11_ffn_norm_weight_to_fp16, y = var_2021_cast_fp16)[name = string("input_169_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_11_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149423872)))];
            tensor<fp16, [1, 96, 1792]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w1_weight_to_fp16, x = input_169_cast_fp16)[name = string("linear_65_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_11_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152176448)))];
            tensor<fp16, [1, 96, 1792]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w3_weight_to_fp16, x = input_169_cast_fp16)[name = string("linear_66_cast_fp16")];
            fp16 var_2030_to_fp16 = const()[name = string("op_2030_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_2031_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_2030_to_fp16)[name = string("op_2031_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2034_cast_fp16 = tanh(x = var_2031_cast_fp16)[name = string("op_2034_cast_fp16")];
            fp16 var_2035_to_fp16 = const()[name = string("op_2035_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_2036_cast_fp16 = add(x = var_2034_cast_fp16, y = var_2035_to_fp16)[name = string("op_2036_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2037_cast_fp16 = mul(x = var_2031_cast_fp16, y = var_2036_cast_fp16)[name = string("op_2037_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_171_cast_fp16 = mul(x = var_2037_cast_fp16, y = linear_66_cast_fp16)[name = string("input_171_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_11_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_11_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154929024)))];
            tensor<fp16, [1, 96, 768]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_11_feed_forward_w2_weight_to_fp16, x = input_171_cast_fp16)[name = string("linear_67_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_135_cast_fp16 = add(x = x_131_cast_fp16, y = linear_67_cast_fp16)[name = string("x_135_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2043_cast_fp16 = abs(x = x_135_cast_fp16)[name = string("op_2043_cast_fp16")];
            tensor<int32, [1]> var_2045_axes_0 = const()[name = string("op_2045_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2045_cast_fp16 = reduce_max(axes = var_2045_axes_0, keep_dims = var_127, x = var_2043_cast_fp16)[name = string("op_2045_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_73_cast_fp16 = maximum(x = var_2045_cast_fp16, y = var_128_to_fp16)[name = string("s_73_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_81_cast_fp16 = real_div(x = x_135_cast_fp16, y = s_73_cast_fp16)[name = string("y_81_cast_fp16")];
            fp16 var_129_promoted_32_to_fp16 = const()[name = string("op_129_promoted_32_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2048_cast_fp16 = pow(x = y_81_cast_fp16, y = var_129_promoted_32_to_fp16)[name = string("op_2048_cast_fp16")];
            tensor<int32, [1]> var_2050_axes_0 = const()[name = string("op_2050_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2050_keep_dims_0 = const()[name = string("op_2050_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2050_cast_fp16 = reduce_mean(axes = var_2050_axes_0, keep_dims = var_2050_keep_dims_0, x = var_2048_cast_fp16)[name = string("op_2050_cast_fp16")];
            fp32 var_2051_epsilon_0 = const()[name = string("op_2051_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2051_cast_fp16 = inverse(epsilon = var_2051_epsilon_0, x = s_73_cast_fp16)[name = string("op_2051_cast_fp16")];
            fp16 var_2052_to_fp16 = const()[name = string("op_2052_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2053_cast_fp16 = mul(x = var_2051_cast_fp16, y = var_2052_to_fp16)[name = string("op_2053_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2054_cast_fp16 = real_div(x = var_2053_cast_fp16, y = s_73_cast_fp16)[name = string("op_2054_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2055_cast_fp16 = add(x = var_2050_cast_fp16, y = var_2054_cast_fp16)[name = string("op_2055_cast_fp16")];
            fp32 var_2056_epsilon_0 = const()[name = string("op_2056_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2056_cast_fp16 = rsqrt(epsilon = var_2056_epsilon_0, x = var_2055_cast_fp16)[name = string("op_2056_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2057_cast_fp16 = mul(x = y_81_cast_fp16, y = var_2056_cast_fp16)[name = string("op_2057_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_12_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_12_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157681600)))];
            tensor<fp16, [1, 96, 768]> input_173_cast_fp16 = mul(x = core_backbone_layers_12_operator_norm_weight_to_fp16, y = var_2057_cast_fp16)[name = string("input_173_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_12_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157683200)))];
            tensor<fp16, [1, 96, 2304]> linear_68_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_12_conv_in_proj_weight_to_fp16, x = input_173_cast_fp16)[name = string("linear_68_cast_fp16")];
            tensor<int32, [3]> bcx_17_perm_0 = const()[name = string("bcx_17_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_41_begin_0 = const()[name = string("b_41_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_41_end_0 = const()[name = string("b_41_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_41_end_mask_0 = const()[name = string("b_41_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_17_cast_fp16 = transpose(perm = bcx_17_perm_0, x = linear_68_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 768, 96]> b_41_cast_fp16 = slice_by_index(begin = b_41_begin_0, end = b_41_end_0, end_mask = b_41_end_mask_0, x = bcx_17_cast_fp16)[name = string("b_41_cast_fp16")];
            tensor<int32, [3]> c_17_begin_0 = const()[name = string("c_17_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_17_end_0 = const()[name = string("c_17_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_17_end_mask_0 = const()[name = string("c_17_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_17_cast_fp16 = slice_by_index(begin = c_17_begin_0, end = c_17_end_0, end_mask = c_17_end_mask_0, x = bcx_17_cast_fp16)[name = string("c_17_cast_fp16")];
            tensor<int32, [3]> xx_17_begin_0 = const()[name = string("xx_17_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_17_end_0 = const()[name = string("xx_17_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_17_end_mask_0 = const()[name = string("xx_17_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_17_cast_fp16 = slice_by_index(begin = xx_17_begin_0, end = xx_17_end_0, end_mask = xx_17_end_mask_0, x = bcx_17_cast_fp16)[name = string("xx_17_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_33_cast_fp16 = mul(x = b_41_cast_fp16, y = xx_17_cast_fp16)[name = string("new_33_cast_fp16")];
            tensor<int32, [4]> input_177_begin_0 = const()[name = string("input_177_begin_0"), val = tensor<int32, [4]>([8, 0, 0, 0])];
            tensor<int32, [4]> input_177_end_0 = const()[name = string("input_177_end_0"), val = tensor<int32, [4]>([9, 1, 768, 2])];
            tensor<bool, [4]> input_177_end_mask_0 = const()[name = string("input_177_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_177_squeeze_mask_0 = const()[name = string("input_177_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_177_cast_fp16 = slice_by_index(begin = input_177_begin_0, end = input_177_end_0, end_mask = input_177_end_mask_0, squeeze_mask = input_177_squeeze_mask_0, x = conv)[name = string("input_177_cast_fp16")];
            tensor<int32, [6]> var_2072_pad_0 = const()[name = string("op_2072_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_2072_mode_0 = const()[name = string("op_2072_mode_0"), val = string("constant")];
            fp16 const_44_to_fp16 = const()[name = string("const_44_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_2072_cast_fp16 = pad(constant_val = const_44_to_fp16, mode = var_2072_mode_0, pad = var_2072_pad_0, x = new_33_cast_fp16)[name = string("op_2072_cast_fp16")];
            tensor<int32, [6]> var_2074_pad_0 = const()[name = string("op_2074_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_2074_mode_0 = const()[name = string("op_2074_mode_0"), val = string("constant")];
            fp16 const_45_to_fp16 = const()[name = string("const_45_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_2074_cast_fp16 = pad(constant_val = const_45_to_fp16, mode = var_2074_mode_0, pad = var_2074_pad_0, x = input_177_cast_fp16)[name = string("op_2074_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_33_cast_fp16 = add(x = var_2072_cast_fp16, y = var_2074_cast_fp16)[name = string("win_33_cast_fp16")];
            tensor<int32, [3]> var_2076_begin_0 = const()[name = string("op_2076_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_2076_end_0 = const()[name = string("op_2076_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_2076_end_mask_0 = const()[name = string("op_2076_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_2076_cast_fp16 = slice_by_index(begin = var_2076_begin_0, end = var_2076_end_0, end_mask = var_2076_end_mask_0, x = win_33_cast_fp16)[name = string("op_2076_cast_fp16")];
            string var_2082_pad_type_0 = const()[name = string("op_2082_pad_type_0"), val = string("valid")];
            int32 var_2082_groups_0 = const()[name = string("op_2082_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_2082_strides_0 = const()[name = string("op_2082_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2082_pad_0 = const()[name = string("op_2082_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2082_dilations_0 = const()[name = string("op_2082_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_12_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161222208)))];
            tensor<fp16, [1, 768, 96]> var_2082_cast_fp16 = conv(dilations = var_2082_dilations_0, groups = var_2082_groups_0, pad = var_2082_pad_0, pad_type = var_2082_pad_type_0, strides = var_2082_strides_0, weight = core_backbone_layers_12_conv_conv_weight_to_fp16, x = win_33_cast_fp16)[name = string("op_2082_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_2084_cast_fp16 = mul(x = c_17_cast_fp16, y = var_2082_cast_fp16)[name = string("op_2084_cast_fp16")];
            tensor<int32, [3]> input_179_perm_0 = const()[name = string("input_179_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_12_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_12_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161226880)))];
            tensor<fp16, [1, 96, 768]> input_179_cast_fp16 = transpose(perm = input_179_perm_0, x = var_2084_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [1, 96, 768]> linear_69_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_12_conv_out_proj_weight_to_fp16, x = input_179_cast_fp16)[name = string("linear_69_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_137_cast_fp16 = add(x = x_135_cast_fp16, y = linear_69_cast_fp16)[name = string("x_137_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2090_cast_fp16 = abs(x = x_137_cast_fp16)[name = string("op_2090_cast_fp16")];
            tensor<int32, [1]> var_2092_axes_0 = const()[name = string("op_2092_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2092_cast_fp16 = reduce_max(axes = var_2092_axes_0, keep_dims = var_127, x = var_2090_cast_fp16)[name = string("op_2092_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_75_cast_fp16 = maximum(x = var_2092_cast_fp16, y = var_128_to_fp16)[name = string("s_75_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_85_cast_fp16 = real_div(x = x_137_cast_fp16, y = s_75_cast_fp16)[name = string("y_85_cast_fp16")];
            fp16 var_129_promoted_33_to_fp16 = const()[name = string("op_129_promoted_33_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2095_cast_fp16 = pow(x = y_85_cast_fp16, y = var_129_promoted_33_to_fp16)[name = string("op_2095_cast_fp16")];
            tensor<int32, [1]> var_2097_axes_0 = const()[name = string("op_2097_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2097_keep_dims_0 = const()[name = string("op_2097_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2097_cast_fp16 = reduce_mean(axes = var_2097_axes_0, keep_dims = var_2097_keep_dims_0, x = var_2095_cast_fp16)[name = string("op_2097_cast_fp16")];
            fp32 var_2098_epsilon_0 = const()[name = string("op_2098_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2098_cast_fp16 = inverse(epsilon = var_2098_epsilon_0, x = s_75_cast_fp16)[name = string("op_2098_cast_fp16")];
            fp16 var_2099_to_fp16 = const()[name = string("op_2099_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2100_cast_fp16 = mul(x = var_2098_cast_fp16, y = var_2099_to_fp16)[name = string("op_2100_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2101_cast_fp16 = real_div(x = var_2100_cast_fp16, y = s_75_cast_fp16)[name = string("op_2101_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2102_cast_fp16 = add(x = var_2097_cast_fp16, y = var_2101_cast_fp16)[name = string("op_2102_cast_fp16")];
            fp32 var_2103_epsilon_0 = const()[name = string("op_2103_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2103_cast_fp16 = rsqrt(epsilon = var_2103_epsilon_0, x = var_2102_cast_fp16)[name = string("op_2103_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2104_cast_fp16 = mul(x = y_85_cast_fp16, y = var_2103_cast_fp16)[name = string("op_2104_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_12_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_12_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162406592)))];
            tensor<fp16, [1, 96, 768]> input_181_cast_fp16 = mul(x = core_backbone_layers_12_ffn_norm_weight_to_fp16, y = var_2104_cast_fp16)[name = string("input_181_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_12_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162408192)))];
            tensor<fp16, [1, 96, 1792]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w1_weight_to_fp16, x = input_181_cast_fp16)[name = string("linear_70_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_12_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165160768)))];
            tensor<fp16, [1, 96, 1792]> linear_71_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w3_weight_to_fp16, x = input_181_cast_fp16)[name = string("linear_71_cast_fp16")];
            fp16 var_2113_to_fp16 = const()[name = string("op_2113_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_2114_cast_fp16 = mul(x = linear_70_cast_fp16, y = var_2113_to_fp16)[name = string("op_2114_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2117_cast_fp16 = tanh(x = var_2114_cast_fp16)[name = string("op_2117_cast_fp16")];
            fp16 var_2118_to_fp16 = const()[name = string("op_2118_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_2119_cast_fp16 = add(x = var_2117_cast_fp16, y = var_2118_to_fp16)[name = string("op_2119_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2120_cast_fp16 = mul(x = var_2114_cast_fp16, y = var_2119_cast_fp16)[name = string("op_2120_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_183_cast_fp16 = mul(x = var_2120_cast_fp16, y = linear_71_cast_fp16)[name = string("input_183_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_12_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_12_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167913344)))];
            tensor<fp16, [1, 96, 768]> linear_72_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_12_feed_forward_w2_weight_to_fp16, x = input_183_cast_fp16)[name = string("linear_72_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_141_cast_fp16 = add(x = x_137_cast_fp16, y = linear_72_cast_fp16)[name = string("x_141_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2126_cast_fp16 = abs(x = x_141_cast_fp16)[name = string("op_2126_cast_fp16")];
            tensor<int32, [1]> var_2128_axes_0 = const()[name = string("op_2128_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2128_cast_fp16 = reduce_max(axes = var_2128_axes_0, keep_dims = var_127, x = var_2126_cast_fp16)[name = string("op_2128_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_77_cast_fp16 = maximum(x = var_2128_cast_fp16, y = var_128_to_fp16)[name = string("s_77_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_87_cast_fp16 = real_div(x = x_141_cast_fp16, y = s_77_cast_fp16)[name = string("y_87_cast_fp16")];
            fp16 var_129_promoted_34_to_fp16 = const()[name = string("op_129_promoted_34_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2131_cast_fp16 = pow(x = y_87_cast_fp16, y = var_129_promoted_34_to_fp16)[name = string("op_2131_cast_fp16")];
            tensor<int32, [1]> var_2133_axes_0 = const()[name = string("op_2133_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2133_keep_dims_0 = const()[name = string("op_2133_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2133_cast_fp16 = reduce_mean(axes = var_2133_axes_0, keep_dims = var_2133_keep_dims_0, x = var_2131_cast_fp16)[name = string("op_2133_cast_fp16")];
            fp32 var_2134_epsilon_0 = const()[name = string("op_2134_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2134_cast_fp16 = inverse(epsilon = var_2134_epsilon_0, x = s_77_cast_fp16)[name = string("op_2134_cast_fp16")];
            fp16 var_2135_to_fp16 = const()[name = string("op_2135_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2136_cast_fp16 = mul(x = var_2134_cast_fp16, y = var_2135_to_fp16)[name = string("op_2136_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2137_cast_fp16 = real_div(x = var_2136_cast_fp16, y = s_77_cast_fp16)[name = string("op_2137_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2138_cast_fp16 = add(x = var_2133_cast_fp16, y = var_2137_cast_fp16)[name = string("op_2138_cast_fp16")];
            fp32 var_2139_epsilon_0 = const()[name = string("op_2139_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2139_cast_fp16 = rsqrt(epsilon = var_2139_epsilon_0, x = var_2138_cast_fp16)[name = string("op_2139_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2140_cast_fp16 = mul(x = y_87_cast_fp16, y = var_2139_cast_fp16)[name = string("op_2140_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_13_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_13_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170665920)))];
            tensor<fp16, [1, 96, 768]> input_185_cast_fp16 = mul(x = core_backbone_layers_13_operator_norm_weight_to_fp16, y = var_2140_cast_fp16)[name = string("input_185_cast_fp16")];
            tensor<fp16, [2304, 768]> core_backbone_layers_13_conv_in_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_in_proj_weight_to_fp16"), val = tensor<fp16, [2304, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170667520)))];
            tensor<fp16, [1, 96, 2304]> linear_73_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = core_backbone_layers_13_conv_in_proj_weight_to_fp16, x = input_185_cast_fp16)[name = string("linear_73_cast_fp16")];
            tensor<int32, [3]> bcx_perm_0 = const()[name = string("bcx_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [3]> b_45_begin_0 = const()[name = string("b_45_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> b_45_end_0 = const()[name = string("b_45_end_0"), val = tensor<int32, [3]>([1, 768, 96])];
            tensor<bool, [3]> b_45_end_mask_0 = const()[name = string("b_45_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 2304, 96]> bcx_cast_fp16 = transpose(perm = bcx_perm_0, x = linear_73_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 768, 96]> b_45_cast_fp16 = slice_by_index(begin = b_45_begin_0, end = b_45_end_0, end_mask = b_45_end_mask_0, x = bcx_cast_fp16)[name = string("b_45_cast_fp16")];
            tensor<int32, [3]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [3]>([0, 768, 0])];
            tensor<int32, [3]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [3]>([1, 1536, 96])];
            tensor<bool, [3]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [3]>([true, false, true])];
            tensor<fp16, [1, 768, 96]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = bcx_cast_fp16)[name = string("c_cast_fp16")];
            tensor<int32, [3]> xx_begin_0 = const()[name = string("xx_begin_0"), val = tensor<int32, [3]>([0, 1536, 0])];
            tensor<int32, [3]> xx_end_0 = const()[name = string("xx_end_0"), val = tensor<int32, [3]>([1, 2304, 96])];
            tensor<bool, [3]> xx_end_mask_0 = const()[name = string("xx_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 96]> xx_cast_fp16 = slice_by_index(begin = xx_begin_0, end = xx_end_0, end_mask = xx_end_mask_0, x = bcx_cast_fp16)[name = string("xx_cast_fp16")];
            tensor<fp16, [1, 768, 96]> new_35_cast_fp16 = mul(x = b_45_cast_fp16, y = xx_cast_fp16)[name = string("new_35_cast_fp16")];
            tensor<int32, [4]> input_189_begin_0 = const()[name = string("input_189_begin_0"), val = tensor<int32, [4]>([9, 0, 0, 0])];
            tensor<int32, [4]> input_189_end_0 = const()[name = string("input_189_end_0"), val = tensor<int32, [4]>([10, 1, 768, 2])];
            tensor<bool, [4]> input_189_end_mask_0 = const()[name = string("input_189_end_mask_0"), val = tensor<bool, [4]>([false, true, true, true])];
            tensor<bool, [4]> input_189_squeeze_mask_0 = const()[name = string("input_189_squeeze_mask_0"), val = tensor<bool, [4]>([true, false, false, false])];
            tensor<fp16, [1, 768, 2]> input_189_cast_fp16 = slice_by_index(begin = input_189_begin_0, end = input_189_end_0, end_mask = input_189_end_mask_0, squeeze_mask = input_189_squeeze_mask_0, x = conv)[name = string("input_189_cast_fp16")];
            tensor<int32, [6]> var_2155_pad_0 = const()[name = string("op_2155_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 2, 0])];
            string var_2155_mode_0 = const()[name = string("op_2155_mode_0"), val = string("constant")];
            fp16 const_46_to_fp16 = const()[name = string("const_46_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_2155_cast_fp16 = pad(constant_val = const_46_to_fp16, mode = var_2155_mode_0, pad = var_2155_pad_0, x = new_35_cast_fp16)[name = string("op_2155_cast_fp16")];
            tensor<int32, [6]> var_2157_pad_0 = const()[name = string("op_2157_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 0, 96])];
            string var_2157_mode_0 = const()[name = string("op_2157_mode_0"), val = string("constant")];
            fp16 const_47_to_fp16 = const()[name = string("const_47_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 768, 98]> var_2157_cast_fp16 = pad(constant_val = const_47_to_fp16, mode = var_2157_mode_0, pad = var_2157_pad_0, x = input_189_cast_fp16)[name = string("op_2157_cast_fp16")];
            tensor<fp16, [1, 768, 98]> win_37_cast_fp16 = add(x = var_2155_cast_fp16, y = var_2157_cast_fp16)[name = string("win_37_cast_fp16")];
            tensor<int32, [3]> var_2159_begin_0 = const()[name = string("op_2159_begin_0"), val = tensor<int32, [3]>([0, 0, 96])];
            tensor<int32, [3]> var_2159_end_0 = const()[name = string("op_2159_end_0"), val = tensor<int32, [3]>([1, 768, 98])];
            tensor<bool, [3]> var_2159_end_mask_0 = const()[name = string("op_2159_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 768, 2]> var_2159_cast_fp16 = slice_by_index(begin = var_2159_begin_0, end = var_2159_end_0, end_mask = var_2159_end_mask_0, x = win_37_cast_fp16)[name = string("op_2159_cast_fp16")];
            string var_2165_pad_type_0 = const()[name = string("op_2165_pad_type_0"), val = string("valid")];
            int32 var_2165_groups_0 = const()[name = string("op_2165_groups_0"), val = int32(768)];
            tensor<int32, [1]> var_2165_strides_0 = const()[name = string("op_2165_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> var_2165_pad_0 = const()[name = string("op_2165_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> var_2165_dilations_0 = const()[name = string("op_2165_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 3]> core_backbone_layers_13_conv_conv_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_conv_weight_to_fp16"), val = tensor<fp16, [768, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174206528)))];
            tensor<fp16, [1, 768, 96]> var_2165_cast_fp16 = conv(dilations = var_2165_dilations_0, groups = var_2165_groups_0, pad = var_2165_pad_0, pad_type = var_2165_pad_type_0, strides = var_2165_strides_0, weight = core_backbone_layers_13_conv_conv_weight_to_fp16, x = win_37_cast_fp16)[name = string("op_2165_cast_fp16")];
            tensor<fp16, [1, 768, 96]> var_2167_cast_fp16 = mul(x = c_cast_fp16, y = var_2165_cast_fp16)[name = string("op_2167_cast_fp16")];
            tensor<int32, [3]> input_191_perm_0 = const()[name = string("input_191_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [768, 768]> core_backbone_layers_13_conv_out_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_13_conv_out_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174211200)))];
            tensor<fp16, [1, 96, 768]> input_191_cast_fp16 = transpose(perm = input_191_perm_0, x = var_2167_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 96, 768]> linear_74_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_13_conv_out_proj_weight_to_fp16, x = input_191_cast_fp16)[name = string("linear_74_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_143_cast_fp16 = add(x = x_141_cast_fp16, y = linear_74_cast_fp16)[name = string("x_143_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2173_cast_fp16 = abs(x = x_143_cast_fp16)[name = string("op_2173_cast_fp16")];
            tensor<int32, [1]> var_2175_axes_0 = const()[name = string("op_2175_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2175_cast_fp16 = reduce_max(axes = var_2175_axes_0, keep_dims = var_127, x = var_2173_cast_fp16)[name = string("op_2175_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_79_cast_fp16 = maximum(x = var_2175_cast_fp16, y = var_128_to_fp16)[name = string("s_79_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_91_cast_fp16 = real_div(x = x_143_cast_fp16, y = s_79_cast_fp16)[name = string("y_91_cast_fp16")];
            fp16 var_129_promoted_35_to_fp16 = const()[name = string("op_129_promoted_35_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2178_cast_fp16 = pow(x = y_91_cast_fp16, y = var_129_promoted_35_to_fp16)[name = string("op_2178_cast_fp16")];
            tensor<int32, [1]> var_2180_axes_0 = const()[name = string("op_2180_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2180_keep_dims_0 = const()[name = string("op_2180_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2180_cast_fp16 = reduce_mean(axes = var_2180_axes_0, keep_dims = var_2180_keep_dims_0, x = var_2178_cast_fp16)[name = string("op_2180_cast_fp16")];
            fp32 var_2181_epsilon_0 = const()[name = string("op_2181_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2181_cast_fp16 = inverse(epsilon = var_2181_epsilon_0, x = s_79_cast_fp16)[name = string("op_2181_cast_fp16")];
            fp16 var_2182_to_fp16 = const()[name = string("op_2182_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2183_cast_fp16 = mul(x = var_2181_cast_fp16, y = var_2182_to_fp16)[name = string("op_2183_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2184_cast_fp16 = real_div(x = var_2183_cast_fp16, y = s_79_cast_fp16)[name = string("op_2184_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2185_cast_fp16 = add(x = var_2180_cast_fp16, y = var_2184_cast_fp16)[name = string("op_2185_cast_fp16")];
            fp32 var_2186_epsilon_0 = const()[name = string("op_2186_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2186_cast_fp16 = rsqrt(epsilon = var_2186_epsilon_0, x = var_2185_cast_fp16)[name = string("op_2186_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2187_cast_fp16 = mul(x = y_91_cast_fp16, y = var_2186_cast_fp16)[name = string("op_2187_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_13_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_13_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175390912)))];
            tensor<fp16, [1, 96, 768]> input_193_cast_fp16 = mul(x = core_backbone_layers_13_ffn_norm_weight_to_fp16, y = var_2187_cast_fp16)[name = string("input_193_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_13_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175392512)))];
            tensor<fp16, [1, 96, 1792]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w1_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_75_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_13_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178145088)))];
            tensor<fp16, [1, 96, 1792]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w3_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_76_cast_fp16")];
            fp16 var_2196_to_fp16 = const()[name = string("op_2196_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_2197_cast_fp16 = mul(x = linear_75_cast_fp16, y = var_2196_to_fp16)[name = string("op_2197_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2200_cast_fp16 = tanh(x = var_2197_cast_fp16)[name = string("op_2200_cast_fp16")];
            fp16 var_2201_to_fp16 = const()[name = string("op_2201_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_2202_cast_fp16 = add(x = var_2200_cast_fp16, y = var_2201_to_fp16)[name = string("op_2202_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2203_cast_fp16 = mul(x = var_2197_cast_fp16, y = var_2202_cast_fp16)[name = string("op_2203_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_195_cast_fp16 = mul(x = var_2203_cast_fp16, y = linear_76_cast_fp16)[name = string("input_195_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_13_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_13_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180897664)))];
            tensor<fp16, [1, 96, 768]> linear_77_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_13_feed_forward_w2_weight_to_fp16, x = input_195_cast_fp16)[name = string("linear_77_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_147_cast_fp16 = add(x = x_143_cast_fp16, y = linear_77_cast_fp16)[name = string("x_147_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2209_cast_fp16 = abs(x = x_147_cast_fp16)[name = string("op_2209_cast_fp16")];
            tensor<int32, [1]> var_2211_axes_0 = const()[name = string("op_2211_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2211_cast_fp16 = reduce_max(axes = var_2211_axes_0, keep_dims = var_127, x = var_2209_cast_fp16)[name = string("op_2211_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_81_cast_fp16 = maximum(x = var_2211_cast_fp16, y = var_128_to_fp16)[name = string("s_81_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_93_cast_fp16 = real_div(x = x_147_cast_fp16, y = s_81_cast_fp16)[name = string("y_93_cast_fp16")];
            fp16 var_129_promoted_36_to_fp16 = const()[name = string("op_129_promoted_36_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2214_cast_fp16 = pow(x = y_93_cast_fp16, y = var_129_promoted_36_to_fp16)[name = string("op_2214_cast_fp16")];
            tensor<int32, [1]> var_2216_axes_0 = const()[name = string("op_2216_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2216_keep_dims_0 = const()[name = string("op_2216_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2216_cast_fp16 = reduce_mean(axes = var_2216_axes_0, keep_dims = var_2216_keep_dims_0, x = var_2214_cast_fp16)[name = string("op_2216_cast_fp16")];
            fp32 var_2217_epsilon_0 = const()[name = string("op_2217_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2217_cast_fp16 = inverse(epsilon = var_2217_epsilon_0, x = s_81_cast_fp16)[name = string("op_2217_cast_fp16")];
            fp16 var_2218_to_fp16 = const()[name = string("op_2218_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2219_cast_fp16 = mul(x = var_2217_cast_fp16, y = var_2218_to_fp16)[name = string("op_2219_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2220_cast_fp16 = real_div(x = var_2219_cast_fp16, y = s_81_cast_fp16)[name = string("op_2220_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2221_cast_fp16 = add(x = var_2216_cast_fp16, y = var_2220_cast_fp16)[name = string("op_2221_cast_fp16")];
            fp32 var_2222_epsilon_0 = const()[name = string("op_2222_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2222_cast_fp16 = rsqrt(epsilon = var_2222_epsilon_0, x = var_2221_cast_fp16)[name = string("op_2222_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2223_cast_fp16 = mul(x = y_93_cast_fp16, y = var_2222_cast_fp16)[name = string("op_2223_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_14_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183650240)))];
            tensor<fp16, [1, 96, 768]> input_197_cast_fp16 = mul(x = core_backbone_layers_14_operator_norm_weight_to_fp16, y = var_2223_cast_fp16)[name = string("input_197_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183651840)))];
            tensor<fp16, [1, 96, 768]> linear_78_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_q_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_78_cast_fp16")];
            tensor<int32, [4]> concat_33 = const()[name = string("concat_33"), val = tensor<int32, [4]>([1, 96, 8, 96])];
            tensor<fp16, [1, 96, 8, 96]> x_149_cast_fp16 = reshape(shape = concat_33, x = linear_78_cast_fp16)[name = string("x_149_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2230_cast_fp16 = abs(x = x_149_cast_fp16)[name = string("op_2230_cast_fp16")];
            tensor<int32, [1]> var_2232_axes_0 = const()[name = string("op_2232_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 8, 1]> var_2232_cast_fp16 = reduce_max(axes = var_2232_axes_0, keep_dims = var_127, x = var_2230_cast_fp16)[name = string("op_2232_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> s_83_cast_fp16 = maximum(x = var_2232_cast_fp16, y = var_128_to_fp16)[name = string("s_83_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> y_95_cast_fp16 = real_div(x = x_149_cast_fp16, y = s_83_cast_fp16)[name = string("y_95_cast_fp16")];
            fp16 var_129_promoted_37_to_fp16 = const()[name = string("op_129_promoted_37_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 8, 96]> var_2235_cast_fp16 = pow(x = y_95_cast_fp16, y = var_129_promoted_37_to_fp16)[name = string("op_2235_cast_fp16")];
            tensor<int32, [1]> var_2237_axes_0 = const()[name = string("op_2237_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2237_keep_dims_0 = const()[name = string("op_2237_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 8, 1]> var_2237_cast_fp16 = reduce_mean(axes = var_2237_axes_0, keep_dims = var_2237_keep_dims_0, x = var_2235_cast_fp16)[name = string("op_2237_cast_fp16")];
            fp32 var_2238_epsilon_0 = const()[name = string("op_2238_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 8, 1]> var_2238_cast_fp16 = inverse(epsilon = var_2238_epsilon_0, x = s_83_cast_fp16)[name = string("op_2238_cast_fp16")];
            fp16 var_2239_to_fp16 = const()[name = string("op_2239_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 8, 1]> var_2240_cast_fp16 = mul(x = var_2238_cast_fp16, y = var_2239_to_fp16)[name = string("op_2240_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_2241_cast_fp16 = real_div(x = var_2240_cast_fp16, y = s_83_cast_fp16)[name = string("op_2241_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_2242_cast_fp16 = add(x = var_2237_cast_fp16, y = var_2241_cast_fp16)[name = string("op_2242_cast_fp16")];
            fp32 var_2243_epsilon_0 = const()[name = string("op_2243_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 8, 1]> var_2243_cast_fp16 = rsqrt(epsilon = var_2243_epsilon_0, x = var_2242_cast_fp16)[name = string("op_2243_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2244_cast_fp16 = mul(x = y_95_cast_fp16, y = var_2243_cast_fp16)[name = string("op_2244_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_14_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184831552)))];
            tensor<fp16, [1, 96, 8, 96]> var_2245_cast_fp16 = mul(x = core_backbone_layers_14_self_attn_q_norm_weight_to_fp16, y = var_2244_cast_fp16)[name = string("op_2245_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184831808)))];
            tensor<fp16, [1, 96, 384]> linear_79_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_k_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_79_cast_fp16")];
            tensor<int32, [4]> concat_34 = const()[name = string("concat_34"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> x_151_cast_fp16 = reshape(shape = concat_34, x = linear_79_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_2252_cast_fp16 = abs(x = x_151_cast_fp16)[name = string("op_2252_cast_fp16")];
            tensor<int32, [1]> var_2254_axes_0 = const()[name = string("op_2254_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 4, 1]> var_2254_cast_fp16 = reduce_max(axes = var_2254_axes_0, keep_dims = var_127, x = var_2252_cast_fp16)[name = string("op_2254_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> s_85_cast_fp16 = maximum(x = var_2254_cast_fp16, y = var_128_to_fp16)[name = string("s_85_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> y_97_cast_fp16 = real_div(x = x_151_cast_fp16, y = s_85_cast_fp16)[name = string("y_97_cast_fp16")];
            fp16 var_129_promoted_38_to_fp16 = const()[name = string("op_129_promoted_38_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 4, 96]> var_2257_cast_fp16 = pow(x = y_97_cast_fp16, y = var_129_promoted_38_to_fp16)[name = string("op_2257_cast_fp16")];
            tensor<int32, [1]> var_2259_axes_0 = const()[name = string("op_2259_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2259_keep_dims_0 = const()[name = string("op_2259_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 4, 1]> var_2259_cast_fp16 = reduce_mean(axes = var_2259_axes_0, keep_dims = var_2259_keep_dims_0, x = var_2257_cast_fp16)[name = string("op_2259_cast_fp16")];
            fp32 var_2260_epsilon_0 = const()[name = string("op_2260_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 4, 1]> var_2260_cast_fp16 = inverse(epsilon = var_2260_epsilon_0, x = s_85_cast_fp16)[name = string("op_2260_cast_fp16")];
            fp16 var_2261_to_fp16 = const()[name = string("op_2261_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 4, 1]> var_2262_cast_fp16 = mul(x = var_2260_cast_fp16, y = var_2261_to_fp16)[name = string("op_2262_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_2263_cast_fp16 = real_div(x = var_2262_cast_fp16, y = s_85_cast_fp16)[name = string("op_2263_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_2264_cast_fp16 = add(x = var_2259_cast_fp16, y = var_2263_cast_fp16)[name = string("op_2264_cast_fp16")];
            fp32 var_2265_epsilon_0 = const()[name = string("op_2265_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 4, 1]> var_2265_cast_fp16 = rsqrt(epsilon = var_2265_epsilon_0, x = var_2264_cast_fp16)[name = string("op_2265_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_2266_cast_fp16 = mul(x = y_97_cast_fp16, y = var_2265_cast_fp16)[name = string("op_2266_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_14_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185421696)))];
            tensor<fp16, [1, 96, 4, 96]> var_2267_cast_fp16 = mul(x = core_backbone_layers_14_self_attn_k_norm_weight_to_fp16, y = var_2266_cast_fp16)[name = string("op_2267_cast_fp16")];
            tensor<int32, [4]> new_37_perm_0 = const()[name = string("new_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185421952)))];
            tensor<fp16, [1, 96, 384]> linear_80_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_v_proj_weight_to_fp16, x = input_197_cast_fp16)[name = string("linear_80_cast_fp16")];
            tensor<int32, [4]> concat_35 = const()[name = string("concat_35"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> var_2272_cast_fp16 = reshape(shape = concat_35, x = linear_80_cast_fp16)[name = string("op_2272_cast_fp16")];
            tensor<int32, [4]> new_39_perm_0 = const()[name = string("new_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_201_begin_0 = const()[name = string("input_201_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])];
            tensor<int32, [5]> input_201_end_0 = const()[name = string("input_201_end_0"), val = tensor<int32, [5]>([5, 1, 4, 31, 96])];
            tensor<bool, [5]> input_201_end_mask_0 = const()[name = string("input_201_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_201_squeeze_mask_0 = const()[name = string("input_201_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_201_cast_fp16 = slice_by_index(begin = input_201_begin_0, end = input_201_end_0, end_mask = input_201_end_mask_0, squeeze_mask = input_201_squeeze_mask_0, x = k)[name = string("input_201_cast_fp16")];
            tensor<int32, [8]> var_2277_pad_0 = const()[name = string("op_2277_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2277_mode_0 = const()[name = string("op_2277_mode_0"), val = string("constant")];
            fp16 const_48_to_fp16 = const()[name = string("const_48_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_37_cast_fp16 = transpose(perm = new_37_perm_0, x = var_2267_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 4, 127, 96]> var_2277_cast_fp16 = pad(constant_val = const_48_to_fp16, mode = var_2277_mode_0, pad = var_2277_pad_0, x = new_37_cast_fp16)[name = string("op_2277_cast_fp16")];
            tensor<int32, [8]> var_2279_pad_0 = const()[name = string("op_2279_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_2279_mode_0 = const()[name = string("op_2279_mode_0"), val = string("constant")];
            fp16 const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_2279_cast_fp16 = pad(constant_val = const_49_to_fp16, mode = var_2279_mode_0, pad = var_2279_pad_0, x = input_201_cast_fp16)[name = string("op_2279_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> x_155_cast_fp16 = add(x = var_2277_cast_fp16, y = var_2279_cast_fp16)[name = string("x_155_cast_fp16")];
            tensor<int32, [5]> input_205_begin_0 = const()[name = string("input_205_begin_0"), val = tensor<int32, [5]>([4, 0, 0, 0, 0])];
            tensor<int32, [5]> input_205_end_0 = const()[name = string("input_205_end_0"), val = tensor<int32, [5]>([5, 1, 4, 31, 96])];
            tensor<bool, [5]> input_205_end_mask_0 = const()[name = string("input_205_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_205_squeeze_mask_0 = const()[name = string("input_205_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_205_cast_fp16 = slice_by_index(begin = input_205_begin_0, end = input_205_end_0, end_mask = input_205_end_mask_0, squeeze_mask = input_205_squeeze_mask_0, x = v)[name = string("input_205_cast_fp16")];
            tensor<int32, [8]> var_2284_pad_0 = const()[name = string("op_2284_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2284_mode_0 = const()[name = string("op_2284_mode_0"), val = string("constant")];
            fp16 const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_39_cast_fp16 = transpose(perm = new_39_perm_0, x = var_2272_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 4, 127, 96]> var_2284_cast_fp16 = pad(constant_val = const_50_to_fp16, mode = var_2284_mode_0, pad = var_2284_pad_0, x = new_39_cast_fp16)[name = string("op_2284_cast_fp16")];
            tensor<int32, [8]> var_2286_pad_0 = const()[name = string("op_2286_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_2286_mode_0 = const()[name = string("op_2286_mode_0"), val = string("constant")];
            fp16 const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_2286_cast_fp16 = pad(constant_val = const_51_to_fp16, mode = var_2286_mode_0, pad = var_2286_pad_0, x = input_205_cast_fp16)[name = string("op_2286_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> v_19_cast_fp16 = add(x = var_2284_cast_fp16, y = var_2286_cast_fp16)[name = string("v_19_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2288_cast_fp16 = mul(x = var_2245_cast_fp16, y = const_66_to_fp16)[name = string("op_2288_cast_fp16")];
            tensor<int32, [2]> var_2289_split_sizes_0 = const()[name = string("op_2289_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_83 = const()[name = string("const_83"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 48]> var_2289_cast_fp16_0, tensor<fp16, [1, 96, 8, 48]> var_2289_cast_fp16_1 = split(axis = const_83, split_sizes = var_2289_split_sizes_0, x = var_2245_cast_fp16)[name = string("op_2289_cast_fp16")];
            fp16 const_52_promoted_to_fp16 = const()[name = string("const_52_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 96, 8, 48]> var_2291_cast_fp16 = mul(x = var_2289_cast_fp16_1, y = const_52_promoted_to_fp16)[name = string("op_2291_cast_fp16")];
            bool var_2293_interleave_0 = const()[name = string("op_2293_interleave_0"), val = bool(false)];
            int32 const_84 = const()[name = string("const_84"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 96]> var_2293_cast_fp16 = concat(axis = const_84, interleave = var_2293_interleave_0, values = (var_2291_cast_fp16, var_2289_cast_fp16_0))[name = string("op_2293_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2294_cast_fp16 = mul(x = var_2293_cast_fp16, y = const_69_to_fp16)[name = string("op_2294_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> q_9_cast_fp16 = add(x = var_2288_cast_fp16, y = var_2294_cast_fp16)[name = string("q_9_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_2296_cast_fp16 = mul(x = x_155_cast_fp16, y = const_0_to_fp16)[name = string("op_2296_cast_fp16")];
            tensor<int32, [2]> var_2297_split_sizes_0 = const()[name = string("op_2297_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2297_axis_0 = const()[name = string("op_2297_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 127, 48]> var_2297_cast_fp16_0, tensor<fp16, [1, 4, 127, 48]> var_2297_cast_fp16_1 = split(axis = var_2297_axis_0, split_sizes = var_2297_split_sizes_0, x = x_155_cast_fp16)[name = string("op_2297_cast_fp16")];
            fp16 const_53_promoted_to_fp16 = const()[name = string("const_53_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 127, 48]> var_2299_cast_fp16 = mul(x = var_2297_cast_fp16_1, y = const_53_promoted_to_fp16)[name = string("op_2299_cast_fp16")];
            bool var_2301_interleave_0 = const()[name = string("op_2301_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 127, 96]> var_2301_cast_fp16 = concat(axis = var_126, interleave = var_2301_interleave_0, values = (var_2299_cast_fp16, var_2297_cast_fp16_0))[name = string("op_2301_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_2302_cast_fp16 = mul(x = var_2301_cast_fp16, y = const_1_to_fp16)[name = string("op_2302_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> k_19_cast_fp16 = add(x = var_2296_cast_fp16, y = var_2302_cast_fp16)[name = string("k_19_cast_fp16")];
            tensor<int32, [1]> var_2304_axes_0 = const()[name = string("op_2304_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_2304_cast_fp16 = expand_dims(axes = var_2304_axes_0, x = k_19_cast_fp16)[name = string("op_2304_cast_fp16")];
            tensor<int32, [5]> var_2306_reps_0 = const()[name = string("op_2306_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_2306_cast_fp16 = tile(reps = var_2306_reps_0, x = var_2304_cast_fp16)[name = string("op_2306_cast_fp16")];
            tensor<int32, [4]> concat_37 = const()[name = string("concat_37"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> k_21_cast_fp16 = reshape(shape = concat_37, x = var_2306_cast_fp16)[name = string("k_21_cast_fp16")];
            tensor<int32, [1]> var_2308_axes_0 = const()[name = string("op_2308_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_2308_cast_fp16 = expand_dims(axes = var_2308_axes_0, x = v_19_cast_fp16)[name = string("op_2308_cast_fp16")];
            tensor<int32, [5]> var_2310_reps_0 = const()[name = string("op_2310_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_2310_cast_fp16 = tile(reps = var_2310_reps_0, x = var_2308_cast_fp16)[name = string("op_2310_cast_fp16")];
            tensor<int32, [4]> concat_39 = const()[name = string("concat_39"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> v_21_cast_fp16 = reshape(shape = concat_39, x = var_2310_cast_fp16)[name = string("v_21_cast_fp16")];
            tensor<int32, [4]> transpose_8_perm_0 = const()[name = string("transpose_8_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2313_transpose_x_1 = const()[name = string("op_2313_transpose_x_1"), val = bool(false)];
            bool var_2313_transpose_y_1 = const()[name = string("op_2313_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 96]> transpose_8_cast_fp16 = transpose(perm = transpose_8_perm_0, x = q_9_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 8, 96, 127]> var_2313_cast_fp16 = matmul(transpose_x = var_2313_transpose_x_1, transpose_y = var_2313_transpose_y_1, x = transpose_8_cast_fp16, y = k_21_cast_fp16)[name = string("op_2313_cast_fp16")];
            fp16 var_2314_to_fp16 = const()[name = string("op_2314_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 96, 127]> s_87_cast_fp16 = mul(x = var_2313_cast_fp16, y = var_2314_to_fp16)[name = string("s_87_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_157_cast_fp16 = add(x = s_87_cast_fp16, y = mask)[name = string("x_157_cast_fp16")];
            tensor<int32, [1]> var_2318_axes_0 = const()[name = string("op_2318_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 96, 1]> var_2318_cast_fp16 = reduce_max(axes = var_2318_axes_0, keep_dims = var_127, x = x_157_cast_fp16)[name = string("op_2318_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_159_cast_fp16 = sub(x = x_157_cast_fp16, y = var_2318_cast_fp16)[name = string("x_159_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_161_cast_fp16 = exp(x = x_159_cast_fp16)[name = string("x_161_cast_fp16")];
            tensor<int32, [1]> var_2322_axes_0 = const()[name = string("op_2322_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2322_keep_dims_0 = const()[name = string("op_2322_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 1]> var_2322_cast_fp16 = reduce_sum(axes = var_2322_axes_0, keep_dims = var_2322_keep_dims_0, x = x_161_cast_fp16)[name = string("op_2322_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> var_2323_cast_fp16 = real_div(x = x_161_cast_fp16, y = var_2322_cast_fp16)[name = string("op_2323_cast_fp16")];
            bool o_17_transpose_x_0 = const()[name = string("o_17_transpose_x_0"), val = bool(false)];
            bool o_17_transpose_y_0 = const()[name = string("o_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 96, 96]> o_17_cast_fp16 = matmul(transpose_x = o_17_transpose_x_0, transpose_y = o_17_transpose_y_0, x = var_2323_cast_fp16, y = v_21_cast_fp16)[name = string("o_17_cast_fp16")];
            tensor<int32, [4]> o_19_perm_0 = const()[name = string("o_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_40 = const()[name = string("concat_40"), val = tensor<int32, [3]>([1, 96, 768])];
            tensor<fp16, [1, 96, 8, 96]> o_19_cast_fp16 = transpose(perm = o_19_perm_0, x = o_17_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 96, 768]> input_207_cast_fp16 = reshape(shape = concat_40, x = o_19_cast_fp16)[name = string("input_207_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186011840)))];
            tensor<fp16, [1, 96, 768]> linear_81_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_self_attn_o_proj_weight_to_fp16, x = input_207_cast_fp16)[name = string("linear_81_cast_fp16")];
            tensor<int32, [4]> var_2332_begin_0 = const()[name = string("op_2332_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_2332_end_0 = const()[name = string("op_2332_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_2332_end_mask_0 = const()[name = string("op_2332_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2332_cast_fp16 = slice_by_index(begin = var_2332_begin_0, end = var_2332_end_0, end_mask = var_2332_end_mask_0, x = x_155_cast_fp16)[name = string("op_2332_cast_fp16")];
            tensor<int32, [4]> var_2335_begin_0 = const()[name = string("op_2335_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_2335_end_0 = const()[name = string("op_2335_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_2335_end_mask_0 = const()[name = string("op_2335_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2335_cast_fp16 = slice_by_index(begin = var_2335_begin_0, end = var_2335_end_0, end_mask = var_2335_end_mask_0, x = v_19_cast_fp16)[name = string("op_2335_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_163_cast_fp16 = add(x = x_147_cast_fp16, y = linear_81_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2338_cast_fp16 = abs(x = x_163_cast_fp16)[name = string("op_2338_cast_fp16")];
            tensor<int32, [1]> var_2340_axes_0 = const()[name = string("op_2340_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2340_cast_fp16 = reduce_max(axes = var_2340_axes_0, keep_dims = var_127, x = var_2338_cast_fp16)[name = string("op_2340_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_89_cast_fp16 = maximum(x = var_2340_cast_fp16, y = var_128_to_fp16)[name = string("s_89_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_99_cast_fp16 = real_div(x = x_163_cast_fp16, y = s_89_cast_fp16)[name = string("y_99_cast_fp16")];
            fp16 var_129_promoted_39_to_fp16 = const()[name = string("op_129_promoted_39_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2343_cast_fp16 = pow(x = y_99_cast_fp16, y = var_129_promoted_39_to_fp16)[name = string("op_2343_cast_fp16")];
            tensor<int32, [1]> var_2345_axes_0 = const()[name = string("op_2345_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2345_keep_dims_0 = const()[name = string("op_2345_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2345_cast_fp16 = reduce_mean(axes = var_2345_axes_0, keep_dims = var_2345_keep_dims_0, x = var_2343_cast_fp16)[name = string("op_2345_cast_fp16")];
            fp32 var_2346_epsilon_0 = const()[name = string("op_2346_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2346_cast_fp16 = inverse(epsilon = var_2346_epsilon_0, x = s_89_cast_fp16)[name = string("op_2346_cast_fp16")];
            fp16 var_2347_to_fp16 = const()[name = string("op_2347_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2348_cast_fp16 = mul(x = var_2346_cast_fp16, y = var_2347_to_fp16)[name = string("op_2348_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2349_cast_fp16 = real_div(x = var_2348_cast_fp16, y = s_89_cast_fp16)[name = string("op_2349_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2350_cast_fp16 = add(x = var_2345_cast_fp16, y = var_2349_cast_fp16)[name = string("op_2350_cast_fp16")];
            fp32 var_2351_epsilon_0 = const()[name = string("op_2351_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2351_cast_fp16 = rsqrt(epsilon = var_2351_epsilon_0, x = var_2350_cast_fp16)[name = string("op_2351_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2352_cast_fp16 = mul(x = y_99_cast_fp16, y = var_2351_cast_fp16)[name = string("op_2352_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_14_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_14_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187191552)))];
            tensor<fp16, [1, 96, 768]> input_209_cast_fp16 = mul(x = core_backbone_layers_14_ffn_norm_weight_to_fp16, y = var_2352_cast_fp16)[name = string("input_209_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_14_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187193152)))];
            tensor<fp16, [1, 96, 1792]> linear_82_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w1_weight_to_fp16, x = input_209_cast_fp16)[name = string("linear_82_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_14_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189945728)))];
            tensor<fp16, [1, 96, 1792]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w3_weight_to_fp16, x = input_209_cast_fp16)[name = string("linear_83_cast_fp16")];
            fp16 var_2361_to_fp16 = const()[name = string("op_2361_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_2362_cast_fp16 = mul(x = linear_82_cast_fp16, y = var_2361_to_fp16)[name = string("op_2362_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2365_cast_fp16 = tanh(x = var_2362_cast_fp16)[name = string("op_2365_cast_fp16")];
            fp16 var_2366_to_fp16 = const()[name = string("op_2366_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_2367_cast_fp16 = add(x = var_2365_cast_fp16, y = var_2366_to_fp16)[name = string("op_2367_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2368_cast_fp16 = mul(x = var_2362_cast_fp16, y = var_2367_cast_fp16)[name = string("op_2368_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_211_cast_fp16 = mul(x = var_2368_cast_fp16, y = linear_83_cast_fp16)[name = string("input_211_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_14_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_14_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192698304)))];
            tensor<fp16, [1, 96, 768]> linear_84_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_14_feed_forward_w2_weight_to_fp16, x = input_211_cast_fp16)[name = string("linear_84_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_167_cast_fp16 = add(x = x_163_cast_fp16, y = linear_84_cast_fp16)[name = string("x_167_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2374_cast_fp16 = abs(x = x_167_cast_fp16)[name = string("op_2374_cast_fp16")];
            tensor<int32, [1]> var_2376_axes_0 = const()[name = string("op_2376_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2376_cast_fp16 = reduce_max(axes = var_2376_axes_0, keep_dims = var_127, x = var_2374_cast_fp16)[name = string("op_2376_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_91_cast_fp16 = maximum(x = var_2376_cast_fp16, y = var_128_to_fp16)[name = string("s_91_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_101_cast_fp16 = real_div(x = x_167_cast_fp16, y = s_91_cast_fp16)[name = string("y_101_cast_fp16")];
            fp16 var_129_promoted_40_to_fp16 = const()[name = string("op_129_promoted_40_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2379_cast_fp16 = pow(x = y_101_cast_fp16, y = var_129_promoted_40_to_fp16)[name = string("op_2379_cast_fp16")];
            tensor<int32, [1]> var_2381_axes_0 = const()[name = string("op_2381_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2381_keep_dims_0 = const()[name = string("op_2381_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2381_cast_fp16 = reduce_mean(axes = var_2381_axes_0, keep_dims = var_2381_keep_dims_0, x = var_2379_cast_fp16)[name = string("op_2381_cast_fp16")];
            fp32 var_2382_epsilon_0 = const()[name = string("op_2382_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2382_cast_fp16 = inverse(epsilon = var_2382_epsilon_0, x = s_91_cast_fp16)[name = string("op_2382_cast_fp16")];
            fp16 var_2383_to_fp16 = const()[name = string("op_2383_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2384_cast_fp16 = mul(x = var_2382_cast_fp16, y = var_2383_to_fp16)[name = string("op_2384_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2385_cast_fp16 = real_div(x = var_2384_cast_fp16, y = s_91_cast_fp16)[name = string("op_2385_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2386_cast_fp16 = add(x = var_2381_cast_fp16, y = var_2385_cast_fp16)[name = string("op_2386_cast_fp16")];
            fp32 var_2387_epsilon_0 = const()[name = string("op_2387_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2387_cast_fp16 = rsqrt(epsilon = var_2387_epsilon_0, x = var_2386_cast_fp16)[name = string("op_2387_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2388_cast_fp16 = mul(x = y_101_cast_fp16, y = var_2387_cast_fp16)[name = string("op_2388_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_15_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195450880)))];
            tensor<fp16, [1, 96, 768]> input_213_cast_fp16 = mul(x = core_backbone_layers_15_operator_norm_weight_to_fp16, y = var_2388_cast_fp16)[name = string("input_213_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195452480)))];
            tensor<fp16, [1, 96, 768]> linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_q_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_85_cast_fp16")];
            tensor<int32, [4]> concat_41 = const()[name = string("concat_41"), val = tensor<int32, [4]>([1, 96, 8, 96])];
            tensor<fp16, [1, 96, 8, 96]> x_169_cast_fp16 = reshape(shape = concat_41, x = linear_85_cast_fp16)[name = string("x_169_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2395_cast_fp16 = abs(x = x_169_cast_fp16)[name = string("op_2395_cast_fp16")];
            tensor<int32, [1]> var_2397_axes_0 = const()[name = string("op_2397_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 8, 1]> var_2397_cast_fp16 = reduce_max(axes = var_2397_axes_0, keep_dims = var_127, x = var_2395_cast_fp16)[name = string("op_2397_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> s_93_cast_fp16 = maximum(x = var_2397_cast_fp16, y = var_128_to_fp16)[name = string("s_93_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> y_103_cast_fp16 = real_div(x = x_169_cast_fp16, y = s_93_cast_fp16)[name = string("y_103_cast_fp16")];
            fp16 var_129_promoted_41_to_fp16 = const()[name = string("op_129_promoted_41_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 8, 96]> var_2400_cast_fp16 = pow(x = y_103_cast_fp16, y = var_129_promoted_41_to_fp16)[name = string("op_2400_cast_fp16")];
            tensor<int32, [1]> var_2402_axes_0 = const()[name = string("op_2402_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2402_keep_dims_0 = const()[name = string("op_2402_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 8, 1]> var_2402_cast_fp16 = reduce_mean(axes = var_2402_axes_0, keep_dims = var_2402_keep_dims_0, x = var_2400_cast_fp16)[name = string("op_2402_cast_fp16")];
            fp32 var_2403_epsilon_0 = const()[name = string("op_2403_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 8, 1]> var_2403_cast_fp16 = inverse(epsilon = var_2403_epsilon_0, x = s_93_cast_fp16)[name = string("op_2403_cast_fp16")];
            fp16 var_2404_to_fp16 = const()[name = string("op_2404_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 8, 1]> var_2405_cast_fp16 = mul(x = var_2403_cast_fp16, y = var_2404_to_fp16)[name = string("op_2405_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_2406_cast_fp16 = real_div(x = var_2405_cast_fp16, y = s_93_cast_fp16)[name = string("op_2406_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_2407_cast_fp16 = add(x = var_2402_cast_fp16, y = var_2406_cast_fp16)[name = string("op_2407_cast_fp16")];
            fp32 var_2408_epsilon_0 = const()[name = string("op_2408_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 8, 1]> var_2408_cast_fp16 = rsqrt(epsilon = var_2408_epsilon_0, x = var_2407_cast_fp16)[name = string("op_2408_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2409_cast_fp16 = mul(x = y_103_cast_fp16, y = var_2408_cast_fp16)[name = string("op_2409_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_15_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196632192)))];
            tensor<fp16, [1, 96, 8, 96]> var_2410_cast_fp16 = mul(x = core_backbone_layers_15_self_attn_q_norm_weight_to_fp16, y = var_2409_cast_fp16)[name = string("op_2410_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196632448)))];
            tensor<fp16, [1, 96, 384]> linear_86_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_k_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_86_cast_fp16")];
            tensor<int32, [4]> concat_42 = const()[name = string("concat_42"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> x_171_cast_fp16 = reshape(shape = concat_42, x = linear_86_cast_fp16)[name = string("x_171_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_2417_cast_fp16 = abs(x = x_171_cast_fp16)[name = string("op_2417_cast_fp16")];
            tensor<int32, [1]> var_2419_axes_0 = const()[name = string("op_2419_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 4, 1]> var_2419_cast_fp16 = reduce_max(axes = var_2419_axes_0, keep_dims = var_127, x = var_2417_cast_fp16)[name = string("op_2419_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> s_95_cast_fp16 = maximum(x = var_2419_cast_fp16, y = var_128_to_fp16)[name = string("s_95_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> y_105_cast_fp16 = real_div(x = x_171_cast_fp16, y = s_95_cast_fp16)[name = string("y_105_cast_fp16")];
            fp16 var_129_promoted_42_to_fp16 = const()[name = string("op_129_promoted_42_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 4, 96]> var_2422_cast_fp16 = pow(x = y_105_cast_fp16, y = var_129_promoted_42_to_fp16)[name = string("op_2422_cast_fp16")];
            tensor<int32, [1]> var_2424_axes_0 = const()[name = string("op_2424_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2424_keep_dims_0 = const()[name = string("op_2424_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 4, 1]> var_2424_cast_fp16 = reduce_mean(axes = var_2424_axes_0, keep_dims = var_2424_keep_dims_0, x = var_2422_cast_fp16)[name = string("op_2424_cast_fp16")];
            fp32 var_2425_epsilon_0 = const()[name = string("op_2425_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 4, 1]> var_2425_cast_fp16 = inverse(epsilon = var_2425_epsilon_0, x = s_95_cast_fp16)[name = string("op_2425_cast_fp16")];
            fp16 var_2426_to_fp16 = const()[name = string("op_2426_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 4, 1]> var_2427_cast_fp16 = mul(x = var_2425_cast_fp16, y = var_2426_to_fp16)[name = string("op_2427_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_2428_cast_fp16 = real_div(x = var_2427_cast_fp16, y = s_95_cast_fp16)[name = string("op_2428_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_2429_cast_fp16 = add(x = var_2424_cast_fp16, y = var_2428_cast_fp16)[name = string("op_2429_cast_fp16")];
            fp32 var_2430_epsilon_0 = const()[name = string("op_2430_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 4, 1]> var_2430_cast_fp16 = rsqrt(epsilon = var_2430_epsilon_0, x = var_2429_cast_fp16)[name = string("op_2430_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_2431_cast_fp16 = mul(x = y_105_cast_fp16, y = var_2430_cast_fp16)[name = string("op_2431_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_15_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197222336)))];
            tensor<fp16, [1, 96, 4, 96]> var_2432_cast_fp16 = mul(x = core_backbone_layers_15_self_attn_k_norm_weight_to_fp16, y = var_2431_cast_fp16)[name = string("op_2432_cast_fp16")];
            tensor<int32, [4]> new_41_perm_0 = const()[name = string("new_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197222592)))];
            tensor<fp16, [1, 96, 384]> linear_87_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_v_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_87_cast_fp16")];
            tensor<int32, [4]> concat_43 = const()[name = string("concat_43"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> var_2437_cast_fp16 = reshape(shape = concat_43, x = linear_87_cast_fp16)[name = string("op_2437_cast_fp16")];
            tensor<int32, [4]> new_43_perm_0 = const()[name = string("new_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_217_begin_0 = const()[name = string("input_217_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])];
            tensor<int32, [5]> input_217_end_0 = const()[name = string("input_217_end_0"), val = tensor<int32, [5]>([6, 1, 4, 31, 96])];
            tensor<bool, [5]> input_217_end_mask_0 = const()[name = string("input_217_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_217_squeeze_mask_0 = const()[name = string("input_217_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_217_cast_fp16 = slice_by_index(begin = input_217_begin_0, end = input_217_end_0, end_mask = input_217_end_mask_0, squeeze_mask = input_217_squeeze_mask_0, x = k)[name = string("input_217_cast_fp16")];
            tensor<int32, [8]> var_2442_pad_0 = const()[name = string("op_2442_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2442_mode_0 = const()[name = string("op_2442_mode_0"), val = string("constant")];
            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_41_cast_fp16 = transpose(perm = new_41_perm_0, x = var_2432_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 4, 127, 96]> var_2442_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = var_2442_mode_0, pad = var_2442_pad_0, x = new_41_cast_fp16)[name = string("op_2442_cast_fp16")];
            tensor<int32, [8]> var_2444_pad_0 = const()[name = string("op_2444_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_2444_mode_0 = const()[name = string("op_2444_mode_0"), val = string("constant")];
            fp16 const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_2444_cast_fp16 = pad(constant_val = const_55_to_fp16, mode = var_2444_mode_0, pad = var_2444_pad_0, x = input_217_cast_fp16)[name = string("op_2444_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> x_175_cast_fp16 = add(x = var_2442_cast_fp16, y = var_2444_cast_fp16)[name = string("x_175_cast_fp16")];
            tensor<int32, [5]> input_221_begin_0 = const()[name = string("input_221_begin_0"), val = tensor<int32, [5]>([5, 0, 0, 0, 0])];
            tensor<int32, [5]> input_221_end_0 = const()[name = string("input_221_end_0"), val = tensor<int32, [5]>([6, 1, 4, 31, 96])];
            tensor<bool, [5]> input_221_end_mask_0 = const()[name = string("input_221_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_221_squeeze_mask_0 = const()[name = string("input_221_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_221_cast_fp16 = slice_by_index(begin = input_221_begin_0, end = input_221_end_0, end_mask = input_221_end_mask_0, squeeze_mask = input_221_squeeze_mask_0, x = v)[name = string("input_221_cast_fp16")];
            tensor<int32, [8]> var_2449_pad_0 = const()[name = string("op_2449_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2449_mode_0 = const()[name = string("op_2449_mode_0"), val = string("constant")];
            fp16 const_56_to_fp16 = const()[name = string("const_56_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_43_cast_fp16 = transpose(perm = new_43_perm_0, x = var_2437_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 4, 127, 96]> var_2449_cast_fp16 = pad(constant_val = const_56_to_fp16, mode = var_2449_mode_0, pad = var_2449_pad_0, x = new_43_cast_fp16)[name = string("op_2449_cast_fp16")];
            tensor<int32, [8]> var_2451_pad_0 = const()[name = string("op_2451_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_2451_mode_0 = const()[name = string("op_2451_mode_0"), val = string("constant")];
            fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_2451_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = var_2451_mode_0, pad = var_2451_pad_0, x = input_221_cast_fp16)[name = string("op_2451_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> v_23_cast_fp16 = add(x = var_2449_cast_fp16, y = var_2451_cast_fp16)[name = string("v_23_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2453_cast_fp16 = mul(x = var_2410_cast_fp16, y = const_66_to_fp16)[name = string("op_2453_cast_fp16")];
            tensor<int32, [2]> var_2454_split_sizes_0 = const()[name = string("op_2454_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_87 = const()[name = string("const_87"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 48]> var_2454_cast_fp16_0, tensor<fp16, [1, 96, 8, 48]> var_2454_cast_fp16_1 = split(axis = const_87, split_sizes = var_2454_split_sizes_0, x = var_2410_cast_fp16)[name = string("op_2454_cast_fp16")];
            fp16 const_58_promoted_to_fp16 = const()[name = string("const_58_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 96, 8, 48]> var_2456_cast_fp16 = mul(x = var_2454_cast_fp16_1, y = const_58_promoted_to_fp16)[name = string("op_2456_cast_fp16")];
            bool var_2458_interleave_0 = const()[name = string("op_2458_interleave_0"), val = bool(false)];
            int32 const_88 = const()[name = string("const_88"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 96]> var_2458_cast_fp16 = concat(axis = const_88, interleave = var_2458_interleave_0, values = (var_2456_cast_fp16, var_2454_cast_fp16_0))[name = string("op_2458_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2459_cast_fp16 = mul(x = var_2458_cast_fp16, y = const_69_to_fp16)[name = string("op_2459_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> q_11_cast_fp16 = add(x = var_2453_cast_fp16, y = var_2459_cast_fp16)[name = string("q_11_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_2461_cast_fp16 = mul(x = x_175_cast_fp16, y = const_0_to_fp16)[name = string("op_2461_cast_fp16")];
            tensor<int32, [2]> var_2462_split_sizes_0 = const()[name = string("op_2462_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2462_axis_0 = const()[name = string("op_2462_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 127, 48]> var_2462_cast_fp16_0, tensor<fp16, [1, 4, 127, 48]> var_2462_cast_fp16_1 = split(axis = var_2462_axis_0, split_sizes = var_2462_split_sizes_0, x = x_175_cast_fp16)[name = string("op_2462_cast_fp16")];
            fp16 const_59_promoted_to_fp16 = const()[name = string("const_59_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 127, 48]> var_2464_cast_fp16 = mul(x = var_2462_cast_fp16_1, y = const_59_promoted_to_fp16)[name = string("op_2464_cast_fp16")];
            bool var_2466_interleave_0 = const()[name = string("op_2466_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 127, 96]> var_2466_cast_fp16 = concat(axis = var_126, interleave = var_2466_interleave_0, values = (var_2464_cast_fp16, var_2462_cast_fp16_0))[name = string("op_2466_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_2467_cast_fp16 = mul(x = var_2466_cast_fp16, y = const_1_to_fp16)[name = string("op_2467_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> k_23_cast_fp16 = add(x = var_2461_cast_fp16, y = var_2467_cast_fp16)[name = string("k_23_cast_fp16")];
            tensor<int32, [1]> var_2469_axes_0 = const()[name = string("op_2469_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_2469_cast_fp16 = expand_dims(axes = var_2469_axes_0, x = k_23_cast_fp16)[name = string("op_2469_cast_fp16")];
            tensor<int32, [5]> var_2471_reps_0 = const()[name = string("op_2471_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_2471_cast_fp16 = tile(reps = var_2471_reps_0, x = var_2469_cast_fp16)[name = string("op_2471_cast_fp16")];
            tensor<int32, [4]> concat_45 = const()[name = string("concat_45"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> k_25_cast_fp16 = reshape(shape = concat_45, x = var_2471_cast_fp16)[name = string("k_25_cast_fp16")];
            tensor<int32, [1]> var_2473_axes_0 = const()[name = string("op_2473_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_2473_cast_fp16 = expand_dims(axes = var_2473_axes_0, x = v_23_cast_fp16)[name = string("op_2473_cast_fp16")];
            tensor<int32, [5]> var_2475_reps_0 = const()[name = string("op_2475_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_2475_cast_fp16 = tile(reps = var_2475_reps_0, x = var_2473_cast_fp16)[name = string("op_2475_cast_fp16")];
            tensor<int32, [4]> concat_47 = const()[name = string("concat_47"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> v_25_cast_fp16 = reshape(shape = concat_47, x = var_2475_cast_fp16)[name = string("v_25_cast_fp16")];
            tensor<int32, [4]> transpose_10_perm_0 = const()[name = string("transpose_10_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2478_transpose_x_1 = const()[name = string("op_2478_transpose_x_1"), val = bool(false)];
            bool var_2478_transpose_y_1 = const()[name = string("op_2478_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 96]> transpose_10_cast_fp16 = transpose(perm = transpose_10_perm_0, x = q_11_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 8, 96, 127]> var_2478_cast_fp16 = matmul(transpose_x = var_2478_transpose_x_1, transpose_y = var_2478_transpose_y_1, x = transpose_10_cast_fp16, y = k_25_cast_fp16)[name = string("op_2478_cast_fp16")];
            fp16 var_2479_to_fp16 = const()[name = string("op_2479_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 96, 127]> s_97_cast_fp16 = mul(x = var_2478_cast_fp16, y = var_2479_to_fp16)[name = string("s_97_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_177_cast_fp16 = add(x = s_97_cast_fp16, y = mask)[name = string("x_177_cast_fp16")];
            tensor<int32, [1]> var_2483_axes_0 = const()[name = string("op_2483_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 96, 1]> var_2483_cast_fp16 = reduce_max(axes = var_2483_axes_0, keep_dims = var_127, x = x_177_cast_fp16)[name = string("op_2483_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_179_cast_fp16 = sub(x = x_177_cast_fp16, y = var_2483_cast_fp16)[name = string("x_179_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_181_cast_fp16 = exp(x = x_179_cast_fp16)[name = string("x_181_cast_fp16")];
            tensor<int32, [1]> var_2487_axes_0 = const()[name = string("op_2487_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2487_keep_dims_0 = const()[name = string("op_2487_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 1]> var_2487_cast_fp16 = reduce_sum(axes = var_2487_axes_0, keep_dims = var_2487_keep_dims_0, x = x_181_cast_fp16)[name = string("op_2487_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> var_2488_cast_fp16 = real_div(x = x_181_cast_fp16, y = var_2487_cast_fp16)[name = string("op_2488_cast_fp16")];
            bool o_21_transpose_x_0 = const()[name = string("o_21_transpose_x_0"), val = bool(false)];
            bool o_21_transpose_y_0 = const()[name = string("o_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 96, 96]> o_21_cast_fp16 = matmul(transpose_x = o_21_transpose_x_0, transpose_y = o_21_transpose_y_0, x = var_2488_cast_fp16, y = v_25_cast_fp16)[name = string("o_21_cast_fp16")];
            tensor<int32, [4]> o_23_perm_0 = const()[name = string("o_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_48 = const()[name = string("concat_48"), val = tensor<int32, [3]>([1, 96, 768])];
            tensor<fp16, [1, 96, 8, 96]> o_23_cast_fp16 = transpose(perm = o_23_perm_0, x = o_21_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 96, 768]> input_223_cast_fp16 = reshape(shape = concat_48, x = o_23_cast_fp16)[name = string("input_223_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197812480)))];
            tensor<fp16, [1, 96, 768]> linear_88_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_self_attn_o_proj_weight_to_fp16, x = input_223_cast_fp16)[name = string("linear_88_cast_fp16")];
            tensor<int32, [4]> var_2497_begin_0 = const()[name = string("op_2497_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_2497_end_0 = const()[name = string("op_2497_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_2497_end_mask_0 = const()[name = string("op_2497_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2497_cast_fp16 = slice_by_index(begin = var_2497_begin_0, end = var_2497_end_0, end_mask = var_2497_end_mask_0, x = x_175_cast_fp16)[name = string("op_2497_cast_fp16")];
            tensor<int32, [4]> var_2500_begin_0 = const()[name = string("op_2500_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_2500_end_0 = const()[name = string("op_2500_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_2500_end_mask_0 = const()[name = string("op_2500_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2500_cast_fp16 = slice_by_index(begin = var_2500_begin_0, end = var_2500_end_0, end_mask = var_2500_end_mask_0, x = v_23_cast_fp16)[name = string("op_2500_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_183_cast_fp16 = add(x = x_167_cast_fp16, y = linear_88_cast_fp16)[name = string("x_183_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2503_cast_fp16 = abs(x = x_183_cast_fp16)[name = string("op_2503_cast_fp16")];
            tensor<int32, [1]> var_2505_axes_0 = const()[name = string("op_2505_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2505_cast_fp16 = reduce_max(axes = var_2505_axes_0, keep_dims = var_127, x = var_2503_cast_fp16)[name = string("op_2505_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_99_cast_fp16 = maximum(x = var_2505_cast_fp16, y = var_128_to_fp16)[name = string("s_99_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_107_cast_fp16 = real_div(x = x_183_cast_fp16, y = s_99_cast_fp16)[name = string("y_107_cast_fp16")];
            fp16 var_129_promoted_43_to_fp16 = const()[name = string("op_129_promoted_43_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2508_cast_fp16 = pow(x = y_107_cast_fp16, y = var_129_promoted_43_to_fp16)[name = string("op_2508_cast_fp16")];
            tensor<int32, [1]> var_2510_axes_0 = const()[name = string("op_2510_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2510_keep_dims_0 = const()[name = string("op_2510_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2510_cast_fp16 = reduce_mean(axes = var_2510_axes_0, keep_dims = var_2510_keep_dims_0, x = var_2508_cast_fp16)[name = string("op_2510_cast_fp16")];
            fp32 var_2511_epsilon_0 = const()[name = string("op_2511_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2511_cast_fp16 = inverse(epsilon = var_2511_epsilon_0, x = s_99_cast_fp16)[name = string("op_2511_cast_fp16")];
            fp16 var_2512_to_fp16 = const()[name = string("op_2512_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2513_cast_fp16 = mul(x = var_2511_cast_fp16, y = var_2512_to_fp16)[name = string("op_2513_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2514_cast_fp16 = real_div(x = var_2513_cast_fp16, y = s_99_cast_fp16)[name = string("op_2514_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2515_cast_fp16 = add(x = var_2510_cast_fp16, y = var_2514_cast_fp16)[name = string("op_2515_cast_fp16")];
            fp32 var_2516_epsilon_0 = const()[name = string("op_2516_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2516_cast_fp16 = rsqrt(epsilon = var_2516_epsilon_0, x = var_2515_cast_fp16)[name = string("op_2516_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2517_cast_fp16 = mul(x = y_107_cast_fp16, y = var_2516_cast_fp16)[name = string("op_2517_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_15_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_15_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198992192)))];
            tensor<fp16, [1, 96, 768]> input_225_cast_fp16 = mul(x = core_backbone_layers_15_ffn_norm_weight_to_fp16, y = var_2517_cast_fp16)[name = string("input_225_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_15_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198993792)))];
            tensor<fp16, [1, 96, 1792]> linear_89_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w1_weight_to_fp16, x = input_225_cast_fp16)[name = string("linear_89_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_15_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201746368)))];
            tensor<fp16, [1, 96, 1792]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w3_weight_to_fp16, x = input_225_cast_fp16)[name = string("linear_90_cast_fp16")];
            fp16 var_2526_to_fp16 = const()[name = string("op_2526_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_2527_cast_fp16 = mul(x = linear_89_cast_fp16, y = var_2526_to_fp16)[name = string("op_2527_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2530_cast_fp16 = tanh(x = var_2527_cast_fp16)[name = string("op_2530_cast_fp16")];
            fp16 var_2531_to_fp16 = const()[name = string("op_2531_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_2532_cast_fp16 = add(x = var_2530_cast_fp16, y = var_2531_to_fp16)[name = string("op_2532_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2533_cast_fp16 = mul(x = var_2527_cast_fp16, y = var_2532_cast_fp16)[name = string("op_2533_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_227_cast_fp16 = mul(x = var_2533_cast_fp16, y = linear_90_cast_fp16)[name = string("input_227_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_15_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_15_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204498944)))];
            tensor<fp16, [1, 96, 768]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_15_feed_forward_w2_weight_to_fp16, x = input_227_cast_fp16)[name = string("linear_91_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_187_cast_fp16 = add(x = x_183_cast_fp16, y = linear_91_cast_fp16)[name = string("x_187_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2539_cast_fp16 = abs(x = x_187_cast_fp16)[name = string("op_2539_cast_fp16")];
            tensor<int32, [1]> var_2541_axes_0 = const()[name = string("op_2541_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2541_cast_fp16 = reduce_max(axes = var_2541_axes_0, keep_dims = var_127, x = var_2539_cast_fp16)[name = string("op_2541_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_101_cast_fp16 = maximum(x = var_2541_cast_fp16, y = var_128_to_fp16)[name = string("s_101_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_109_cast_fp16 = real_div(x = x_187_cast_fp16, y = s_101_cast_fp16)[name = string("y_109_cast_fp16")];
            fp16 var_129_promoted_44_to_fp16 = const()[name = string("op_129_promoted_44_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2544_cast_fp16 = pow(x = y_109_cast_fp16, y = var_129_promoted_44_to_fp16)[name = string("op_2544_cast_fp16")];
            tensor<int32, [1]> var_2546_axes_0 = const()[name = string("op_2546_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2546_keep_dims_0 = const()[name = string("op_2546_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2546_cast_fp16 = reduce_mean(axes = var_2546_axes_0, keep_dims = var_2546_keep_dims_0, x = var_2544_cast_fp16)[name = string("op_2546_cast_fp16")];
            fp32 var_2547_epsilon_0 = const()[name = string("op_2547_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2547_cast_fp16 = inverse(epsilon = var_2547_epsilon_0, x = s_101_cast_fp16)[name = string("op_2547_cast_fp16")];
            fp16 var_2548_to_fp16 = const()[name = string("op_2548_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2549_cast_fp16 = mul(x = var_2547_cast_fp16, y = var_2548_to_fp16)[name = string("op_2549_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2550_cast_fp16 = real_div(x = var_2549_cast_fp16, y = s_101_cast_fp16)[name = string("op_2550_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2551_cast_fp16 = add(x = var_2546_cast_fp16, y = var_2550_cast_fp16)[name = string("op_2551_cast_fp16")];
            fp32 var_2552_epsilon_0 = const()[name = string("op_2552_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2552_cast_fp16 = rsqrt(epsilon = var_2552_epsilon_0, x = var_2551_cast_fp16)[name = string("op_2552_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2553_cast_fp16 = mul(x = y_109_cast_fp16, y = var_2552_cast_fp16)[name = string("op_2553_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_16_operator_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_operator_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207251520)))];
            tensor<fp16, [1, 96, 768]> input_229_cast_fp16 = mul(x = core_backbone_layers_16_operator_norm_weight_to_fp16, y = var_2553_cast_fp16)[name = string("input_229_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207253120)))];
            tensor<fp16, [1, 96, 768]> linear_92_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_q_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_92_cast_fp16")];
            tensor<int32, [4]> concat_49 = const()[name = string("concat_49"), val = tensor<int32, [4]>([1, 96, 8, 96])];
            tensor<fp16, [1, 96, 8, 96]> x_189_cast_fp16 = reshape(shape = concat_49, x = linear_92_cast_fp16)[name = string("x_189_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2560_cast_fp16 = abs(x = x_189_cast_fp16)[name = string("op_2560_cast_fp16")];
            tensor<int32, [1]> var_2562_axes_0 = const()[name = string("op_2562_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 8, 1]> var_2562_cast_fp16 = reduce_max(axes = var_2562_axes_0, keep_dims = var_127, x = var_2560_cast_fp16)[name = string("op_2562_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> s_103_cast_fp16 = maximum(x = var_2562_cast_fp16, y = var_128_to_fp16)[name = string("s_103_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> y_111_cast_fp16 = real_div(x = x_189_cast_fp16, y = s_103_cast_fp16)[name = string("y_111_cast_fp16")];
            fp16 var_129_promoted_45_to_fp16 = const()[name = string("op_129_promoted_45_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 8, 96]> var_2565_cast_fp16 = pow(x = y_111_cast_fp16, y = var_129_promoted_45_to_fp16)[name = string("op_2565_cast_fp16")];
            tensor<int32, [1]> var_2567_axes_0 = const()[name = string("op_2567_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2567_keep_dims_0 = const()[name = string("op_2567_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 8, 1]> var_2567_cast_fp16 = reduce_mean(axes = var_2567_axes_0, keep_dims = var_2567_keep_dims_0, x = var_2565_cast_fp16)[name = string("op_2567_cast_fp16")];
            fp32 var_2568_epsilon_0 = const()[name = string("op_2568_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 8, 1]> var_2568_cast_fp16 = inverse(epsilon = var_2568_epsilon_0, x = s_103_cast_fp16)[name = string("op_2568_cast_fp16")];
            fp16 var_2569_to_fp16 = const()[name = string("op_2569_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 8, 1]> var_2570_cast_fp16 = mul(x = var_2568_cast_fp16, y = var_2569_to_fp16)[name = string("op_2570_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_2571_cast_fp16 = real_div(x = var_2570_cast_fp16, y = s_103_cast_fp16)[name = string("op_2571_cast_fp16")];
            tensor<fp16, [1, 96, 8, 1]> var_2572_cast_fp16 = add(x = var_2567_cast_fp16, y = var_2571_cast_fp16)[name = string("op_2572_cast_fp16")];
            fp32 var_2573_epsilon_0 = const()[name = string("op_2573_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 8, 1]> var_2573_cast_fp16 = rsqrt(epsilon = var_2573_epsilon_0, x = var_2572_cast_fp16)[name = string("op_2573_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2574_cast_fp16 = mul(x = y_111_cast_fp16, y = var_2573_cast_fp16)[name = string("op_2574_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_16_self_attn_q_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_q_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208432832)))];
            tensor<fp16, [1, 96, 8, 96]> var_2575_cast_fp16 = mul(x = core_backbone_layers_16_self_attn_q_norm_weight_to_fp16, y = var_2574_cast_fp16)[name = string("op_2575_cast_fp16")];
            tensor<fp16, [384, 768]> core_backbone_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208433088)))];
            tensor<fp16, [1, 96, 384]> linear_93_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_k_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_93_cast_fp16")];
            tensor<int32, [4]> concat_50 = const()[name = string("concat_50"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> x_191_cast_fp16 = reshape(shape = concat_50, x = linear_93_cast_fp16)[name = string("x_191_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_2582_cast_fp16 = abs(x = x_191_cast_fp16)[name = string("op_2582_cast_fp16")];
            tensor<int32, [1]> var_2584_axes_0 = const()[name = string("op_2584_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 4, 1]> var_2584_cast_fp16 = reduce_max(axes = var_2584_axes_0, keep_dims = var_127, x = var_2582_cast_fp16)[name = string("op_2584_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> s_105_cast_fp16 = maximum(x = var_2584_cast_fp16, y = var_128_to_fp16)[name = string("s_105_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> y_113_cast_fp16 = real_div(x = x_191_cast_fp16, y = s_105_cast_fp16)[name = string("y_113_cast_fp16")];
            fp16 var_129_promoted_46_to_fp16 = const()[name = string("op_129_promoted_46_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 4, 96]> var_2587_cast_fp16 = pow(x = y_113_cast_fp16, y = var_129_promoted_46_to_fp16)[name = string("op_2587_cast_fp16")];
            tensor<int32, [1]> var_2589_axes_0 = const()[name = string("op_2589_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2589_keep_dims_0 = const()[name = string("op_2589_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 4, 1]> var_2589_cast_fp16 = reduce_mean(axes = var_2589_axes_0, keep_dims = var_2589_keep_dims_0, x = var_2587_cast_fp16)[name = string("op_2589_cast_fp16")];
            fp32 var_2590_epsilon_0 = const()[name = string("op_2590_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 4, 1]> var_2590_cast_fp16 = inverse(epsilon = var_2590_epsilon_0, x = s_105_cast_fp16)[name = string("op_2590_cast_fp16")];
            fp16 var_2591_to_fp16 = const()[name = string("op_2591_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 4, 1]> var_2592_cast_fp16 = mul(x = var_2590_cast_fp16, y = var_2591_to_fp16)[name = string("op_2592_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_2593_cast_fp16 = real_div(x = var_2592_cast_fp16, y = s_105_cast_fp16)[name = string("op_2593_cast_fp16")];
            tensor<fp16, [1, 96, 4, 1]> var_2594_cast_fp16 = add(x = var_2589_cast_fp16, y = var_2593_cast_fp16)[name = string("op_2594_cast_fp16")];
            fp32 var_2595_epsilon_0 = const()[name = string("op_2595_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 4, 1]> var_2595_cast_fp16 = rsqrt(epsilon = var_2595_epsilon_0, x = var_2594_cast_fp16)[name = string("op_2595_cast_fp16")];
            tensor<fp16, [1, 96, 4, 96]> var_2596_cast_fp16 = mul(x = y_113_cast_fp16, y = var_2595_cast_fp16)[name = string("op_2596_cast_fp16")];
            tensor<fp16, [96]> core_backbone_layers_16_self_attn_k_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_k_norm_weight_to_fp16"), val = tensor<fp16, [96]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209022976)))];
            tensor<fp16, [1, 96, 4, 96]> var_2597_cast_fp16 = mul(x = core_backbone_layers_16_self_attn_k_norm_weight_to_fp16, y = var_2596_cast_fp16)[name = string("op_2597_cast_fp16")];
            tensor<int32, [4]> new_45_perm_0 = const()[name = string("new_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [384, 768]> core_backbone_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209023232)))];
            tensor<fp16, [1, 96, 384]> linear_94_cast_fp16 = linear(bias = linear_11_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_v_proj_weight_to_fp16, x = input_229_cast_fp16)[name = string("linear_94_cast_fp16")];
            tensor<int32, [4]> concat_51 = const()[name = string("concat_51"), val = tensor<int32, [4]>([1, 96, 4, 96])];
            tensor<fp16, [1, 96, 4, 96]> var_2602_cast_fp16 = reshape(shape = concat_51, x = linear_94_cast_fp16)[name = string("op_2602_cast_fp16")];
            tensor<int32, [4]> new_perm_0 = const()[name = string("new_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [5]> input_233_begin_0 = const()[name = string("input_233_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])];
            tensor<int32, [5]> input_233_end_0 = const()[name = string("input_233_end_0"), val = tensor<int32, [5]>([7, 1, 4, 31, 96])];
            tensor<bool, [5]> input_233_end_mask_0 = const()[name = string("input_233_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_233_squeeze_mask_0 = const()[name = string("input_233_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_233_cast_fp16 = slice_by_index(begin = input_233_begin_0, end = input_233_end_0, end_mask = input_233_end_mask_0, squeeze_mask = input_233_squeeze_mask_0, x = k)[name = string("input_233_cast_fp16")];
            tensor<int32, [8]> var_2607_pad_0 = const()[name = string("op_2607_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2607_mode_0 = const()[name = string("op_2607_mode_0"), val = string("constant")];
            fp16 const_60_to_fp16 = const()[name = string("const_60_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_45_cast_fp16 = transpose(perm = new_45_perm_0, x = var_2597_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 4, 127, 96]> var_2607_cast_fp16 = pad(constant_val = const_60_to_fp16, mode = var_2607_mode_0, pad = var_2607_pad_0, x = new_45_cast_fp16)[name = string("op_2607_cast_fp16")];
            tensor<int32, [8]> var_2609_pad_0 = const()[name = string("op_2609_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_2609_mode_0 = const()[name = string("op_2609_mode_0"), val = string("constant")];
            fp16 const_61_to_fp16 = const()[name = string("const_61_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_2609_cast_fp16 = pad(constant_val = const_61_to_fp16, mode = var_2609_mode_0, pad = var_2609_pad_0, x = input_233_cast_fp16)[name = string("op_2609_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> x_195_cast_fp16 = add(x = var_2607_cast_fp16, y = var_2609_cast_fp16)[name = string("x_195_cast_fp16")];
            tensor<int32, [5]> input_237_begin_0 = const()[name = string("input_237_begin_0"), val = tensor<int32, [5]>([6, 0, 0, 0, 0])];
            tensor<int32, [5]> input_237_end_0 = const()[name = string("input_237_end_0"), val = tensor<int32, [5]>([7, 1, 4, 31, 96])];
            tensor<bool, [5]> input_237_end_mask_0 = const()[name = string("input_237_end_mask_0"), val = tensor<bool, [5]>([false, true, true, true, true])];
            tensor<bool, [5]> input_237_squeeze_mask_0 = const()[name = string("input_237_squeeze_mask_0"), val = tensor<bool, [5]>([true, false, false, false, false])];
            tensor<fp16, [1, 4, 31, 96]> input_237_cast_fp16 = slice_by_index(begin = input_237_begin_0, end = input_237_end_0, end_mask = input_237_end_mask_0, squeeze_mask = input_237_squeeze_mask_0, x = v)[name = string("input_237_cast_fp16")];
            tensor<int32, [8]> var_2614_pad_0 = const()[name = string("op_2614_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 31, 0, 0, 0])];
            string var_2614_mode_0 = const()[name = string("op_2614_mode_0"), val = string("constant")];
            fp16 const_62_to_fp16 = const()[name = string("const_62_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 96, 96]> new_cast_fp16 = transpose(perm = new_perm_0, x = var_2602_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 4, 127, 96]> var_2614_cast_fp16 = pad(constant_val = const_62_to_fp16, mode = var_2614_mode_0, pad = var_2614_pad_0, x = new_cast_fp16)[name = string("op_2614_cast_fp16")];
            tensor<int32, [8]> var_2616_pad_0 = const()[name = string("op_2616_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 96, 0, 0])];
            string var_2616_mode_0 = const()[name = string("op_2616_mode_0"), val = string("constant")];
            fp16 const_63_to_fp16 = const()[name = string("const_63_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 4, 127, 96]> var_2616_cast_fp16 = pad(constant_val = const_63_to_fp16, mode = var_2616_mode_0, pad = var_2616_pad_0, x = input_237_cast_fp16)[name = string("op_2616_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> v_27_cast_fp16 = add(x = var_2614_cast_fp16, y = var_2616_cast_fp16)[name = string("v_27_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2618_cast_fp16 = mul(x = var_2575_cast_fp16, y = const_66_to_fp16)[name = string("op_2618_cast_fp16")];
            tensor<int32, [2]> var_2619_split_sizes_0 = const()[name = string("op_2619_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 const_91 = const()[name = string("const_91"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 48]> var_2619_cast_fp16_0, tensor<fp16, [1, 96, 8, 48]> var_2619_cast_fp16_1 = split(axis = const_91, split_sizes = var_2619_split_sizes_0, x = var_2575_cast_fp16)[name = string("op_2619_cast_fp16")];
            fp16 const_64_promoted_to_fp16 = const()[name = string("const_64_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 96, 8, 48]> var_2621_cast_fp16 = mul(x = var_2619_cast_fp16_1, y = const_64_promoted_to_fp16)[name = string("op_2621_cast_fp16")];
            bool var_2623_interleave_0 = const()[name = string("op_2623_interleave_0"), val = bool(false)];
            int32 const_92 = const()[name = string("const_92"), val = int32(3)];
            tensor<fp16, [1, 96, 8, 96]> var_2623_cast_fp16 = concat(axis = const_92, interleave = var_2623_interleave_0, values = (var_2621_cast_fp16, var_2619_cast_fp16_0))[name = string("op_2623_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> var_2624_cast_fp16 = mul(x = var_2623_cast_fp16, y = const_69_to_fp16)[name = string("op_2624_cast_fp16")];
            tensor<fp16, [1, 96, 8, 96]> q_cast_fp16 = add(x = var_2618_cast_fp16, y = var_2624_cast_fp16)[name = string("q_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_2626_cast_fp16 = mul(x = x_195_cast_fp16, y = const_0_to_fp16)[name = string("op_2626_cast_fp16")];
            tensor<int32, [2]> var_2627_split_sizes_0 = const()[name = string("op_2627_split_sizes_0"), val = tensor<int32, [2]>([48, 48])];
            int32 var_2627_axis_0 = const()[name = string("op_2627_axis_0"), val = int32(-1)];
            tensor<fp16, [1, 4, 127, 48]> var_2627_cast_fp16_0, tensor<fp16, [1, 4, 127, 48]> var_2627_cast_fp16_1 = split(axis = var_2627_axis_0, split_sizes = var_2627_split_sizes_0, x = x_195_cast_fp16)[name = string("op_2627_cast_fp16")];
            fp16 const_65_promoted_to_fp16 = const()[name = string("const_65_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 4, 127, 48]> var_2629_cast_fp16 = mul(x = var_2627_cast_fp16_1, y = const_65_promoted_to_fp16)[name = string("op_2629_cast_fp16")];
            bool var_2631_interleave_0 = const()[name = string("op_2631_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 4, 127, 96]> var_2631_cast_fp16 = concat(axis = var_126, interleave = var_2631_interleave_0, values = (var_2629_cast_fp16, var_2627_cast_fp16_0))[name = string("op_2631_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> var_2632_cast_fp16 = mul(x = var_2631_cast_fp16, y = const_1_to_fp16)[name = string("op_2632_cast_fp16")];
            tensor<fp16, [1, 4, 127, 96]> k_27_cast_fp16 = add(x = var_2626_cast_fp16, y = var_2632_cast_fp16)[name = string("k_27_cast_fp16")];
            tensor<int32, [1]> var_2634_axes_0 = const()[name = string("op_2634_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_2634_cast_fp16 = expand_dims(axes = var_2634_axes_0, x = k_27_cast_fp16)[name = string("op_2634_cast_fp16")];
            tensor<int32, [5]> var_2636_reps_0 = const()[name = string("op_2636_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_2636_cast_fp16 = tile(reps = var_2636_reps_0, x = var_2634_cast_fp16)[name = string("op_2636_cast_fp16")];
            tensor<int32, [4]> concat_53 = const()[name = string("concat_53"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> k_cast_fp16 = reshape(shape = concat_53, x = var_2636_cast_fp16)[name = string("k_cast_fp16")];
            tensor<int32, [1]> var_2638_axes_0 = const()[name = string("op_2638_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [1, 4, 1, 127, 96]> var_2638_cast_fp16 = expand_dims(axes = var_2638_axes_0, x = v_27_cast_fp16)[name = string("op_2638_cast_fp16")];
            tensor<int32, [5]> var_2640_reps_0 = const()[name = string("op_2640_reps_0"), val = tensor<int32, [5]>([1, 1, 2, 1, 1])];
            tensor<fp16, [1, 4, 2, 127, 96]> var_2640_cast_fp16 = tile(reps = var_2640_reps_0, x = var_2638_cast_fp16)[name = string("op_2640_cast_fp16")];
            tensor<int32, [4]> concat_55 = const()[name = string("concat_55"), val = tensor<int32, [4]>([1, 8, 127, 96])];
            tensor<fp16, [1, 8, 127, 96]> v_cast_fp16 = reshape(shape = concat_55, x = var_2640_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [4]> transpose_12_perm_0 = const()[name = string("transpose_12_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            bool var_2643_transpose_x_1 = const()[name = string("op_2643_transpose_x_1"), val = bool(false)];
            bool var_2643_transpose_y_1 = const()[name = string("op_2643_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 96]> transpose_12_cast_fp16 = transpose(perm = transpose_12_perm_0, x = q_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 8, 96, 127]> var_2643_cast_fp16 = matmul(transpose_x = var_2643_transpose_x_1, transpose_y = var_2643_transpose_y_1, x = transpose_12_cast_fp16, y = k_cast_fp16)[name = string("op_2643_cast_fp16")];
            fp16 var_2644_to_fp16 = const()[name = string("op_2644_to_fp16"), val = fp16(0x1.a2p-4)];
            tensor<fp16, [1, 8, 96, 127]> s_107_cast_fp16 = mul(x = var_2643_cast_fp16, y = var_2644_to_fp16)[name = string("s_107_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_197_cast_fp16 = add(x = s_107_cast_fp16, y = mask)[name = string("x_197_cast_fp16")];
            tensor<int32, [1]> var_2648_axes_0 = const()[name = string("op_2648_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 8, 96, 1]> var_2648_cast_fp16 = reduce_max(axes = var_2648_axes_0, keep_dims = var_127, x = x_197_cast_fp16)[name = string("op_2648_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_199_cast_fp16 = sub(x = x_197_cast_fp16, y = var_2648_cast_fp16)[name = string("x_199_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> x_201_cast_fp16 = exp(x = x_199_cast_fp16)[name = string("x_201_cast_fp16")];
            tensor<int32, [1]> var_2652_axes_0 = const()[name = string("op_2652_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2652_keep_dims_0 = const()[name = string("op_2652_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 8, 96, 1]> var_2652_cast_fp16 = reduce_sum(axes = var_2652_axes_0, keep_dims = var_2652_keep_dims_0, x = x_201_cast_fp16)[name = string("op_2652_cast_fp16")];
            tensor<fp16, [1, 8, 96, 127]> var_2653_cast_fp16 = real_div(x = x_201_cast_fp16, y = var_2652_cast_fp16)[name = string("op_2653_cast_fp16")];
            bool o_25_transpose_x_0 = const()[name = string("o_25_transpose_x_0"), val = bool(false)];
            bool o_25_transpose_y_0 = const()[name = string("o_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 96, 96]> o_25_cast_fp16 = matmul(transpose_x = o_25_transpose_x_0, transpose_y = o_25_transpose_y_0, x = var_2653_cast_fp16, y = v_cast_fp16)[name = string("o_25_cast_fp16")];
            tensor<int32, [4]> o_perm_0 = const()[name = string("o_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> concat_56 = const()[name = string("concat_56"), val = tensor<int32, [3]>([1, 96, 768])];
            tensor<fp16, [1, 96, 8, 96]> o_cast_fp16 = transpose(perm = o_perm_0, x = o_25_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 96, 768]> input_239_cast_fp16 = reshape(shape = concat_56, x = o_cast_fp16)[name = string("input_239_cast_fp16")];
            tensor<fp16, [768, 768]> core_backbone_layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = string("core_backbone_layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209613120)))];
            tensor<fp16, [1, 96, 768]> linear_95_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_self_attn_o_proj_weight_to_fp16, x = input_239_cast_fp16)[name = string("linear_95_cast_fp16")];
            tensor<int32, [4]> var_2662_begin_0 = const()[name = string("op_2662_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_2662_end_0 = const()[name = string("op_2662_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_2662_end_mask_0 = const()[name = string("op_2662_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2662_cast_fp16 = slice_by_index(begin = var_2662_begin_0, end = var_2662_end_0, end_mask = var_2662_end_mask_0, x = x_195_cast_fp16)[name = string("op_2662_cast_fp16")];
            tensor<int32, [4]> var_2665_begin_0 = const()[name = string("op_2665_begin_0"), val = tensor<int32, [4]>([0, 0, 96, 0])];
            tensor<int32, [4]> var_2665_end_0 = const()[name = string("op_2665_end_0"), val = tensor<int32, [4]>([1, 4, 127, 96])];
            tensor<bool, [4]> var_2665_end_mask_0 = const()[name = string("op_2665_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 4, 31, 96]> var_2665_cast_fp16 = slice_by_index(begin = var_2665_begin_0, end = var_2665_end_0, end_mask = var_2665_end_mask_0, x = v_27_cast_fp16)[name = string("op_2665_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_203_cast_fp16 = add(x = x_187_cast_fp16, y = linear_95_cast_fp16)[name = string("x_203_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2668_cast_fp16 = abs(x = x_203_cast_fp16)[name = string("op_2668_cast_fp16")];
            tensor<int32, [1]> var_2670_axes_0 = const()[name = string("op_2670_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2670_cast_fp16 = reduce_max(axes = var_2670_axes_0, keep_dims = var_127, x = var_2668_cast_fp16)[name = string("op_2670_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_109_cast_fp16 = maximum(x = var_2670_cast_fp16, y = var_128_to_fp16)[name = string("s_109_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_115_cast_fp16 = real_div(x = x_203_cast_fp16, y = s_109_cast_fp16)[name = string("y_115_cast_fp16")];
            fp16 var_129_promoted_47_to_fp16 = const()[name = string("op_129_promoted_47_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2673_cast_fp16 = pow(x = y_115_cast_fp16, y = var_129_promoted_47_to_fp16)[name = string("op_2673_cast_fp16")];
            tensor<int32, [1]> var_2675_axes_0 = const()[name = string("op_2675_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2675_keep_dims_0 = const()[name = string("op_2675_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2675_cast_fp16 = reduce_mean(axes = var_2675_axes_0, keep_dims = var_2675_keep_dims_0, x = var_2673_cast_fp16)[name = string("op_2675_cast_fp16")];
            fp32 var_2676_epsilon_0 = const()[name = string("op_2676_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2676_cast_fp16 = inverse(epsilon = var_2676_epsilon_0, x = s_109_cast_fp16)[name = string("op_2676_cast_fp16")];
            fp16 var_2677_to_fp16 = const()[name = string("op_2677_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2678_cast_fp16 = mul(x = var_2676_cast_fp16, y = var_2677_to_fp16)[name = string("op_2678_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2679_cast_fp16 = real_div(x = var_2678_cast_fp16, y = s_109_cast_fp16)[name = string("op_2679_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2680_cast_fp16 = add(x = var_2675_cast_fp16, y = var_2679_cast_fp16)[name = string("op_2680_cast_fp16")];
            fp32 var_2681_epsilon_0 = const()[name = string("op_2681_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2681_cast_fp16 = rsqrt(epsilon = var_2681_epsilon_0, x = var_2680_cast_fp16)[name = string("op_2681_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2682_cast_fp16 = mul(x = y_115_cast_fp16, y = var_2681_cast_fp16)[name = string("op_2682_cast_fp16")];
            tensor<fp16, [768]> core_backbone_layers_16_ffn_norm_weight_to_fp16 = const()[name = string("core_backbone_layers_16_ffn_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210792832)))];
            tensor<fp16, [1, 96, 768]> input_241_cast_fp16 = mul(x = core_backbone_layers_16_ffn_norm_weight_to_fp16, y = var_2682_cast_fp16)[name = string("input_241_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_16_feed_forward_w1_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w1_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210794432)))];
            tensor<fp16, [1, 96, 1792]> linear_96_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w1_weight_to_fp16, x = input_241_cast_fp16)[name = string("linear_96_cast_fp16")];
            tensor<fp16, [1792, 768]> core_backbone_layers_16_feed_forward_w3_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w3_weight_to_fp16"), val = tensor<fp16, [1792, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213547008)))];
            tensor<fp16, [1, 96, 1792]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w3_weight_to_fp16, x = input_241_cast_fp16)[name = string("linear_97_cast_fp16")];
            fp16 var_2691_to_fp16 = const()[name = string("op_2691_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 96, 1792]> var_2692_cast_fp16 = mul(x = linear_96_cast_fp16, y = var_2691_to_fp16)[name = string("op_2692_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2695_cast_fp16 = tanh(x = var_2692_cast_fp16)[name = string("op_2695_cast_fp16")];
            fp16 var_2696_to_fp16 = const()[name = string("op_2696_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1, 96, 1792]> var_2697_cast_fp16 = add(x = var_2695_cast_fp16, y = var_2696_to_fp16)[name = string("op_2697_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> var_2698_cast_fp16 = mul(x = var_2692_cast_fp16, y = var_2697_cast_fp16)[name = string("op_2698_cast_fp16")];
            tensor<fp16, [1, 96, 1792]> input_243_cast_fp16 = mul(x = var_2698_cast_fp16, y = linear_97_cast_fp16)[name = string("input_243_cast_fp16")];
            tensor<fp16, [768, 1792]> core_backbone_layers_16_feed_forward_w2_weight_to_fp16 = const()[name = string("core_backbone_layers_16_feed_forward_w2_weight_to_fp16"), val = tensor<fp16, [768, 1792]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216299584)))];
            tensor<fp16, [1, 96, 768]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = core_backbone_layers_16_feed_forward_w2_weight_to_fp16, x = input_243_cast_fp16)[name = string("linear_98_cast_fp16")];
            tensor<fp16, [1, 96, 768]> x_cast_fp16 = add(x = x_203_cast_fp16, y = linear_98_cast_fp16)[name = string("x_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2704_cast_fp16 = abs(x = x_cast_fp16)[name = string("op_2704_cast_fp16")];
            tensor<int32, [1]> var_2706_axes_0 = const()[name = string("op_2706_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1, 96, 1]> var_2706_cast_fp16 = reduce_max(axes = var_2706_axes_0, keep_dims = var_127, x = var_2704_cast_fp16)[name = string("op_2706_cast_fp16")];
            tensor<fp16, [1, 96, 1]> s_cast_fp16 = maximum(x = var_2706_cast_fp16, y = var_128_to_fp16)[name = string("s_cast_fp16")];
            tensor<fp16, [1, 96, 768]> y_cast_fp16 = real_div(x = x_cast_fp16, y = s_cast_fp16)[name = string("y_cast_fp16")];
            fp16 var_129_promoted_48_to_fp16 = const()[name = string("op_129_promoted_48_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 96, 768]> var_2709_cast_fp16 = pow(x = y_cast_fp16, y = var_129_promoted_48_to_fp16)[name = string("op_2709_cast_fp16")];
            tensor<int32, [1]> var_2711_axes_0 = const()[name = string("op_2711_axes_0"), val = tensor<int32, [1]>([-1])];
            bool var_2711_keep_dims_0 = const()[name = string("op_2711_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 96, 1]> var_2711_cast_fp16 = reduce_mean(axes = var_2711_axes_0, keep_dims = var_2711_keep_dims_0, x = var_2709_cast_fp16)[name = string("op_2711_cast_fp16")];
            fp32 var_2712_epsilon_0 = const()[name = string("op_2712_epsilon_0"), val = fp32(0x1.a36e2ep-14)];
            tensor<fp16, [1, 96, 1]> var_2712_cast_fp16 = inverse(epsilon = var_2712_epsilon_0, x = s_cast_fp16)[name = string("op_2712_cast_fp16")];
            fp16 var_2713_to_fp16 = const()[name = string("op_2713_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 1]> var_2714_cast_fp16 = mul(x = var_2712_cast_fp16, y = var_2713_to_fp16)[name = string("op_2714_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2715_cast_fp16 = real_div(x = var_2714_cast_fp16, y = s_cast_fp16)[name = string("op_2715_cast_fp16")];
            tensor<fp16, [1, 96, 1]> var_2716_cast_fp16 = add(x = var_2711_cast_fp16, y = var_2715_cast_fp16)[name = string("op_2716_cast_fp16")];
            fp32 var_2717_epsilon_0 = const()[name = string("op_2717_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 96, 1]> var_2717_cast_fp16 = rsqrt(epsilon = var_2717_epsilon_0, x = var_2716_cast_fp16)[name = string("op_2717_cast_fp16")];
            tensor<fp16, [1, 96, 768]> var_2718_cast_fp16 = mul(x = y_cast_fp16, y = var_2717_cast_fp16)[name = string("op_2718_cast_fp16")];
            tensor<fp16, [768]> core_backbone_embedding_norm_weight_to_fp16 = const()[name = string("core_backbone_embedding_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219052160)))];
            tensor<fp16, [1, 96, 768]> input_245_cast_fp16 = mul(x = core_backbone_embedding_norm_weight_to_fp16, y = var_2718_cast_fp16)[name = string("input_245_cast_fp16")];
            int32 var_2721_axis_0 = const()[name = string("op_2721_axis_0"), val = int32(0)];
            tensor<fp16, [10, 1, 768, 2]> conv_out = stack(axis = var_2721_axis_0, values = (var_752_cast_fp16, var_835_cast_fp16, var_1083_cast_fp16, var_1166_cast_fp16, var_1414_cast_fp16, var_1497_cast_fp16, var_1745_cast_fp16, var_1828_cast_fp16, var_2076_cast_fp16, var_2159_cast_fp16))[name = string("op_2721_cast_fp16")];
            int32 var_2723_axis_0 = const()[name = string("op_2723_axis_0"), val = int32(0)];
            tensor<fp16, [7, 1, 4, 31, 96]> k_out = stack(axis = var_2723_axis_0, values = (var_1008_cast_fp16, var_1339_cast_fp16, var_1670_cast_fp16, var_2001_cast_fp16, var_2332_cast_fp16, var_2497_cast_fp16, var_2662_cast_fp16))[name = string("op_2723_cast_fp16")];
            int32 var_2725_axis_0 = const()[name = string("op_2725_axis_0"), val = int32(0)];
            tensor<fp16, [7, 1, 4, 31, 96]> v_out = stack(axis = var_2725_axis_0, values = (var_1011_cast_fp16, var_1342_cast_fp16, var_1673_cast_fp16, var_2004_cast_fp16, var_2335_cast_fp16, var_2500_cast_fp16, var_2665_cast_fp16))[name = string("op_2725_cast_fp16")];
            tensor<int32, [1]> input_247_axes_0 = const()[name = string("input_247_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> projector_layer_norm_weight_to_fp16 = const()[name = string("projector_layer_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219053760)))];
            tensor<fp16, [768]> projector_layer_norm_bias_to_fp16 = const()[name = string("projector_layer_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219055360)))];
            fp16 var_2733_to_fp16 = const()[name = string("op_2733_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 96, 768]> input_247_cast_fp16 = layer_norm(axes = input_247_axes_0, beta = projector_layer_norm_bias_to_fp16, epsilon = var_2733_to_fp16, gamma = projector_layer_norm_weight_to_fp16, x = input_245_cast_fp16)[name = string("input_247_cast_fp16")];
            tensor<fp16, [2048, 768]> projector_linear_1_weight_to_fp16 = const()[name = string("projector_linear_1_weight_to_fp16"), val = tensor<fp16, [2048, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219056960)))];
            tensor<fp16, [2048]> projector_linear_1_bias_to_fp16 = const()[name = string("projector_linear_1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222202752)))];
            tensor<fp16, [1, 96, 2048]> linear_99_cast_fp16 = linear(bias = projector_linear_1_bias_to_fp16, weight = projector_linear_1_weight_to_fp16, x = input_247_cast_fp16)[name = string("linear_99_cast_fp16")];
            string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 96, 2048]> input_cast_fp16 = gelu(mode = input_mode_0, x = linear_99_cast_fp16)[name = string("input_cast_fp16")];
            tensor<fp16, [2048, 2048]> projector_linear_2_weight_to_fp16 = const()[name = string("projector_linear_2_weight_to_fp16"), val = tensor<fp16, [2048, 2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222206912)))];
            tensor<fp16, [2048]> projector_linear_2_bias_to_fp16 = const()[name = string("projector_linear_2_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230595584)))];
            tensor<fp16, [1, 96, 2048]> emb = linear(bias = projector_linear_2_bias_to_fp16, weight = projector_linear_2_weight_to_fp16, x = input_cast_fp16)[name = string("linear_100_cast_fp16")];
        } -> (emb, conv_out, k_out, v_out);
}