program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.2"}})]
{
    func main<ios18>(tensor<fp16, [1, 188, 1, 1]> key_bias, tensor<fp16, [1, 128, 1, 1500]> mel, tensor<fp16, [1, 1, 1, 188]> pad_mask) {
            tensor<int32, [4]> var_70 = const()[name = string("op_70"), val = tensor<int32, [4]>([1, 1, 128, 1500])];
            tensor<fp16, [1, 1, 128, 1500]> input_1_cast_fp16 = reshape(shape = var_70, x = mel)[name = string("input_1_cast_fp16")];
            string input_3_pad_type_0 = const()[name = string("input_3_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_3_pad_0 = const()[name = string("input_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_3_strides_0 = const()[name = string("input_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> input_3_dilations_0 = const()[name = string("input_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_3_groups_0 = const()[name = string("input_3_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1, 3, 3]> pre_encode_conv0_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1856))))[name = string("pre_encode_conv0_weight_to_fp16_palettized")];
            tensor<fp16, [256]> pre_encode_conv0_bias_to_fp16 = const()[name = string("pre_encode_conv0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3968)))];
            tensor<fp16, [1, 256, 64, 750]> input_3_cast_fp16 = conv(bias = pre_encode_conv0_bias_to_fp16, dilations = input_3_dilations_0, groups = input_3_groups_0, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = input_3_strides_0, weight = pre_encode_conv0_weight_to_fp16_palettized, x = input_1_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<fp16, [1, 256, 64, 750]> input_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string input_7_pad_type_0 = const()[name = string("input_7_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_7_pad_0 = const()[name = string("input_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_7_strides_0 = const()[name = string("input_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 input_7_groups_0 = const()[name = string("input_7_groups_0"), val = int32(256)];
            tensor<int32, [2]> input_7_dilations_0 = const()[name = string("input_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> pre_encode_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4544))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6336))))[name = string("pre_encode_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [256]> pre_encode_conv2_bias_to_fp16 = const()[name = string("pre_encode_conv2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8448)))];
            tensor<fp16, [1, 256, 32, 375]> input_7_cast_fp16 = conv(bias = pre_encode_conv2_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = pre_encode_conv2_weight_to_fp16_palettized, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
            string input_9_pad_type_0 = const()[name = string("input_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_9_strides_0 = const()[name = string("input_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_9_pad_0 = const()[name = string("input_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_9_dilations_0 = const()[name = string("input_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_9_groups_0 = const()[name = string("input_9_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> pre_encode_conv3_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9024))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58240))))[name = string("pre_encode_conv3_weight_to_fp16_palettized")];
            tensor<fp16, [256]> pre_encode_conv3_bias_to_fp16 = const()[name = string("pre_encode_conv3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60352)))];
            tensor<fp16, [1, 256, 32, 375]> input_9_cast_fp16 = conv(bias = pre_encode_conv3_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 = pre_encode_conv3_weight_to_fp16_palettized, x = input_7_cast_fp16)[name = string("input_9_cast_fp16")];
            tensor<fp16, [1, 256, 32, 375]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
            string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(256)];
            tensor<int32, [2]> input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> pre_encode_conv5_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60928))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62720))))[name = string("pre_encode_conv5_weight_to_fp16_palettized")];
            tensor<fp16, [256]> pre_encode_conv5_bias_to_fp16 = const()[name = string("pre_encode_conv5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64832)))];
            tensor<fp16, [1, 256, 16, 188]> input_13_cast_fp16 = conv(bias = pre_encode_conv5_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = pre_encode_conv5_weight_to_fp16_palettized, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")];
            string input_15_pad_type_0 = const()[name = string("input_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_15_strides_0 = const()[name = string("input_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_15_pad_0 = const()[name = string("input_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_15_dilations_0 = const()[name = string("input_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_15_groups_0 = const()[name = string("input_15_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> pre_encode_conv6_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65408))), lut = tensor<fp16, [16, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114624))))[name = string("pre_encode_conv6_weight_to_fp16_palettized")];
            tensor<fp16, [256]> pre_encode_conv6_bias_to_fp16 = const()[name = string("pre_encode_conv6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116736)))];
            tensor<fp16, [1, 256, 16, 188]> input_15_cast_fp16 = conv(bias = pre_encode_conv6_bias_to_fp16, dilations = input_15_dilations_0, groups = input_15_groups_0, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = input_15_strides_0, weight = pre_encode_conv6_weight_to_fp16_palettized, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
            tensor<fp16, [1, 256, 16, 188]> x_1_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("x_1_cast_fp16")];
            tensor<int32, [4]> var_118 = const()[name = string("op_118"), val = tensor<int32, [4]>([1, 4096, 1, 188])];
            tensor<fp16, [1, 4096, 1, 188]> input_17_cast_fp16 = reshape(shape = var_118, x = x_1_cast_fp16)[name = string("input_17_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, [1024, 4096, 1, 1]> pre_encode_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117312))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3263104))))[name = string("pre_encode_out_weight_to_fp16_palettized")];
            tensor<fp16, [1024]> pre_encode_out_bias_to_fp16 = const()[name = string("pre_encode_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3271360)))];
            tensor<fp16, [1, 1024, 1, 188]> x_3_cast_fp16 = conv(bias = pre_encode_out_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 = pre_encode_out_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = string("x_3_cast_fp16")];
            int32 var_143 = const()[name = string("op_143"), val = int32(1)];
            tensor<int32, [1]> var_170_axes_0 = const()[name = string("op_170_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_0_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_0_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_170_cast_fp16 = layer_norm(axes = var_170_axes_0, epsilon = layers_0_norm_feed_forward1_eps_scaled_to_fp16, x = x_3_cast_fp16)[name = string("op_170_cast_fp16")];
            tensor<fp16, [1024]> input_19_mean_0_to_fp16 = const()[name = string("input_19_mean_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3273472)))];
            tensor<fp16, [1024]> input_19_variance_0_to_fp16 = const()[name = string("input_19_variance_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3275584)))];
            tensor<fp16, [1024]> input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3277696)))];
            tensor<fp16, [1024]> input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3279808)))];
            fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_170_cast_fp16)[name = string("input_19_cast_fp16")];
            string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_0_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3281920))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6427712))))[name = string("layers_0_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_21_cast_fp16 = conv(dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_0_feed_forward1_linear1_weight_to_fp16_palettized, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_23_cast_fp16 = silu(x = input_21_cast_fp16)[name = string("input_23_cast_fp16")];
            string var_187_pad_type_0 = const()[name = string("op_187_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_187_strides_0 = const()[name = string("op_187_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_187_pad_0 = const()[name = string("op_187_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_187_dilations_0 = const()[name = string("op_187_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_187_groups_0 = const()[name = string("op_187_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_188_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6460544))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9606336))))[name = string("op_188_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_188_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_187_dilations_0, groups = var_187_groups_0, pad = var_187_pad_0, pad_type = var_187_pad_type_0, strides = var_187_strides_0, weight = op_188_weight_0_to_fp16_palettized, x = input_23_cast_fp16)[name = string("op_188_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_5_cast_fp16 = add(x = x_3_cast_fp16, y = var_188_cast_fp16)[name = string("x_5_cast_fp16")];
            tensor<int32, [1]> var_204_axes_0 = const()[name = string("op_204_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_0_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_0_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_204_cast_fp16 = layer_norm(axes = var_204_axes_0, epsilon = layers_0_norm_self_att_eps_scaled_to_fp16, x = x_5_cast_fp16)[name = string("op_204_cast_fp16")];
            tensor<fp16, [1024]> x_7_gamma_0_to_fp16 = const()[name = string("x_7_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9614592)))];
            tensor<fp16, [1024]> x_7_beta_0_to_fp16 = const()[name = string("x_7_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9616704)))];
            fp16 x_7_epsilon_0_to_fp16 = const()[name = string("x_7_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_7_cast_fp16 = batch_norm(beta = x_7_beta_0_to_fp16, epsilon = x_7_epsilon_0_to_fp16, gamma = x_7_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_204_cast_fp16)[name = string("x_7_cast_fp16")];
            string q_1_pad_type_0 = const()[name = string("q_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_1_strides_0 = const()[name = string("q_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_1_pad_0 = const()[name = string("q_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_1_dilations_0 = const()[name = string("q_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_1_groups_0 = const()[name = string("q_1_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9618816))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10405312))))[name = string("layers_0_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_1_cast_fp16 = conv(dilations = q_1_dilations_0, groups = q_1_groups_0, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = q_1_strides_0, weight = layers_0_self_attn_linear_q_weight_to_fp16_palettized, x = x_7_cast_fp16)[name = string("q_1_cast_fp16")];
            string k_1_pad_type_0 = const()[name = string("k_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_1_strides_0 = const()[name = string("k_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_1_pad_0 = const()[name = string("k_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_1_dilations_0 = const()[name = string("k_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_1_groups_0 = const()[name = string("k_1_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10413568))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11200064))))[name = string("layers_0_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_1_cast_fp16 = conv(dilations = k_1_dilations_0, groups = k_1_groups_0, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = k_1_strides_0, weight = layers_0_self_attn_linear_k_weight_to_fp16_palettized, x = x_7_cast_fp16)[name = string("k_1_cast_fp16")];
            string v_1_pad_type_0 = const()[name = string("v_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_1_strides_0 = const()[name = string("v_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_1_pad_0 = const()[name = string("v_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_1_dilations_0 = const()[name = string("v_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_1_groups_0 = const()[name = string("v_1_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11208320))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11994816))))[name = string("layers_0_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_1_cast_fp16 = conv(dilations = v_1_dilations_0, groups = v_1_groups_0, pad = v_1_pad_0, pad_type = v_1_pad_type_0, strides = v_1_strides_0, weight = layers_0_self_attn_linear_v_weight_to_fp16_palettized, x = x_7_cast_fp16)[name = string("v_1_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_1_to_fp16 = const()[name = string("bv_all_1_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12003072)))];
            tensor<fp16, [1, 1024, 1, 188]> var_236_cast_fp16 = add(x = q_1_cast_fp16, y = bv_all_1_to_fp16)[name = string("op_236_cast_fp16")];
            tensor<int32, [3]> var_237 = const()[name = string("op_237"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_1_cast_fp16 = reshape(shape = var_237, x = var_236_cast_fp16)[name = string("qb_1_cast_fp16")];
            bool bd_all_1_transpose_x_0 = const()[name = string("bd_all_1_transpose_x_0"), val = bool(false)];
            bool bd_all_1_transpose_y_0 = const()[name = string("bd_all_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_0_self_attn_pos_proj_to_fp16 = const()[name = string("layers_0_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12005184)))];
            tensor<fp16, [8, 375, 188]> bd_all_1_cast_fp16 = matmul(transpose_x = bd_all_1_transpose_x_0, transpose_y = bd_all_1_transpose_y_0, x = layers_0_self_attn_pos_proj_to_fp16, y = qb_1_cast_fp16)[name = string("bd_all_1_cast_fp16")];
            tensor<int32, [3]> x_9_perm_0 = const()[name = string("x_9_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_11_pad_0 = const()[name = string("x_11_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_11_mode_0 = const()[name = string("x_11_mode_0"), val = string("constant")];
            fp16 const_10_to_fp16 = const()[name = string("const_10_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_9_cast_fp16 = transpose(perm = x_9_perm_0, x = bd_all_1_cast_fp16)[name = string("transpose_143")];
            tensor<fp16, [8, 188, 376]> x_11_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = x_11_mode_0, pad = x_11_pad_0, x = x_9_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<int32, [3]> var_244 = const()[name = string("op_244"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_13_cast_fp16 = reshape(shape = var_244, x = x_11_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<int32, [3]> var_247_begin_0 = const()[name = string("op_247_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_247_end_0 = const()[name = string("op_247_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_247_end_mask_0 = const()[name = string("op_247_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_247_cast_fp16 = slice_by_index(begin = var_247_begin_0, end = var_247_end_0, end_mask = var_247_end_mask_0, x = x_13_cast_fp16)[name = string("op_247_cast_fp16")];
            tensor<int32, [3]> var_248 = const()[name = string("op_248"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_15_cast_fp16 = reshape(shape = var_248, x = var_247_cast_fp16)[name = string("x_15_cast_fp16")];
            tensor<int32, [3]> bd_all_3_begin_0 = const()[name = string("bd_all_3_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_3_end_0 = const()[name = string("bd_all_3_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_3_end_mask_0 = const()[name = string("bd_all_3_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_3_cast_fp16 = slice_by_index(begin = bd_all_3_begin_0, end = bd_all_3_end_0, end_mask = bd_all_3_end_mask_0, x = x_15_cast_fp16)[name = string("bd_all_3_cast_fp16")];
            tensor<int32, [4]> var_253 = const()[name = string("op_253"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_254_cast_fp16 = reshape(shape = var_253, x = q_1_cast_fp16)[name = string("op_254_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_256_to_fp16 = const()[name = string("op_256_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12773248)))];
            tensor<fp16, [8, 128, 1, 188]> var_257_cast_fp16 = add(x = var_254_cast_fp16, y = var_256_to_fp16)[name = string("op_257_cast_fp16")];
            tensor<int32, [4]> var_258 = const()[name = string("op_258"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_1_cast_fp16 = reshape(shape = var_258, x = k_1_cast_fp16)[name = string("kh_1_cast_fp16")];
            tensor<int32, [4]> var_260 = const()[name = string("op_260"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_1_cast_fp16 = reshape(shape = var_260, x = v_1_cast_fp16)[name = string("vh_1_cast_fp16")];
            tensor<int32, [4]> var_262 = const()[name = string("op_262"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_1_equation_0 = const()[name = string("ac_1_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_263_cast_fp16 = transpose(perm = var_262, x = kh_1_cast_fp16)[name = string("transpose_142")];
            tensor<fp16, [8, 188, 1, 188]> ac_1_cast_fp16 = einsum(equation = ac_1_equation_0, values = (var_263_cast_fp16, var_257_cast_fp16))[name = string("ac_1_cast_fp16")];
            tensor<int32, [3]> var_266_perm_0 = const()[name = string("op_266_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_267_axes_0 = const()[name = string("op_267_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_266_cast_fp16 = transpose(perm = var_266_perm_0, x = bd_all_3_cast_fp16)[name = string("transpose_141")];
            tensor<fp16, [8, 188, 1, 188]> var_267_cast_fp16 = expand_dims(axes = var_267_axes_0, x = var_266_cast_fp16)[name = string("op_267_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_268_cast_fp16 = add(x = ac_1_cast_fp16, y = var_267_cast_fp16)[name = string("op_268_cast_fp16")];
            fp16 var_269_to_fp16 = const()[name = string("op_269_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_1_cast_fp16 = mul(x = var_268_cast_fp16, y = var_269_to_fp16)[name = string("scores_1_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_3_cast_fp16 = add(x = scores_1_cast_fp16, y = key_bias)[name = string("scores_3_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_272_cast_fp16 = softmax(axis = var_143, x = scores_3_cast_fp16)[name = string("op_272_cast_fp16")];
            tensor<int32, [4]> transpose_48_perm_0 = const()[name = string("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_0_perm_0 = const()[name = string("transpose_0_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_4 = const()[name = string("concat_4"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_0_cast_fp16 = transpose(perm = transpose_0_perm_0, x = var_272_cast_fp16)[name = string("transpose_140")];
            tensor<fp16, [8, 188, 188]> reshape_0_cast_fp16 = reshape(shape = concat_4, x = transpose_0_cast_fp16)[name = string("reshape_0_cast_fp16")];
            tensor<int32, [3]> concat_5 = const()[name = string("concat_5"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_48_cast_fp16 = transpose(perm = transpose_48_perm_0, x = vh_1_cast_fp16)[name = string("transpose_139")];
            tensor<fp16, [8, 188, 128]> reshape_1_cast_fp16 = reshape(shape = concat_5, x = transpose_48_cast_fp16)[name = string("reshape_1_cast_fp16")];
            bool matmul_0_transpose_x_0 = const()[name = string("matmul_0_transpose_x_0"), val = bool(false)];
            bool matmul_0_transpose_y_0 = const()[name = string("matmul_0_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_0_cast_fp16 = matmul(transpose_x = matmul_0_transpose_x_0, transpose_y = matmul_0_transpose_y_0, x = reshape_0_cast_fp16, y = reshape_1_cast_fp16)[name = string("matmul_0_cast_fp16")];
            tensor<int32, [4]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_2_cast_fp16 = reshape(shape = concat_9, x = matmul_0_cast_fp16)[name = string("reshape_2_cast_fp16")];
            tensor<int32, [4]> ctx_1_perm_0 = const()[name = string("ctx_1_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_277 = const()[name = string("op_277"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_1_cast_fp16 = transpose(perm = ctx_1_perm_0, x = reshape_2_cast_fp16)[name = string("transpose_138")];
            tensor<fp16, [1, 1024, 1, 188]> input_25_cast_fp16 = reshape(shape = var_277, x = ctx_1_cast_fp16)[name = string("input_25_cast_fp16")];
            string var_284_pad_type_0 = const()[name = string("op_284_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_284_strides_0 = const()[name = string("op_284_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_284_pad_0 = const()[name = string("op_284_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_284_dilations_0 = const()[name = string("op_284_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_284_groups_0 = const()[name = string("op_284_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12775360))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13561856))))[name = string("layers_0_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_284_cast_fp16 = conv(dilations = var_284_dilations_0, groups = var_284_groups_0, pad = var_284_pad_0, pad_type = var_284_pad_type_0, strides = var_284_strides_0, weight = layers_0_self_attn_linear_out_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = string("op_284_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_17_cast_fp16 = add(x = x_5_cast_fp16, y = var_284_cast_fp16)[name = string("x_17_cast_fp16")];
            tensor<int32, [1]> var_300_axes_0 = const()[name = string("op_300_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_0_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_0_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_300_cast_fp16 = layer_norm(axes = var_300_axes_0, epsilon = layers_0_norm_conv_eps_scaled_to_fp16, x = x_17_cast_fp16)[name = string("op_300_cast_fp16")];
            tensor<fp16, [1024]> input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13570112)))];
            tensor<fp16, [1024]> input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13572224)))];
            fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_300_cast_fp16)[name = string("input_27_cast_fp16")];
            string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_0_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13574336))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15147264))))[name = string("layers_0_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_29_cast_fp16 = conv(dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_0_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")];
            int32 x_19_split_num_splits_0 = const()[name = string("x_19_split_num_splits_0"), val = int32(2)];
            int32 x_19_split_axis_0 = const()[name = string("x_19_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_29_cast_fp16)[name = string("x_19_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_19_split_1_sigmoid_cast_fp16 = sigmoid(x = x_19_split_cast_fp16_1)[name = string("x_19_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_19_cast_fp16 = mul(x = x_19_split_cast_fp16_0, y = x_19_split_1_sigmoid_cast_fp16)[name = string("x_19_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_31_cast_fp16 = mul(x = x_19_cast_fp16, y = pad_mask)[name = string("input_31_cast_fp16")];
            string input_33_pad_type_0 = const()[name = string("input_33_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_33_pad_0 = const()[name = string("input_33_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_33_groups_0 = const()[name = string("input_33_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_33_strides_0 = const()[name = string("input_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_33_dilations_0 = const()[name = string("input_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_103_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15163712))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15170688))))[name = string("const_103_to_fp16_palettized")];
            tensor<fp16, [1024]> const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15178944)))];
            tensor<fp16, [1, 1024, 1, 188]> input_35_cast_fp16 = conv(bias = const_104_to_fp16, dilations = input_33_dilations_0, groups = input_33_groups_0, pad = input_33_pad_0, pad_type = input_33_pad_type_0, strides = input_33_strides_0, weight = const_103_to_fp16_palettized, x = input_31_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_37_cast_fp16 = silu(x = input_35_cast_fp16)[name = string("input_37_cast_fp16")];
            string var_332_pad_type_0 = const()[name = string("op_332_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_332_strides_0 = const()[name = string("op_332_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_332_pad_0 = const()[name = string("op_332_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_332_dilations_0 = const()[name = string("op_332_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_332_groups_0 = const()[name = string("op_332_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_0_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15181056))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15967552))))[name = string("layers_0_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_332_cast_fp16 = conv(dilations = var_332_dilations_0, groups = var_332_groups_0, pad = var_332_pad_0, pad_type = var_332_pad_type_0, strides = var_332_strides_0, weight = layers_0_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = string("op_332_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_21_cast_fp16 = add(x = x_17_cast_fp16, y = var_332_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<int32, [1]> var_348_axes_0 = const()[name = string("op_348_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_0_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_0_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_348_cast_fp16 = layer_norm(axes = var_348_axes_0, epsilon = layers_0_norm_feed_forward2_eps_scaled_to_fp16, x = x_21_cast_fp16)[name = string("op_348_cast_fp16")];
            tensor<fp16, [1024]> input_39_gamma_0_to_fp16 = const()[name = string("input_39_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15975808)))];
            tensor<fp16, [1024]> input_39_beta_0_to_fp16 = const()[name = string("input_39_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15977920)))];
            fp16 input_39_epsilon_0_to_fp16 = const()[name = string("input_39_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_39_cast_fp16 = batch_norm(beta = input_39_beta_0_to_fp16, epsilon = input_39_epsilon_0_to_fp16, gamma = input_39_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_348_cast_fp16)[name = string("input_39_cast_fp16")];
            string input_41_pad_type_0 = const()[name = string("input_41_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_41_strides_0 = const()[name = string("input_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_41_pad_0 = const()[name = string("input_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_41_dilations_0 = const()[name = string("input_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_41_groups_0 = const()[name = string("input_41_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_0_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15980032))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19125824))))[name = string("layers_0_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_41_cast_fp16 = conv(dilations = input_41_dilations_0, groups = input_41_groups_0, pad = input_41_pad_0, pad_type = input_41_pad_type_0, strides = input_41_strides_0, weight = layers_0_feed_forward2_linear1_weight_to_fp16_palettized, x = input_39_cast_fp16)[name = string("input_41_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_43_cast_fp16 = silu(x = input_41_cast_fp16)[name = string("input_43_cast_fp16")];
            string var_365_pad_type_0 = const()[name = string("op_365_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_365_strides_0 = const()[name = string("op_365_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_365_pad_0 = const()[name = string("op_365_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_365_dilations_0 = const()[name = string("op_365_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_365_groups_0 = const()[name = string("op_365_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_366_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19158656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22304448))))[name = string("op_366_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_366_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_365_dilations_0, groups = var_365_groups_0, pad = var_365_pad_0, pad_type = var_365_pad_type_0, strides = var_365_strides_0, weight = op_366_weight_0_to_fp16_palettized, x = input_43_cast_fp16)[name = string("op_366_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_23_cast_fp16 = add(x = x_21_cast_fp16, y = var_366_cast_fp16)[name = string("x_23_cast_fp16")];
            tensor<int32, [1]> var_382_axes_0 = const()[name = string("op_382_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_0_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_0_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_382_cast_fp16 = layer_norm(axes = var_382_axes_0, epsilon = layers_0_norm_out_eps_scaled_to_fp16, x = x_23_cast_fp16)[name = string("op_382_cast_fp16")];
            tensor<fp16, [1024]> x_25_gamma_0_to_fp16 = const()[name = string("x_25_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22312704)))];
            tensor<fp16, [1024]> x_25_beta_0_to_fp16 = const()[name = string("x_25_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22314816)))];
            fp16 x_25_epsilon_0_to_fp16 = const()[name = string("x_25_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_25_cast_fp16 = batch_norm(beta = x_25_beta_0_to_fp16, epsilon = x_25_epsilon_0_to_fp16, gamma = x_25_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_382_cast_fp16)[name = string("x_25_cast_fp16")];
            int32 var_401 = const()[name = string("op_401"), val = int32(1)];
            tensor<int32, [1]> var_428_axes_0 = const()[name = string("op_428_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_1_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_1_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_428_cast_fp16 = layer_norm(axes = var_428_axes_0, epsilon = layers_1_norm_feed_forward1_eps_scaled_to_fp16, x = x_25_cast_fp16)[name = string("op_428_cast_fp16")];
            tensor<fp16, [1024]> input_45_gamma_0_to_fp16 = const()[name = string("input_45_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22316928)))];
            tensor<fp16, [1024]> input_45_beta_0_to_fp16 = const()[name = string("input_45_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22319040)))];
            fp16 input_45_epsilon_0_to_fp16 = const()[name = string("input_45_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_45_cast_fp16 = batch_norm(beta = input_45_beta_0_to_fp16, epsilon = input_45_epsilon_0_to_fp16, gamma = input_45_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_428_cast_fp16)[name = string("input_45_cast_fp16")];
            string input_47_pad_type_0 = const()[name = string("input_47_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_47_strides_0 = const()[name = string("input_47_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_47_dilations_0 = const()[name = string("input_47_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_47_groups_0 = const()[name = string("input_47_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_1_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22321152))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25466944))))[name = string("layers_1_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_47_cast_fp16 = conv(dilations = input_47_dilations_0, groups = input_47_groups_0, pad = input_47_pad_0, pad_type = input_47_pad_type_0, strides = input_47_strides_0, weight = layers_1_feed_forward1_linear1_weight_to_fp16_palettized, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_49_cast_fp16 = silu(x = input_47_cast_fp16)[name = string("input_49_cast_fp16")];
            string var_445_pad_type_0 = const()[name = string("op_445_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_445_strides_0 = const()[name = string("op_445_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_445_pad_0 = const()[name = string("op_445_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_445_dilations_0 = const()[name = string("op_445_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_445_groups_0 = const()[name = string("op_445_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_446_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25499776))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28645568))))[name = string("op_446_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_446_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_445_dilations_0, groups = var_445_groups_0, pad = var_445_pad_0, pad_type = var_445_pad_type_0, strides = var_445_strides_0, weight = op_446_weight_0_to_fp16_palettized, x = input_49_cast_fp16)[name = string("op_446_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_27_cast_fp16 = add(x = x_25_cast_fp16, y = var_446_cast_fp16)[name = string("x_27_cast_fp16")];
            tensor<int32, [1]> var_462_axes_0 = const()[name = string("op_462_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_1_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_1_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_462_cast_fp16 = layer_norm(axes = var_462_axes_0, epsilon = layers_1_norm_self_att_eps_scaled_to_fp16, x = x_27_cast_fp16)[name = string("op_462_cast_fp16")];
            tensor<fp16, [1024]> x_29_gamma_0_to_fp16 = const()[name = string("x_29_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28653824)))];
            tensor<fp16, [1024]> x_29_beta_0_to_fp16 = const()[name = string("x_29_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28655936)))];
            fp16 x_29_epsilon_0_to_fp16 = const()[name = string("x_29_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_29_cast_fp16 = batch_norm(beta = x_29_beta_0_to_fp16, epsilon = x_29_epsilon_0_to_fp16, gamma = x_29_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_462_cast_fp16)[name = string("x_29_cast_fp16")];
            string q_3_pad_type_0 = const()[name = string("q_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_3_strides_0 = const()[name = string("q_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_3_pad_0 = const()[name = string("q_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_3_dilations_0 = const()[name = string("q_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_3_groups_0 = const()[name = string("q_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28658048))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29444544))))[name = string("layers_1_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_3_cast_fp16 = conv(dilations = q_3_dilations_0, groups = q_3_groups_0, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = q_3_strides_0, weight = layers_1_self_attn_linear_q_weight_to_fp16_palettized, x = x_29_cast_fp16)[name = string("q_3_cast_fp16")];
            string k_3_pad_type_0 = const()[name = string("k_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_3_strides_0 = const()[name = string("k_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_3_pad_0 = const()[name = string("k_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_3_dilations_0 = const()[name = string("k_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_3_groups_0 = const()[name = string("k_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29452800))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30239296))))[name = string("layers_1_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_3_cast_fp16 = conv(dilations = k_3_dilations_0, groups = k_3_groups_0, pad = k_3_pad_0, pad_type = k_3_pad_type_0, strides = k_3_strides_0, weight = layers_1_self_attn_linear_k_weight_to_fp16_palettized, x = x_29_cast_fp16)[name = string("k_3_cast_fp16")];
            string v_3_pad_type_0 = const()[name = string("v_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_3_strides_0 = const()[name = string("v_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_3_pad_0 = const()[name = string("v_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_3_dilations_0 = const()[name = string("v_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_3_groups_0 = const()[name = string("v_3_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30247552))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31034048))))[name = string("layers_1_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_3_cast_fp16 = conv(dilations = v_3_dilations_0, groups = v_3_groups_0, pad = v_3_pad_0, pad_type = v_3_pad_type_0, strides = v_3_strides_0, weight = layers_1_self_attn_linear_v_weight_to_fp16_palettized, x = x_29_cast_fp16)[name = string("v_3_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_3_to_fp16 = const()[name = string("bv_all_3_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31042304)))];
            tensor<fp16, [1, 1024, 1, 188]> var_494_cast_fp16 = add(x = q_3_cast_fp16, y = bv_all_3_to_fp16)[name = string("op_494_cast_fp16")];
            tensor<int32, [3]> var_495 = const()[name = string("op_495"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_3_cast_fp16 = reshape(shape = var_495, x = var_494_cast_fp16)[name = string("qb_3_cast_fp16")];
            bool bd_all_5_transpose_x_0 = const()[name = string("bd_all_5_transpose_x_0"), val = bool(false)];
            bool bd_all_5_transpose_y_0 = const()[name = string("bd_all_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_1_self_attn_pos_proj_to_fp16 = const()[name = string("layers_1_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31044416)))];
            tensor<fp16, [8, 375, 188]> bd_all_5_cast_fp16 = matmul(transpose_x = bd_all_5_transpose_x_0, transpose_y = bd_all_5_transpose_y_0, x = layers_1_self_attn_pos_proj_to_fp16, y = qb_3_cast_fp16)[name = string("bd_all_5_cast_fp16")];
            tensor<int32, [3]> x_31_perm_0 = const()[name = string("x_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_33_pad_0 = const()[name = string("x_33_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_33_mode_0 = const()[name = string("x_33_mode_0"), val = string("constant")];
            fp16 const_14_to_fp16 = const()[name = string("const_14_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_31_cast_fp16 = transpose(perm = x_31_perm_0, x = bd_all_5_cast_fp16)[name = string("transpose_137")];
            tensor<fp16, [8, 188, 376]> x_33_cast_fp16 = pad(constant_val = const_14_to_fp16, mode = x_33_mode_0, pad = x_33_pad_0, x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<int32, [3]> var_502 = const()[name = string("op_502"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_35_cast_fp16 = reshape(shape = var_502, x = x_33_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<int32, [3]> var_505_begin_0 = const()[name = string("op_505_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_505_end_0 = const()[name = string("op_505_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_505_end_mask_0 = const()[name = string("op_505_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_505_cast_fp16 = slice_by_index(begin = var_505_begin_0, end = var_505_end_0, end_mask = var_505_end_mask_0, x = x_35_cast_fp16)[name = string("op_505_cast_fp16")];
            tensor<int32, [3]> var_506 = const()[name = string("op_506"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_37_cast_fp16 = reshape(shape = var_506, x = var_505_cast_fp16)[name = string("x_37_cast_fp16")];
            tensor<int32, [3]> bd_all_7_begin_0 = const()[name = string("bd_all_7_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_7_end_0 = const()[name = string("bd_all_7_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_7_end_mask_0 = const()[name = string("bd_all_7_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_7_cast_fp16 = slice_by_index(begin = bd_all_7_begin_0, end = bd_all_7_end_0, end_mask = bd_all_7_end_mask_0, x = x_37_cast_fp16)[name = string("bd_all_7_cast_fp16")];
            tensor<int32, [4]> var_511 = const()[name = string("op_511"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_512_cast_fp16 = reshape(shape = var_511, x = q_3_cast_fp16)[name = string("op_512_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_514_to_fp16 = const()[name = string("op_514_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31812480)))];
            tensor<fp16, [8, 128, 1, 188]> var_515_cast_fp16 = add(x = var_512_cast_fp16, y = var_514_to_fp16)[name = string("op_515_cast_fp16")];
            tensor<int32, [4]> var_516 = const()[name = string("op_516"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_3_cast_fp16 = reshape(shape = var_516, x = k_3_cast_fp16)[name = string("kh_3_cast_fp16")];
            tensor<int32, [4]> var_518 = const()[name = string("op_518"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_3_cast_fp16 = reshape(shape = var_518, x = v_3_cast_fp16)[name = string("vh_3_cast_fp16")];
            tensor<int32, [4]> var_520 = const()[name = string("op_520"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_3_equation_0 = const()[name = string("ac_3_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_521_cast_fp16 = transpose(perm = var_520, x = kh_3_cast_fp16)[name = string("transpose_136")];
            tensor<fp16, [8, 188, 1, 188]> ac_3_cast_fp16 = einsum(equation = ac_3_equation_0, values = (var_521_cast_fp16, var_515_cast_fp16))[name = string("ac_3_cast_fp16")];
            tensor<int32, [3]> var_524_perm_0 = const()[name = string("op_524_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_525_axes_0 = const()[name = string("op_525_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_524_cast_fp16 = transpose(perm = var_524_perm_0, x = bd_all_7_cast_fp16)[name = string("transpose_135")];
            tensor<fp16, [8, 188, 1, 188]> var_525_cast_fp16 = expand_dims(axes = var_525_axes_0, x = var_524_cast_fp16)[name = string("op_525_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_526_cast_fp16 = add(x = ac_3_cast_fp16, y = var_525_cast_fp16)[name = string("op_526_cast_fp16")];
            fp16 var_527_to_fp16 = const()[name = string("op_527_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_5_cast_fp16 = mul(x = var_526_cast_fp16, y = var_527_to_fp16)[name = string("scores_5_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_7_cast_fp16 = add(x = scores_5_cast_fp16, y = key_bias)[name = string("scores_7_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_530_cast_fp16 = softmax(axis = var_401, x = scores_7_cast_fp16)[name = string("op_530_cast_fp16")];
            tensor<int32, [4]> transpose_49_perm_0 = const()[name = string("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_2_perm_0 = const()[name = string("transpose_2_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_14 = const()[name = string("concat_14"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = var_530_cast_fp16)[name = string("transpose_134")];
            tensor<fp16, [8, 188, 188]> reshape_3_cast_fp16 = reshape(shape = concat_14, x = transpose_2_cast_fp16)[name = string("reshape_3_cast_fp16")];
            tensor<int32, [3]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_49_cast_fp16 = transpose(perm = transpose_49_perm_0, x = vh_3_cast_fp16)[name = string("transpose_133")];
            tensor<fp16, [8, 188, 128]> reshape_4_cast_fp16 = reshape(shape = concat_15, x = transpose_49_cast_fp16)[name = string("reshape_4_cast_fp16")];
            bool matmul_1_transpose_x_0 = const()[name = string("matmul_1_transpose_x_0"), val = bool(false)];
            bool matmul_1_transpose_y_0 = const()[name = string("matmul_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = reshape_3_cast_fp16, y = reshape_4_cast_fp16)[name = string("matmul_1_cast_fp16")];
            tensor<int32, [4]> concat_19 = const()[name = string("concat_19"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_5_cast_fp16 = reshape(shape = concat_19, x = matmul_1_cast_fp16)[name = string("reshape_5_cast_fp16")];
            tensor<int32, [4]> ctx_3_perm_0 = const()[name = string("ctx_3_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_535 = const()[name = string("op_535"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_3_cast_fp16 = transpose(perm = ctx_3_perm_0, x = reshape_5_cast_fp16)[name = string("transpose_132")];
            tensor<fp16, [1, 1024, 1, 188]> input_51_cast_fp16 = reshape(shape = var_535, x = ctx_3_cast_fp16)[name = string("input_51_cast_fp16")];
            string var_542_pad_type_0 = const()[name = string("op_542_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_542_strides_0 = const()[name = string("op_542_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_542_pad_0 = const()[name = string("op_542_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_542_dilations_0 = const()[name = string("op_542_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_542_groups_0 = const()[name = string("op_542_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31814592))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32601088))))[name = string("layers_1_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_542_cast_fp16 = conv(dilations = var_542_dilations_0, groups = var_542_groups_0, pad = var_542_pad_0, pad_type = var_542_pad_type_0, strides = var_542_strides_0, weight = layers_1_self_attn_linear_out_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = string("op_542_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_39_cast_fp16 = add(x = x_27_cast_fp16, y = var_542_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<int32, [1]> var_558_axes_0 = const()[name = string("op_558_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_1_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_1_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_558_cast_fp16 = layer_norm(axes = var_558_axes_0, epsilon = layers_1_norm_conv_eps_scaled_to_fp16, x = x_39_cast_fp16)[name = string("op_558_cast_fp16")];
            tensor<fp16, [1024]> input_53_gamma_0_to_fp16 = const()[name = string("input_53_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32609344)))];
            tensor<fp16, [1024]> input_53_beta_0_to_fp16 = const()[name = string("input_53_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32611456)))];
            fp16 input_53_epsilon_0_to_fp16 = const()[name = string("input_53_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_53_cast_fp16 = batch_norm(beta = input_53_beta_0_to_fp16, epsilon = input_53_epsilon_0_to_fp16, gamma = input_53_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_558_cast_fp16)[name = string("input_53_cast_fp16")];
            string input_55_pad_type_0 = const()[name = string("input_55_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_55_strides_0 = const()[name = string("input_55_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_55_pad_0 = const()[name = string("input_55_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_55_dilations_0 = const()[name = string("input_55_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_55_groups_0 = const()[name = string("input_55_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_1_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32613568))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34186496))))[name = string("layers_1_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_55_cast_fp16 = conv(dilations = input_55_dilations_0, groups = input_55_groups_0, pad = input_55_pad_0, pad_type = input_55_pad_type_0, strides = input_55_strides_0, weight = layers_1_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_53_cast_fp16)[name = string("input_55_cast_fp16")];
            int32 x_41_split_num_splits_0 = const()[name = string("x_41_split_num_splits_0"), val = int32(2)];
            int32 x_41_split_axis_0 = const()[name = string("x_41_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_41_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_55_cast_fp16)[name = string("x_41_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_41_split_1_sigmoid_cast_fp16 = sigmoid(x = x_41_split_cast_fp16_1)[name = string("x_41_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_41_cast_fp16 = mul(x = x_41_split_cast_fp16_0, y = x_41_split_1_sigmoid_cast_fp16)[name = string("x_41_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_57_cast_fp16 = mul(x = x_41_cast_fp16, y = pad_mask)[name = string("input_57_cast_fp16")];
            string input_59_pad_type_0 = const()[name = string("input_59_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_59_pad_0 = const()[name = string("input_59_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_59_groups_0 = const()[name = string("input_59_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_59_strides_0 = const()[name = string("input_59_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_59_dilations_0 = const()[name = string("input_59_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_105_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34202944))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34209920))))[name = string("const_105_to_fp16_palettized")];
            tensor<fp16, [1024]> const_106_to_fp16 = const()[name = string("const_106_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34218176)))];
            tensor<fp16, [1, 1024, 1, 188]> input_61_cast_fp16 = conv(bias = const_106_to_fp16, dilations = input_59_dilations_0, groups = input_59_groups_0, pad = input_59_pad_0, pad_type = input_59_pad_type_0, strides = input_59_strides_0, weight = const_105_to_fp16_palettized, x = input_57_cast_fp16)[name = string("input_61_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_63_cast_fp16 = silu(x = input_61_cast_fp16)[name = string("input_63_cast_fp16")];
            string var_590_pad_type_0 = const()[name = string("op_590_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_590_strides_0 = const()[name = string("op_590_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_590_pad_0 = const()[name = string("op_590_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_590_dilations_0 = const()[name = string("op_590_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_590_groups_0 = const()[name = string("op_590_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_1_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34220288))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35006784))))[name = string("layers_1_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_590_cast_fp16 = conv(dilations = var_590_dilations_0, groups = var_590_groups_0, pad = var_590_pad_0, pad_type = var_590_pad_type_0, strides = var_590_strides_0, weight = layers_1_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = string("op_590_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_43_cast_fp16 = add(x = x_39_cast_fp16, y = var_590_cast_fp16)[name = string("x_43_cast_fp16")];
            tensor<int32, [1]> var_606_axes_0 = const()[name = string("op_606_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_1_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_1_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_606_cast_fp16 = layer_norm(axes = var_606_axes_0, epsilon = layers_1_norm_feed_forward2_eps_scaled_to_fp16, x = x_43_cast_fp16)[name = string("op_606_cast_fp16")];
            tensor<fp16, [1024]> input_65_gamma_0_to_fp16 = const()[name = string("input_65_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35015040)))];
            tensor<fp16, [1024]> input_65_beta_0_to_fp16 = const()[name = string("input_65_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35017152)))];
            fp16 input_65_epsilon_0_to_fp16 = const()[name = string("input_65_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_65_cast_fp16 = batch_norm(beta = input_65_beta_0_to_fp16, epsilon = input_65_epsilon_0_to_fp16, gamma = input_65_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_606_cast_fp16)[name = string("input_65_cast_fp16")];
            string input_67_pad_type_0 = const()[name = string("input_67_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_67_strides_0 = const()[name = string("input_67_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_67_pad_0 = const()[name = string("input_67_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_67_dilations_0 = const()[name = string("input_67_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_67_groups_0 = const()[name = string("input_67_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_1_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35019264))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38165056))))[name = string("layers_1_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_67_cast_fp16 = conv(dilations = input_67_dilations_0, groups = input_67_groups_0, pad = input_67_pad_0, pad_type = input_67_pad_type_0, strides = input_67_strides_0, weight = layers_1_feed_forward2_linear1_weight_to_fp16_palettized, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_69_cast_fp16 = silu(x = input_67_cast_fp16)[name = string("input_69_cast_fp16")];
            string var_623_pad_type_0 = const()[name = string("op_623_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_623_strides_0 = const()[name = string("op_623_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_623_pad_0 = const()[name = string("op_623_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_623_dilations_0 = const()[name = string("op_623_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_623_groups_0 = const()[name = string("op_623_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_624_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38197888))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41343680))))[name = string("op_624_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_624_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_623_dilations_0, groups = var_623_groups_0, pad = var_623_pad_0, pad_type = var_623_pad_type_0, strides = var_623_strides_0, weight = op_624_weight_0_to_fp16_palettized, x = input_69_cast_fp16)[name = string("op_624_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_45_cast_fp16 = add(x = x_43_cast_fp16, y = var_624_cast_fp16)[name = string("x_45_cast_fp16")];
            tensor<int32, [1]> var_640_axes_0 = const()[name = string("op_640_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_1_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_1_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_640_cast_fp16 = layer_norm(axes = var_640_axes_0, epsilon = layers_1_norm_out_eps_scaled_to_fp16, x = x_45_cast_fp16)[name = string("op_640_cast_fp16")];
            tensor<fp16, [1024]> x_47_gamma_0_to_fp16 = const()[name = string("x_47_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41351936)))];
            tensor<fp16, [1024]> x_47_beta_0_to_fp16 = const()[name = string("x_47_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41354048)))];
            fp16 x_47_epsilon_0_to_fp16 = const()[name = string("x_47_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_47_cast_fp16 = batch_norm(beta = x_47_beta_0_to_fp16, epsilon = x_47_epsilon_0_to_fp16, gamma = x_47_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_640_cast_fp16)[name = string("x_47_cast_fp16")];
            int32 var_659 = const()[name = string("op_659"), val = int32(1)];
            tensor<int32, [1]> var_686_axes_0 = const()[name = string("op_686_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_2_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_2_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_686_cast_fp16 = layer_norm(axes = var_686_axes_0, epsilon = layers_2_norm_feed_forward1_eps_scaled_to_fp16, x = x_47_cast_fp16)[name = string("op_686_cast_fp16")];
            tensor<fp16, [1024]> input_71_gamma_0_to_fp16 = const()[name = string("input_71_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41356160)))];
            tensor<fp16, [1024]> input_71_beta_0_to_fp16 = const()[name = string("input_71_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41358272)))];
            fp16 input_71_epsilon_0_to_fp16 = const()[name = string("input_71_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_71_cast_fp16 = batch_norm(beta = input_71_beta_0_to_fp16, epsilon = input_71_epsilon_0_to_fp16, gamma = input_71_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_686_cast_fp16)[name = string("input_71_cast_fp16")];
            string input_73_pad_type_0 = const()[name = string("input_73_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_73_strides_0 = const()[name = string("input_73_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_73_pad_0 = const()[name = string("input_73_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_73_dilations_0 = const()[name = string("input_73_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_73_groups_0 = const()[name = string("input_73_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_2_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41360384))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44506176))))[name = string("layers_2_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_73_cast_fp16 = conv(dilations = input_73_dilations_0, groups = input_73_groups_0, pad = input_73_pad_0, pad_type = input_73_pad_type_0, strides = input_73_strides_0, weight = layers_2_feed_forward1_linear1_weight_to_fp16_palettized, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_75_cast_fp16 = silu(x = input_73_cast_fp16)[name = string("input_75_cast_fp16")];
            string var_703_pad_type_0 = const()[name = string("op_703_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_703_strides_0 = const()[name = string("op_703_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_703_pad_0 = const()[name = string("op_703_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_703_dilations_0 = const()[name = string("op_703_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_703_groups_0 = const()[name = string("op_703_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_704_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44539008))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47684800))))[name = string("op_704_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_704_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_703_dilations_0, groups = var_703_groups_0, pad = var_703_pad_0, pad_type = var_703_pad_type_0, strides = var_703_strides_0, weight = op_704_weight_0_to_fp16_palettized, x = input_75_cast_fp16)[name = string("op_704_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_49_cast_fp16 = add(x = x_47_cast_fp16, y = var_704_cast_fp16)[name = string("x_49_cast_fp16")];
            tensor<int32, [1]> var_720_axes_0 = const()[name = string("op_720_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_2_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_2_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_720_cast_fp16 = layer_norm(axes = var_720_axes_0, epsilon = layers_2_norm_self_att_eps_scaled_to_fp16, x = x_49_cast_fp16)[name = string("op_720_cast_fp16")];
            tensor<fp16, [1024]> x_51_gamma_0_to_fp16 = const()[name = string("x_51_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47693056)))];
            tensor<fp16, [1024]> x_51_beta_0_to_fp16 = const()[name = string("x_51_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47695168)))];
            fp16 x_51_epsilon_0_to_fp16 = const()[name = string("x_51_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_51_cast_fp16 = batch_norm(beta = x_51_beta_0_to_fp16, epsilon = x_51_epsilon_0_to_fp16, gamma = x_51_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_720_cast_fp16)[name = string("x_51_cast_fp16")];
            string q_5_pad_type_0 = const()[name = string("q_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_5_strides_0 = const()[name = string("q_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_5_pad_0 = const()[name = string("q_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_5_dilations_0 = const()[name = string("q_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_5_groups_0 = const()[name = string("q_5_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47697280))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48483776))))[name = string("layers_2_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_5_cast_fp16 = conv(dilations = q_5_dilations_0, groups = q_5_groups_0, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = q_5_strides_0, weight = layers_2_self_attn_linear_q_weight_to_fp16_palettized, x = x_51_cast_fp16)[name = string("q_5_cast_fp16")];
            string k_5_pad_type_0 = const()[name = string("k_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_5_strides_0 = const()[name = string("k_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_5_pad_0 = const()[name = string("k_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_5_dilations_0 = const()[name = string("k_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_5_groups_0 = const()[name = string("k_5_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48492032))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49278528))))[name = string("layers_2_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_5_cast_fp16 = conv(dilations = k_5_dilations_0, groups = k_5_groups_0, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = k_5_strides_0, weight = layers_2_self_attn_linear_k_weight_to_fp16_palettized, x = x_51_cast_fp16)[name = string("k_5_cast_fp16")];
            string v_5_pad_type_0 = const()[name = string("v_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_5_strides_0 = const()[name = string("v_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_5_pad_0 = const()[name = string("v_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_5_dilations_0 = const()[name = string("v_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_5_groups_0 = const()[name = string("v_5_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49286784))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50073280))))[name = string("layers_2_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_5_cast_fp16 = conv(dilations = v_5_dilations_0, groups = v_5_groups_0, pad = v_5_pad_0, pad_type = v_5_pad_type_0, strides = v_5_strides_0, weight = layers_2_self_attn_linear_v_weight_to_fp16_palettized, x = x_51_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_5_to_fp16 = const()[name = string("bv_all_5_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50081536)))];
            tensor<fp16, [1, 1024, 1, 188]> var_752_cast_fp16 = add(x = q_5_cast_fp16, y = bv_all_5_to_fp16)[name = string("op_752_cast_fp16")];
            tensor<int32, [3]> var_753 = const()[name = string("op_753"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_5_cast_fp16 = reshape(shape = var_753, x = var_752_cast_fp16)[name = string("qb_5_cast_fp16")];
            bool bd_all_9_transpose_x_0 = const()[name = string("bd_all_9_transpose_x_0"), val = bool(false)];
            bool bd_all_9_transpose_y_0 = const()[name = string("bd_all_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_2_self_attn_pos_proj_to_fp16 = const()[name = string("layers_2_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50083648)))];
            tensor<fp16, [8, 375, 188]> bd_all_9_cast_fp16 = matmul(transpose_x = bd_all_9_transpose_x_0, transpose_y = bd_all_9_transpose_y_0, x = layers_2_self_attn_pos_proj_to_fp16, y = qb_5_cast_fp16)[name = string("bd_all_9_cast_fp16")];
            tensor<int32, [3]> x_53_perm_0 = const()[name = string("x_53_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_55_pad_0 = const()[name = string("x_55_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_55_mode_0 = const()[name = string("x_55_mode_0"), val = string("constant")];
            fp16 const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_53_cast_fp16 = transpose(perm = x_53_perm_0, x = bd_all_9_cast_fp16)[name = string("transpose_131")];
            tensor<fp16, [8, 188, 376]> x_55_cast_fp16 = pad(constant_val = const_18_to_fp16, mode = x_55_mode_0, pad = x_55_pad_0, x = x_53_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<int32, [3]> var_760 = const()[name = string("op_760"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_57_cast_fp16 = reshape(shape = var_760, x = x_55_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<int32, [3]> var_763_begin_0 = const()[name = string("op_763_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_763_end_0 = const()[name = string("op_763_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_763_end_mask_0 = const()[name = string("op_763_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_763_cast_fp16 = slice_by_index(begin = var_763_begin_0, end = var_763_end_0, end_mask = var_763_end_mask_0, x = x_57_cast_fp16)[name = string("op_763_cast_fp16")];
            tensor<int32, [3]> var_764 = const()[name = string("op_764"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_59_cast_fp16 = reshape(shape = var_764, x = var_763_cast_fp16)[name = string("x_59_cast_fp16")];
            tensor<int32, [3]> bd_all_11_begin_0 = const()[name = string("bd_all_11_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_11_end_0 = const()[name = string("bd_all_11_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_11_end_mask_0 = const()[name = string("bd_all_11_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_11_cast_fp16 = slice_by_index(begin = bd_all_11_begin_0, end = bd_all_11_end_0, end_mask = bd_all_11_end_mask_0, x = x_59_cast_fp16)[name = string("bd_all_11_cast_fp16")];
            tensor<int32, [4]> var_769 = const()[name = string("op_769"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_770_cast_fp16 = reshape(shape = var_769, x = q_5_cast_fp16)[name = string("op_770_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_772_to_fp16 = const()[name = string("op_772_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50851712)))];
            tensor<fp16, [8, 128, 1, 188]> var_773_cast_fp16 = add(x = var_770_cast_fp16, y = var_772_to_fp16)[name = string("op_773_cast_fp16")];
            tensor<int32, [4]> var_774 = const()[name = string("op_774"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_5_cast_fp16 = reshape(shape = var_774, x = k_5_cast_fp16)[name = string("kh_5_cast_fp16")];
            tensor<int32, [4]> var_776 = const()[name = string("op_776"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_5_cast_fp16 = reshape(shape = var_776, x = v_5_cast_fp16)[name = string("vh_5_cast_fp16")];
            tensor<int32, [4]> var_778 = const()[name = string("op_778"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_5_equation_0 = const()[name = string("ac_5_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_779_cast_fp16 = transpose(perm = var_778, x = kh_5_cast_fp16)[name = string("transpose_130")];
            tensor<fp16, [8, 188, 1, 188]> ac_5_cast_fp16 = einsum(equation = ac_5_equation_0, values = (var_779_cast_fp16, var_773_cast_fp16))[name = string("ac_5_cast_fp16")];
            tensor<int32, [3]> var_782_perm_0 = const()[name = string("op_782_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_783_axes_0 = const()[name = string("op_783_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_782_cast_fp16 = transpose(perm = var_782_perm_0, x = bd_all_11_cast_fp16)[name = string("transpose_129")];
            tensor<fp16, [8, 188, 1, 188]> var_783_cast_fp16 = expand_dims(axes = var_783_axes_0, x = var_782_cast_fp16)[name = string("op_783_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_784_cast_fp16 = add(x = ac_5_cast_fp16, y = var_783_cast_fp16)[name = string("op_784_cast_fp16")];
            fp16 var_785_to_fp16 = const()[name = string("op_785_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_9_cast_fp16 = mul(x = var_784_cast_fp16, y = var_785_to_fp16)[name = string("scores_9_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_11_cast_fp16 = add(x = scores_9_cast_fp16, y = key_bias)[name = string("scores_11_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_788_cast_fp16 = softmax(axis = var_659, x = scores_11_cast_fp16)[name = string("op_788_cast_fp16")];
            tensor<int32, [4]> transpose_50_perm_0 = const()[name = string("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_4_perm_0 = const()[name = string("transpose_4_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_24 = const()[name = string("concat_24"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_4_cast_fp16 = transpose(perm = transpose_4_perm_0, x = var_788_cast_fp16)[name = string("transpose_128")];
            tensor<fp16, [8, 188, 188]> reshape_6_cast_fp16 = reshape(shape = concat_24, x = transpose_4_cast_fp16)[name = string("reshape_6_cast_fp16")];
            tensor<int32, [3]> concat_25 = const()[name = string("concat_25"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_50_cast_fp16 = transpose(perm = transpose_50_perm_0, x = vh_5_cast_fp16)[name = string("transpose_127")];
            tensor<fp16, [8, 188, 128]> reshape_7_cast_fp16 = reshape(shape = concat_25, x = transpose_50_cast_fp16)[name = string("reshape_7_cast_fp16")];
            bool matmul_2_transpose_x_0 = const()[name = string("matmul_2_transpose_x_0"), val = bool(false)];
            bool matmul_2_transpose_y_0 = const()[name = string("matmul_2_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = reshape_6_cast_fp16, y = reshape_7_cast_fp16)[name = string("matmul_2_cast_fp16")];
            tensor<int32, [4]> concat_29 = const()[name = string("concat_29"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_8_cast_fp16 = reshape(shape = concat_29, x = matmul_2_cast_fp16)[name = string("reshape_8_cast_fp16")];
            tensor<int32, [4]> ctx_5_perm_0 = const()[name = string("ctx_5_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_793 = const()[name = string("op_793"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_5_cast_fp16 = transpose(perm = ctx_5_perm_0, x = reshape_8_cast_fp16)[name = string("transpose_126")];
            tensor<fp16, [1, 1024, 1, 188]> input_77_cast_fp16 = reshape(shape = var_793, x = ctx_5_cast_fp16)[name = string("input_77_cast_fp16")];
            string var_800_pad_type_0 = const()[name = string("op_800_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_800_strides_0 = const()[name = string("op_800_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_800_pad_0 = const()[name = string("op_800_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_800_dilations_0 = const()[name = string("op_800_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_800_groups_0 = const()[name = string("op_800_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50853824))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51640320))))[name = string("layers_2_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_800_cast_fp16 = conv(dilations = var_800_dilations_0, groups = var_800_groups_0, pad = var_800_pad_0, pad_type = var_800_pad_type_0, strides = var_800_strides_0, weight = layers_2_self_attn_linear_out_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = string("op_800_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_61_cast_fp16 = add(x = x_49_cast_fp16, y = var_800_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<int32, [1]> var_816_axes_0 = const()[name = string("op_816_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_2_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_2_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_816_cast_fp16 = layer_norm(axes = var_816_axes_0, epsilon = layers_2_norm_conv_eps_scaled_to_fp16, x = x_61_cast_fp16)[name = string("op_816_cast_fp16")];
            tensor<fp16, [1024]> input_79_gamma_0_to_fp16 = const()[name = string("input_79_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51648576)))];
            tensor<fp16, [1024]> input_79_beta_0_to_fp16 = const()[name = string("input_79_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51650688)))];
            fp16 input_79_epsilon_0_to_fp16 = const()[name = string("input_79_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_79_cast_fp16 = batch_norm(beta = input_79_beta_0_to_fp16, epsilon = input_79_epsilon_0_to_fp16, gamma = input_79_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_816_cast_fp16)[name = string("input_79_cast_fp16")];
            string input_81_pad_type_0 = const()[name = string("input_81_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_81_strides_0 = const()[name = string("input_81_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_81_pad_0 = const()[name = string("input_81_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_81_dilations_0 = const()[name = string("input_81_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_81_groups_0 = const()[name = string("input_81_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_2_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51652800))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53225728))))[name = string("layers_2_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_81_cast_fp16 = conv(dilations = input_81_dilations_0, groups = input_81_groups_0, pad = input_81_pad_0, pad_type = input_81_pad_type_0, strides = input_81_strides_0, weight = layers_2_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_79_cast_fp16)[name = string("input_81_cast_fp16")];
            int32 x_63_split_num_splits_0 = const()[name = string("x_63_split_num_splits_0"), val = int32(2)];
            int32 x_63_split_axis_0 = const()[name = string("x_63_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_63_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_81_cast_fp16)[name = string("x_63_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_63_split_1_sigmoid_cast_fp16 = sigmoid(x = x_63_split_cast_fp16_1)[name = string("x_63_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_63_cast_fp16 = mul(x = x_63_split_cast_fp16_0, y = x_63_split_1_sigmoid_cast_fp16)[name = string("x_63_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_83_cast_fp16 = mul(x = x_63_cast_fp16, y = pad_mask)[name = string("input_83_cast_fp16")];
            string input_85_pad_type_0 = const()[name = string("input_85_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_85_pad_0 = const()[name = string("input_85_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_85_groups_0 = const()[name = string("input_85_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_85_strides_0 = const()[name = string("input_85_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_85_dilations_0 = const()[name = string("input_85_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_107_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53242176))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53249152))))[name = string("const_107_to_fp16_palettized")];
            tensor<fp16, [1024]> const_108_to_fp16 = const()[name = string("const_108_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53257408)))];
            tensor<fp16, [1, 1024, 1, 188]> input_87_cast_fp16 = conv(bias = const_108_to_fp16, dilations = input_85_dilations_0, groups = input_85_groups_0, pad = input_85_pad_0, pad_type = input_85_pad_type_0, strides = input_85_strides_0, weight = const_107_to_fp16_palettized, x = input_83_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_89_cast_fp16 = silu(x = input_87_cast_fp16)[name = string("input_89_cast_fp16")];
            string var_848_pad_type_0 = const()[name = string("op_848_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_848_strides_0 = const()[name = string("op_848_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_848_pad_0 = const()[name = string("op_848_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_848_dilations_0 = const()[name = string("op_848_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_848_groups_0 = const()[name = string("op_848_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_2_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53259520))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54046016))))[name = string("layers_2_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_848_cast_fp16 = conv(dilations = var_848_dilations_0, groups = var_848_groups_0, pad = var_848_pad_0, pad_type = var_848_pad_type_0, strides = var_848_strides_0, weight = layers_2_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = string("op_848_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_65_cast_fp16 = add(x = x_61_cast_fp16, y = var_848_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [1]> var_864_axes_0 = const()[name = string("op_864_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_2_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_2_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_864_cast_fp16 = layer_norm(axes = var_864_axes_0, epsilon = layers_2_norm_feed_forward2_eps_scaled_to_fp16, x = x_65_cast_fp16)[name = string("op_864_cast_fp16")];
            tensor<fp16, [1024]> input_91_gamma_0_to_fp16 = const()[name = string("input_91_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54054272)))];
            tensor<fp16, [1024]> input_91_beta_0_to_fp16 = const()[name = string("input_91_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54056384)))];
            fp16 input_91_epsilon_0_to_fp16 = const()[name = string("input_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_91_cast_fp16 = batch_norm(beta = input_91_beta_0_to_fp16, epsilon = input_91_epsilon_0_to_fp16, gamma = input_91_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_864_cast_fp16)[name = string("input_91_cast_fp16")];
            string input_93_pad_type_0 = const()[name = string("input_93_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_93_strides_0 = const()[name = string("input_93_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_93_pad_0 = const()[name = string("input_93_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_93_dilations_0 = const()[name = string("input_93_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_93_groups_0 = const()[name = string("input_93_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_2_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54058496))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57204288))))[name = string("layers_2_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_93_cast_fp16 = conv(dilations = input_93_dilations_0, groups = input_93_groups_0, pad = input_93_pad_0, pad_type = input_93_pad_type_0, strides = input_93_strides_0, weight = layers_2_feed_forward2_linear1_weight_to_fp16_palettized, x = input_91_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_95_cast_fp16 = silu(x = input_93_cast_fp16)[name = string("input_95_cast_fp16")];
            string var_881_pad_type_0 = const()[name = string("op_881_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_881_strides_0 = const()[name = string("op_881_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_881_pad_0 = const()[name = string("op_881_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_881_dilations_0 = const()[name = string("op_881_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_881_groups_0 = const()[name = string("op_881_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_882_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57237120))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60382912))))[name = string("op_882_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_882_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_881_dilations_0, groups = var_881_groups_0, pad = var_881_pad_0, pad_type = var_881_pad_type_0, strides = var_881_strides_0, weight = op_882_weight_0_to_fp16_palettized, x = input_95_cast_fp16)[name = string("op_882_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_67_cast_fp16 = add(x = x_65_cast_fp16, y = var_882_cast_fp16)[name = string("x_67_cast_fp16")];
            tensor<int32, [1]> var_898_axes_0 = const()[name = string("op_898_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_2_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_2_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_898_cast_fp16 = layer_norm(axes = var_898_axes_0, epsilon = layers_2_norm_out_eps_scaled_to_fp16, x = x_67_cast_fp16)[name = string("op_898_cast_fp16")];
            tensor<fp16, [1024]> x_69_gamma_0_to_fp16 = const()[name = string("x_69_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60391168)))];
            tensor<fp16, [1024]> x_69_beta_0_to_fp16 = const()[name = string("x_69_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60393280)))];
            fp16 x_69_epsilon_0_to_fp16 = const()[name = string("x_69_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_69_cast_fp16 = batch_norm(beta = x_69_beta_0_to_fp16, epsilon = x_69_epsilon_0_to_fp16, gamma = x_69_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_898_cast_fp16)[name = string("x_69_cast_fp16")];
            int32 var_917 = const()[name = string("op_917"), val = int32(1)];
            tensor<int32, [1]> var_944_axes_0 = const()[name = string("op_944_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_3_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_3_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_944_cast_fp16 = layer_norm(axes = var_944_axes_0, epsilon = layers_3_norm_feed_forward1_eps_scaled_to_fp16, x = x_69_cast_fp16)[name = string("op_944_cast_fp16")];
            tensor<fp16, [1024]> input_97_gamma_0_to_fp16 = const()[name = string("input_97_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60395392)))];
            tensor<fp16, [1024]> input_97_beta_0_to_fp16 = const()[name = string("input_97_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60397504)))];
            fp16 input_97_epsilon_0_to_fp16 = const()[name = string("input_97_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_97_cast_fp16 = batch_norm(beta = input_97_beta_0_to_fp16, epsilon = input_97_epsilon_0_to_fp16, gamma = input_97_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_944_cast_fp16)[name = string("input_97_cast_fp16")];
            string input_99_pad_type_0 = const()[name = string("input_99_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_99_strides_0 = const()[name = string("input_99_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_99_pad_0 = const()[name = string("input_99_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_99_dilations_0 = const()[name = string("input_99_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_99_groups_0 = const()[name = string("input_99_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_3_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60399616))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63545408))))[name = string("layers_3_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_99_cast_fp16 = conv(dilations = input_99_dilations_0, groups = input_99_groups_0, pad = input_99_pad_0, pad_type = input_99_pad_type_0, strides = input_99_strides_0, weight = layers_3_feed_forward1_linear1_weight_to_fp16_palettized, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_101_cast_fp16 = silu(x = input_99_cast_fp16)[name = string("input_101_cast_fp16")];
            string var_961_pad_type_0 = const()[name = string("op_961_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_961_strides_0 = const()[name = string("op_961_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_961_pad_0 = const()[name = string("op_961_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_961_dilations_0 = const()[name = string("op_961_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_961_groups_0 = const()[name = string("op_961_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_962_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63578240))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66724032))))[name = string("op_962_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_962_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_961_dilations_0, groups = var_961_groups_0, pad = var_961_pad_0, pad_type = var_961_pad_type_0, strides = var_961_strides_0, weight = op_962_weight_0_to_fp16_palettized, x = input_101_cast_fp16)[name = string("op_962_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_71_cast_fp16 = add(x = x_69_cast_fp16, y = var_962_cast_fp16)[name = string("x_71_cast_fp16")];
            tensor<int32, [1]> var_978_axes_0 = const()[name = string("op_978_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_3_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_3_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_978_cast_fp16 = layer_norm(axes = var_978_axes_0, epsilon = layers_3_norm_self_att_eps_scaled_to_fp16, x = x_71_cast_fp16)[name = string("op_978_cast_fp16")];
            tensor<fp16, [1024]> x_73_gamma_0_to_fp16 = const()[name = string("x_73_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66732288)))];
            tensor<fp16, [1024]> x_73_beta_0_to_fp16 = const()[name = string("x_73_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66734400)))];
            fp16 x_73_epsilon_0_to_fp16 = const()[name = string("x_73_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_73_cast_fp16 = batch_norm(beta = x_73_beta_0_to_fp16, epsilon = x_73_epsilon_0_to_fp16, gamma = x_73_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_978_cast_fp16)[name = string("x_73_cast_fp16")];
            string q_7_pad_type_0 = const()[name = string("q_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_7_strides_0 = const()[name = string("q_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_7_pad_0 = const()[name = string("q_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_7_dilations_0 = const()[name = string("q_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_7_groups_0 = const()[name = string("q_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66736512))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67523008))))[name = string("layers_3_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_7_cast_fp16 = conv(dilations = q_7_dilations_0, groups = q_7_groups_0, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = q_7_strides_0, weight = layers_3_self_attn_linear_q_weight_to_fp16_palettized, x = x_73_cast_fp16)[name = string("q_7_cast_fp16")];
            string k_7_pad_type_0 = const()[name = string("k_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_7_strides_0 = const()[name = string("k_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_7_pad_0 = const()[name = string("k_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_7_dilations_0 = const()[name = string("k_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_7_groups_0 = const()[name = string("k_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67531264))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68317760))))[name = string("layers_3_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_7_cast_fp16 = conv(dilations = k_7_dilations_0, groups = k_7_groups_0, pad = k_7_pad_0, pad_type = k_7_pad_type_0, strides = k_7_strides_0, weight = layers_3_self_attn_linear_k_weight_to_fp16_palettized, x = x_73_cast_fp16)[name = string("k_7_cast_fp16")];
            string v_7_pad_type_0 = const()[name = string("v_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_7_strides_0 = const()[name = string("v_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_7_pad_0 = const()[name = string("v_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_7_dilations_0 = const()[name = string("v_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_7_groups_0 = const()[name = string("v_7_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68326016))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69112512))))[name = string("layers_3_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_7_cast_fp16 = conv(dilations = v_7_dilations_0, groups = v_7_groups_0, pad = v_7_pad_0, pad_type = v_7_pad_type_0, strides = v_7_strides_0, weight = layers_3_self_attn_linear_v_weight_to_fp16_palettized, x = x_73_cast_fp16)[name = string("v_7_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_7_to_fp16 = const()[name = string("bv_all_7_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69120768)))];
            tensor<fp16, [1, 1024, 1, 188]> var_1010_cast_fp16 = add(x = q_7_cast_fp16, y = bv_all_7_to_fp16)[name = string("op_1010_cast_fp16")];
            tensor<int32, [3]> var_1011 = const()[name = string("op_1011"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_7_cast_fp16 = reshape(shape = var_1011, x = var_1010_cast_fp16)[name = string("qb_7_cast_fp16")];
            bool bd_all_13_transpose_x_0 = const()[name = string("bd_all_13_transpose_x_0"), val = bool(false)];
            bool bd_all_13_transpose_y_0 = const()[name = string("bd_all_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_3_self_attn_pos_proj_to_fp16 = const()[name = string("layers_3_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69122880)))];
            tensor<fp16, [8, 375, 188]> bd_all_13_cast_fp16 = matmul(transpose_x = bd_all_13_transpose_x_0, transpose_y = bd_all_13_transpose_y_0, x = layers_3_self_attn_pos_proj_to_fp16, y = qb_7_cast_fp16)[name = string("bd_all_13_cast_fp16")];
            tensor<int32, [3]> x_75_perm_0 = const()[name = string("x_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_77_pad_0 = const()[name = string("x_77_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_77_mode_0 = const()[name = string("x_77_mode_0"), val = string("constant")];
            fp16 const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_75_cast_fp16 = transpose(perm = x_75_perm_0, x = bd_all_13_cast_fp16)[name = string("transpose_125")];
            tensor<fp16, [8, 188, 376]> x_77_cast_fp16 = pad(constant_val = const_22_to_fp16, mode = x_77_mode_0, pad = x_77_pad_0, x = x_75_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<int32, [3]> var_1018 = const()[name = string("op_1018"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_79_cast_fp16 = reshape(shape = var_1018, x = x_77_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<int32, [3]> var_1021_begin_0 = const()[name = string("op_1021_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_1021_end_0 = const()[name = string("op_1021_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_1021_end_mask_0 = const()[name = string("op_1021_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_1021_cast_fp16 = slice_by_index(begin = var_1021_begin_0, end = var_1021_end_0, end_mask = var_1021_end_mask_0, x = x_79_cast_fp16)[name = string("op_1021_cast_fp16")];
            tensor<int32, [3]> var_1022 = const()[name = string("op_1022"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_81_cast_fp16 = reshape(shape = var_1022, x = var_1021_cast_fp16)[name = string("x_81_cast_fp16")];
            tensor<int32, [3]> bd_all_15_begin_0 = const()[name = string("bd_all_15_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_15_end_0 = const()[name = string("bd_all_15_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_15_end_mask_0 = const()[name = string("bd_all_15_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_15_cast_fp16 = slice_by_index(begin = bd_all_15_begin_0, end = bd_all_15_end_0, end_mask = bd_all_15_end_mask_0, x = x_81_cast_fp16)[name = string("bd_all_15_cast_fp16")];
            tensor<int32, [4]> var_1027 = const()[name = string("op_1027"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_1028_cast_fp16 = reshape(shape = var_1027, x = q_7_cast_fp16)[name = string("op_1028_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_1030_to_fp16 = const()[name = string("op_1030_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69890944)))];
            tensor<fp16, [8, 128, 1, 188]> var_1031_cast_fp16 = add(x = var_1028_cast_fp16, y = var_1030_to_fp16)[name = string("op_1031_cast_fp16")];
            tensor<int32, [4]> var_1032 = const()[name = string("op_1032"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_7_cast_fp16 = reshape(shape = var_1032, x = k_7_cast_fp16)[name = string("kh_7_cast_fp16")];
            tensor<int32, [4]> var_1034 = const()[name = string("op_1034"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_7_cast_fp16 = reshape(shape = var_1034, x = v_7_cast_fp16)[name = string("vh_7_cast_fp16")];
            tensor<int32, [4]> var_1036 = const()[name = string("op_1036"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_7_equation_0 = const()[name = string("ac_7_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_1037_cast_fp16 = transpose(perm = var_1036, x = kh_7_cast_fp16)[name = string("transpose_124")];
            tensor<fp16, [8, 188, 1, 188]> ac_7_cast_fp16 = einsum(equation = ac_7_equation_0, values = (var_1037_cast_fp16, var_1031_cast_fp16))[name = string("ac_7_cast_fp16")];
            tensor<int32, [3]> var_1040_perm_0 = const()[name = string("op_1040_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1041_axes_0 = const()[name = string("op_1041_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_1040_cast_fp16 = transpose(perm = var_1040_perm_0, x = bd_all_15_cast_fp16)[name = string("transpose_123")];
            tensor<fp16, [8, 188, 1, 188]> var_1041_cast_fp16 = expand_dims(axes = var_1041_axes_0, x = var_1040_cast_fp16)[name = string("op_1041_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_1042_cast_fp16 = add(x = ac_7_cast_fp16, y = var_1041_cast_fp16)[name = string("op_1042_cast_fp16")];
            fp16 var_1043_to_fp16 = const()[name = string("op_1043_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_13_cast_fp16 = mul(x = var_1042_cast_fp16, y = var_1043_to_fp16)[name = string("scores_13_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_15_cast_fp16 = add(x = scores_13_cast_fp16, y = key_bias)[name = string("scores_15_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_1046_cast_fp16 = softmax(axis = var_917, x = scores_15_cast_fp16)[name = string("op_1046_cast_fp16")];
            tensor<int32, [4]> transpose_51_perm_0 = const()[name = string("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_6_perm_0 = const()[name = string("transpose_6_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_34 = const()[name = string("concat_34"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_6_cast_fp16 = transpose(perm = transpose_6_perm_0, x = var_1046_cast_fp16)[name = string("transpose_122")];
            tensor<fp16, [8, 188, 188]> reshape_9_cast_fp16 = reshape(shape = concat_34, x = transpose_6_cast_fp16)[name = string("reshape_9_cast_fp16")];
            tensor<int32, [3]> concat_35 = const()[name = string("concat_35"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_51_cast_fp16 = transpose(perm = transpose_51_perm_0, x = vh_7_cast_fp16)[name = string("transpose_121")];
            tensor<fp16, [8, 188, 128]> reshape_10_cast_fp16 = reshape(shape = concat_35, x = transpose_51_cast_fp16)[name = string("reshape_10_cast_fp16")];
            bool matmul_3_transpose_x_0 = const()[name = string("matmul_3_transpose_x_0"), val = bool(false)];
            bool matmul_3_transpose_y_0 = const()[name = string("matmul_3_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = reshape_9_cast_fp16, y = reshape_10_cast_fp16)[name = string("matmul_3_cast_fp16")];
            tensor<int32, [4]> concat_39 = const()[name = string("concat_39"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_11_cast_fp16 = reshape(shape = concat_39, x = matmul_3_cast_fp16)[name = string("reshape_11_cast_fp16")];
            tensor<int32, [4]> ctx_7_perm_0 = const()[name = string("ctx_7_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_1051 = const()[name = string("op_1051"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_7_cast_fp16 = transpose(perm = ctx_7_perm_0, x = reshape_11_cast_fp16)[name = string("transpose_120")];
            tensor<fp16, [1, 1024, 1, 188]> input_103_cast_fp16 = reshape(shape = var_1051, x = ctx_7_cast_fp16)[name = string("input_103_cast_fp16")];
            string var_1058_pad_type_0 = const()[name = string("op_1058_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1058_strides_0 = const()[name = string("op_1058_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1058_pad_0 = const()[name = string("op_1058_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1058_dilations_0 = const()[name = string("op_1058_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1058_groups_0 = const()[name = string("op_1058_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69893056))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70679552))))[name = string("layers_3_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1058_cast_fp16 = conv(dilations = var_1058_dilations_0, groups = var_1058_groups_0, pad = var_1058_pad_0, pad_type = var_1058_pad_type_0, strides = var_1058_strides_0, weight = layers_3_self_attn_linear_out_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = string("op_1058_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_83_cast_fp16 = add(x = x_71_cast_fp16, y = var_1058_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<int32, [1]> var_1074_axes_0 = const()[name = string("op_1074_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_3_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_3_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1074_cast_fp16 = layer_norm(axes = var_1074_axes_0, epsilon = layers_3_norm_conv_eps_scaled_to_fp16, x = x_83_cast_fp16)[name = string("op_1074_cast_fp16")];
            tensor<fp16, [1024]> input_105_gamma_0_to_fp16 = const()[name = string("input_105_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70687808)))];
            tensor<fp16, [1024]> input_105_beta_0_to_fp16 = const()[name = string("input_105_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70689920)))];
            fp16 input_105_epsilon_0_to_fp16 = const()[name = string("input_105_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_105_cast_fp16 = batch_norm(beta = input_105_beta_0_to_fp16, epsilon = input_105_epsilon_0_to_fp16, gamma = input_105_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1074_cast_fp16)[name = string("input_105_cast_fp16")];
            string input_107_pad_type_0 = const()[name = string("input_107_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_107_strides_0 = const()[name = string("input_107_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_107_pad_0 = const()[name = string("input_107_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_107_dilations_0 = const()[name = string("input_107_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_107_groups_0 = const()[name = string("input_107_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_3_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(70692032))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72264960))))[name = string("layers_3_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_107_cast_fp16 = conv(dilations = input_107_dilations_0, groups = input_107_groups_0, pad = input_107_pad_0, pad_type = input_107_pad_type_0, strides = input_107_strides_0, weight = layers_3_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_105_cast_fp16)[name = string("input_107_cast_fp16")];
            int32 x_85_split_num_splits_0 = const()[name = string("x_85_split_num_splits_0"), val = int32(2)];
            int32 x_85_split_axis_0 = const()[name = string("x_85_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_85_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_107_cast_fp16)[name = string("x_85_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_85_split_1_sigmoid_cast_fp16 = sigmoid(x = x_85_split_cast_fp16_1)[name = string("x_85_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_85_cast_fp16 = mul(x = x_85_split_cast_fp16_0, y = x_85_split_1_sigmoid_cast_fp16)[name = string("x_85_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_109_cast_fp16 = mul(x = x_85_cast_fp16, y = pad_mask)[name = string("input_109_cast_fp16")];
            string input_111_pad_type_0 = const()[name = string("input_111_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_111_pad_0 = const()[name = string("input_111_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_111_groups_0 = const()[name = string("input_111_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_111_strides_0 = const()[name = string("input_111_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_111_dilations_0 = const()[name = string("input_111_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_109_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72281408))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72288384))))[name = string("const_109_to_fp16_palettized")];
            tensor<fp16, [1024]> const_110_to_fp16 = const()[name = string("const_110_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72296640)))];
            tensor<fp16, [1, 1024, 1, 188]> input_113_cast_fp16 = conv(bias = const_110_to_fp16, dilations = input_111_dilations_0, groups = input_111_groups_0, pad = input_111_pad_0, pad_type = input_111_pad_type_0, strides = input_111_strides_0, weight = const_109_to_fp16_palettized, x = input_109_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_115_cast_fp16 = silu(x = input_113_cast_fp16)[name = string("input_115_cast_fp16")];
            string var_1106_pad_type_0 = const()[name = string("op_1106_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1106_strides_0 = const()[name = string("op_1106_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1106_pad_0 = const()[name = string("op_1106_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1106_dilations_0 = const()[name = string("op_1106_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1106_groups_0 = const()[name = string("op_1106_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_3_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72298752))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73085248))))[name = string("layers_3_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1106_cast_fp16 = conv(dilations = var_1106_dilations_0, groups = var_1106_groups_0, pad = var_1106_pad_0, pad_type = var_1106_pad_type_0, strides = var_1106_strides_0, weight = layers_3_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = string("op_1106_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_87_cast_fp16 = add(x = x_83_cast_fp16, y = var_1106_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<int32, [1]> var_1122_axes_0 = const()[name = string("op_1122_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_3_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_3_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1122_cast_fp16 = layer_norm(axes = var_1122_axes_0, epsilon = layers_3_norm_feed_forward2_eps_scaled_to_fp16, x = x_87_cast_fp16)[name = string("op_1122_cast_fp16")];
            tensor<fp16, [1024]> input_117_gamma_0_to_fp16 = const()[name = string("input_117_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73093504)))];
            tensor<fp16, [1024]> input_117_beta_0_to_fp16 = const()[name = string("input_117_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73095616)))];
            fp16 input_117_epsilon_0_to_fp16 = const()[name = string("input_117_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_117_cast_fp16 = batch_norm(beta = input_117_beta_0_to_fp16, epsilon = input_117_epsilon_0_to_fp16, gamma = input_117_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1122_cast_fp16)[name = string("input_117_cast_fp16")];
            string input_119_pad_type_0 = const()[name = string("input_119_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_119_strides_0 = const()[name = string("input_119_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_119_pad_0 = const()[name = string("input_119_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_119_dilations_0 = const()[name = string("input_119_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_119_groups_0 = const()[name = string("input_119_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_3_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73097728))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76243520))))[name = string("layers_3_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_119_cast_fp16 = conv(dilations = input_119_dilations_0, groups = input_119_groups_0, pad = input_119_pad_0, pad_type = input_119_pad_type_0, strides = input_119_strides_0, weight = layers_3_feed_forward2_linear1_weight_to_fp16_palettized, x = input_117_cast_fp16)[name = string("input_119_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_121_cast_fp16 = silu(x = input_119_cast_fp16)[name = string("input_121_cast_fp16")];
            string var_1139_pad_type_0 = const()[name = string("op_1139_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1139_strides_0 = const()[name = string("op_1139_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1139_pad_0 = const()[name = string("op_1139_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1139_dilations_0 = const()[name = string("op_1139_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1139_groups_0 = const()[name = string("op_1139_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_1140_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76276352))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79422144))))[name = string("op_1140_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1140_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_1139_dilations_0, groups = var_1139_groups_0, pad = var_1139_pad_0, pad_type = var_1139_pad_type_0, strides = var_1139_strides_0, weight = op_1140_weight_0_to_fp16_palettized, x = input_121_cast_fp16)[name = string("op_1140_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_89_cast_fp16 = add(x = x_87_cast_fp16, y = var_1140_cast_fp16)[name = string("x_89_cast_fp16")];
            tensor<int32, [1]> var_1156_axes_0 = const()[name = string("op_1156_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_3_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_3_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1156_cast_fp16 = layer_norm(axes = var_1156_axes_0, epsilon = layers_3_norm_out_eps_scaled_to_fp16, x = x_89_cast_fp16)[name = string("op_1156_cast_fp16")];
            tensor<fp16, [1024]> x_91_gamma_0_to_fp16 = const()[name = string("x_91_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79430400)))];
            tensor<fp16, [1024]> x_91_beta_0_to_fp16 = const()[name = string("x_91_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79432512)))];
            fp16 x_91_epsilon_0_to_fp16 = const()[name = string("x_91_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_91_cast_fp16 = batch_norm(beta = x_91_beta_0_to_fp16, epsilon = x_91_epsilon_0_to_fp16, gamma = x_91_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1156_cast_fp16)[name = string("x_91_cast_fp16")];
            int32 var_1175 = const()[name = string("op_1175"), val = int32(1)];
            tensor<int32, [1]> var_1202_axes_0 = const()[name = string("op_1202_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_4_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_4_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1202_cast_fp16 = layer_norm(axes = var_1202_axes_0, epsilon = layers_4_norm_feed_forward1_eps_scaled_to_fp16, x = x_91_cast_fp16)[name = string("op_1202_cast_fp16")];
            tensor<fp16, [1024]> input_123_gamma_0_to_fp16 = const()[name = string("input_123_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79434624)))];
            tensor<fp16, [1024]> input_123_beta_0_to_fp16 = const()[name = string("input_123_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79436736)))];
            fp16 input_123_epsilon_0_to_fp16 = const()[name = string("input_123_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_123_cast_fp16 = batch_norm(beta = input_123_beta_0_to_fp16, epsilon = input_123_epsilon_0_to_fp16, gamma = input_123_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1202_cast_fp16)[name = string("input_123_cast_fp16")];
            string input_125_pad_type_0 = const()[name = string("input_125_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_125_strides_0 = const()[name = string("input_125_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_125_pad_0 = const()[name = string("input_125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_125_dilations_0 = const()[name = string("input_125_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_125_groups_0 = const()[name = string("input_125_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_4_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79438848))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82584640))))[name = string("layers_4_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_125_cast_fp16 = conv(dilations = input_125_dilations_0, groups = input_125_groups_0, pad = input_125_pad_0, pad_type = input_125_pad_type_0, strides = input_125_strides_0, weight = layers_4_feed_forward1_linear1_weight_to_fp16_palettized, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_127_cast_fp16 = silu(x = input_125_cast_fp16)[name = string("input_127_cast_fp16")];
            string var_1219_pad_type_0 = const()[name = string("op_1219_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1219_strides_0 = const()[name = string("op_1219_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1219_pad_0 = const()[name = string("op_1219_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1219_dilations_0 = const()[name = string("op_1219_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1219_groups_0 = const()[name = string("op_1219_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_1220_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82617472))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85763264))))[name = string("op_1220_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1220_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_1219_dilations_0, groups = var_1219_groups_0, pad = var_1219_pad_0, pad_type = var_1219_pad_type_0, strides = var_1219_strides_0, weight = op_1220_weight_0_to_fp16_palettized, x = input_127_cast_fp16)[name = string("op_1220_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_93_cast_fp16 = add(x = x_91_cast_fp16, y = var_1220_cast_fp16)[name = string("x_93_cast_fp16")];
            tensor<int32, [1]> var_1236_axes_0 = const()[name = string("op_1236_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_4_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_4_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1236_cast_fp16 = layer_norm(axes = var_1236_axes_0, epsilon = layers_4_norm_self_att_eps_scaled_to_fp16, x = x_93_cast_fp16)[name = string("op_1236_cast_fp16")];
            tensor<fp16, [1024]> x_95_gamma_0_to_fp16 = const()[name = string("x_95_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85771520)))];
            tensor<fp16, [1024]> x_95_beta_0_to_fp16 = const()[name = string("x_95_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85773632)))];
            fp16 x_95_epsilon_0_to_fp16 = const()[name = string("x_95_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_95_cast_fp16 = batch_norm(beta = x_95_beta_0_to_fp16, epsilon = x_95_epsilon_0_to_fp16, gamma = x_95_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1236_cast_fp16)[name = string("x_95_cast_fp16")];
            string q_9_pad_type_0 = const()[name = string("q_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_9_strides_0 = const()[name = string("q_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_9_pad_0 = const()[name = string("q_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_9_dilations_0 = const()[name = string("q_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_9_groups_0 = const()[name = string("q_9_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85775744))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86562240))))[name = string("layers_4_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_9_cast_fp16 = conv(dilations = q_9_dilations_0, groups = q_9_groups_0, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = q_9_strides_0, weight = layers_4_self_attn_linear_q_weight_to_fp16_palettized, x = x_95_cast_fp16)[name = string("q_9_cast_fp16")];
            string k_9_pad_type_0 = const()[name = string("k_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_9_strides_0 = const()[name = string("k_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_9_pad_0 = const()[name = string("k_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_9_dilations_0 = const()[name = string("k_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_9_groups_0 = const()[name = string("k_9_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86570496))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87356992))))[name = string("layers_4_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_9_cast_fp16 = conv(dilations = k_9_dilations_0, groups = k_9_groups_0, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = k_9_strides_0, weight = layers_4_self_attn_linear_k_weight_to_fp16_palettized, x = x_95_cast_fp16)[name = string("k_9_cast_fp16")];
            string v_9_pad_type_0 = const()[name = string("v_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_9_strides_0 = const()[name = string("v_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_9_pad_0 = const()[name = string("v_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_9_dilations_0 = const()[name = string("v_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_9_groups_0 = const()[name = string("v_9_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87365248))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88151744))))[name = string("layers_4_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_9_cast_fp16 = conv(dilations = v_9_dilations_0, groups = v_9_groups_0, pad = v_9_pad_0, pad_type = v_9_pad_type_0, strides = v_9_strides_0, weight = layers_4_self_attn_linear_v_weight_to_fp16_palettized, x = x_95_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_9_to_fp16 = const()[name = string("bv_all_9_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88160000)))];
            tensor<fp16, [1, 1024, 1, 188]> var_1268_cast_fp16 = add(x = q_9_cast_fp16, y = bv_all_9_to_fp16)[name = string("op_1268_cast_fp16")];
            tensor<int32, [3]> var_1269 = const()[name = string("op_1269"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_9_cast_fp16 = reshape(shape = var_1269, x = var_1268_cast_fp16)[name = string("qb_9_cast_fp16")];
            bool bd_all_17_transpose_x_0 = const()[name = string("bd_all_17_transpose_x_0"), val = bool(false)];
            bool bd_all_17_transpose_y_0 = const()[name = string("bd_all_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_4_self_attn_pos_proj_to_fp16 = const()[name = string("layers_4_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88162112)))];
            tensor<fp16, [8, 375, 188]> bd_all_17_cast_fp16 = matmul(transpose_x = bd_all_17_transpose_x_0, transpose_y = bd_all_17_transpose_y_0, x = layers_4_self_attn_pos_proj_to_fp16, y = qb_9_cast_fp16)[name = string("bd_all_17_cast_fp16")];
            tensor<int32, [3]> x_97_perm_0 = const()[name = string("x_97_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_99_pad_0 = const()[name = string("x_99_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_99_mode_0 = const()[name = string("x_99_mode_0"), val = string("constant")];
            fp16 const_26_to_fp16 = const()[name = string("const_26_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_97_cast_fp16 = transpose(perm = x_97_perm_0, x = bd_all_17_cast_fp16)[name = string("transpose_119")];
            tensor<fp16, [8, 188, 376]> x_99_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = x_99_mode_0, pad = x_99_pad_0, x = x_97_cast_fp16)[name = string("x_99_cast_fp16")];
            tensor<int32, [3]> var_1276 = const()[name = string("op_1276"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_101_cast_fp16 = reshape(shape = var_1276, x = x_99_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [3]> var_1279_begin_0 = const()[name = string("op_1279_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_1279_end_0 = const()[name = string("op_1279_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_1279_end_mask_0 = const()[name = string("op_1279_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_1279_cast_fp16 = slice_by_index(begin = var_1279_begin_0, end = var_1279_end_0, end_mask = var_1279_end_mask_0, x = x_101_cast_fp16)[name = string("op_1279_cast_fp16")];
            tensor<int32, [3]> var_1280 = const()[name = string("op_1280"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_103_cast_fp16 = reshape(shape = var_1280, x = var_1279_cast_fp16)[name = string("x_103_cast_fp16")];
            tensor<int32, [3]> bd_all_19_begin_0 = const()[name = string("bd_all_19_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_19_end_0 = const()[name = string("bd_all_19_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_19_end_mask_0 = const()[name = string("bd_all_19_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_19_cast_fp16 = slice_by_index(begin = bd_all_19_begin_0, end = bd_all_19_end_0, end_mask = bd_all_19_end_mask_0, x = x_103_cast_fp16)[name = string("bd_all_19_cast_fp16")];
            tensor<int32, [4]> var_1285 = const()[name = string("op_1285"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_1286_cast_fp16 = reshape(shape = var_1285, x = q_9_cast_fp16)[name = string("op_1286_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_1288_to_fp16 = const()[name = string("op_1288_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88930176)))];
            tensor<fp16, [8, 128, 1, 188]> var_1289_cast_fp16 = add(x = var_1286_cast_fp16, y = var_1288_to_fp16)[name = string("op_1289_cast_fp16")];
            tensor<int32, [4]> var_1290 = const()[name = string("op_1290"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_9_cast_fp16 = reshape(shape = var_1290, x = k_9_cast_fp16)[name = string("kh_9_cast_fp16")];
            tensor<int32, [4]> var_1292 = const()[name = string("op_1292"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_9_cast_fp16 = reshape(shape = var_1292, x = v_9_cast_fp16)[name = string("vh_9_cast_fp16")];
            tensor<int32, [4]> var_1294 = const()[name = string("op_1294"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_9_equation_0 = const()[name = string("ac_9_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_1295_cast_fp16 = transpose(perm = var_1294, x = kh_9_cast_fp16)[name = string("transpose_118")];
            tensor<fp16, [8, 188, 1, 188]> ac_9_cast_fp16 = einsum(equation = ac_9_equation_0, values = (var_1295_cast_fp16, var_1289_cast_fp16))[name = string("ac_9_cast_fp16")];
            tensor<int32, [3]> var_1298_perm_0 = const()[name = string("op_1298_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1299_axes_0 = const()[name = string("op_1299_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_1298_cast_fp16 = transpose(perm = var_1298_perm_0, x = bd_all_19_cast_fp16)[name = string("transpose_117")];
            tensor<fp16, [8, 188, 1, 188]> var_1299_cast_fp16 = expand_dims(axes = var_1299_axes_0, x = var_1298_cast_fp16)[name = string("op_1299_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_1300_cast_fp16 = add(x = ac_9_cast_fp16, y = var_1299_cast_fp16)[name = string("op_1300_cast_fp16")];
            fp16 var_1301_to_fp16 = const()[name = string("op_1301_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_17_cast_fp16 = mul(x = var_1300_cast_fp16, y = var_1301_to_fp16)[name = string("scores_17_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_19_cast_fp16 = add(x = scores_17_cast_fp16, y = key_bias)[name = string("scores_19_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_1304_cast_fp16 = softmax(axis = var_1175, x = scores_19_cast_fp16)[name = string("op_1304_cast_fp16")];
            tensor<int32, [4]> transpose_52_perm_0 = const()[name = string("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_8_perm_0 = const()[name = string("transpose_8_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_44 = const()[name = string("concat_44"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_8_cast_fp16 = transpose(perm = transpose_8_perm_0, x = var_1304_cast_fp16)[name = string("transpose_116")];
            tensor<fp16, [8, 188, 188]> reshape_12_cast_fp16 = reshape(shape = concat_44, x = transpose_8_cast_fp16)[name = string("reshape_12_cast_fp16")];
            tensor<int32, [3]> concat_45 = const()[name = string("concat_45"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_52_cast_fp16 = transpose(perm = transpose_52_perm_0, x = vh_9_cast_fp16)[name = string("transpose_115")];
            tensor<fp16, [8, 188, 128]> reshape_13_cast_fp16 = reshape(shape = concat_45, x = transpose_52_cast_fp16)[name = string("reshape_13_cast_fp16")];
            bool matmul_4_transpose_x_0 = const()[name = string("matmul_4_transpose_x_0"), val = bool(false)];
            bool matmul_4_transpose_y_0 = const()[name = string("matmul_4_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = reshape_12_cast_fp16, y = reshape_13_cast_fp16)[name = string("matmul_4_cast_fp16")];
            tensor<int32, [4]> concat_49 = const()[name = string("concat_49"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_14_cast_fp16 = reshape(shape = concat_49, x = matmul_4_cast_fp16)[name = string("reshape_14_cast_fp16")];
            tensor<int32, [4]> ctx_9_perm_0 = const()[name = string("ctx_9_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_1309 = const()[name = string("op_1309"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_9_cast_fp16 = transpose(perm = ctx_9_perm_0, x = reshape_14_cast_fp16)[name = string("transpose_114")];
            tensor<fp16, [1, 1024, 1, 188]> input_129_cast_fp16 = reshape(shape = var_1309, x = ctx_9_cast_fp16)[name = string("input_129_cast_fp16")];
            string var_1316_pad_type_0 = const()[name = string("op_1316_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1316_strides_0 = const()[name = string("op_1316_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1316_pad_0 = const()[name = string("op_1316_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1316_dilations_0 = const()[name = string("op_1316_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1316_groups_0 = const()[name = string("op_1316_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(88932288))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89718784))))[name = string("layers_4_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1316_cast_fp16 = conv(dilations = var_1316_dilations_0, groups = var_1316_groups_0, pad = var_1316_pad_0, pad_type = var_1316_pad_type_0, strides = var_1316_strides_0, weight = layers_4_self_attn_linear_out_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = string("op_1316_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_105_cast_fp16 = add(x = x_93_cast_fp16, y = var_1316_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<int32, [1]> var_1332_axes_0 = const()[name = string("op_1332_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_4_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_4_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1332_cast_fp16 = layer_norm(axes = var_1332_axes_0, epsilon = layers_4_norm_conv_eps_scaled_to_fp16, x = x_105_cast_fp16)[name = string("op_1332_cast_fp16")];
            tensor<fp16, [1024]> input_131_gamma_0_to_fp16 = const()[name = string("input_131_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89727040)))];
            tensor<fp16, [1024]> input_131_beta_0_to_fp16 = const()[name = string("input_131_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89729152)))];
            fp16 input_131_epsilon_0_to_fp16 = const()[name = string("input_131_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_131_cast_fp16 = batch_norm(beta = input_131_beta_0_to_fp16, epsilon = input_131_epsilon_0_to_fp16, gamma = input_131_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1332_cast_fp16)[name = string("input_131_cast_fp16")];
            string input_133_pad_type_0 = const()[name = string("input_133_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_133_strides_0 = const()[name = string("input_133_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_133_pad_0 = const()[name = string("input_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_133_dilations_0 = const()[name = string("input_133_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_133_groups_0 = const()[name = string("input_133_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_4_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89731264))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91304192))))[name = string("layers_4_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_133_cast_fp16 = conv(dilations = input_133_dilations_0, groups = input_133_groups_0, pad = input_133_pad_0, pad_type = input_133_pad_type_0, strides = input_133_strides_0, weight = layers_4_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_131_cast_fp16)[name = string("input_133_cast_fp16")];
            int32 x_107_split_num_splits_0 = const()[name = string("x_107_split_num_splits_0"), val = int32(2)];
            int32 x_107_split_axis_0 = const()[name = string("x_107_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_107_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_133_cast_fp16)[name = string("x_107_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_107_split_1_sigmoid_cast_fp16 = sigmoid(x = x_107_split_cast_fp16_1)[name = string("x_107_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_107_cast_fp16 = mul(x = x_107_split_cast_fp16_0, y = x_107_split_1_sigmoid_cast_fp16)[name = string("x_107_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_135_cast_fp16 = mul(x = x_107_cast_fp16, y = pad_mask)[name = string("input_135_cast_fp16")];
            string input_137_pad_type_0 = const()[name = string("input_137_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_137_pad_0 = const()[name = string("input_137_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_137_groups_0 = const()[name = string("input_137_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_137_strides_0 = const()[name = string("input_137_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_137_dilations_0 = const()[name = string("input_137_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_111_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91320640))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91327616))))[name = string("const_111_to_fp16_palettized")];
            tensor<fp16, [1024]> const_112_to_fp16 = const()[name = string("const_112_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91335872)))];
            tensor<fp16, [1, 1024, 1, 188]> input_139_cast_fp16 = conv(bias = const_112_to_fp16, dilations = input_137_dilations_0, groups = input_137_groups_0, pad = input_137_pad_0, pad_type = input_137_pad_type_0, strides = input_137_strides_0, weight = const_111_to_fp16_palettized, x = input_135_cast_fp16)[name = string("input_139_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_141_cast_fp16 = silu(x = input_139_cast_fp16)[name = string("input_141_cast_fp16")];
            string var_1364_pad_type_0 = const()[name = string("op_1364_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1364_strides_0 = const()[name = string("op_1364_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1364_pad_0 = const()[name = string("op_1364_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1364_dilations_0 = const()[name = string("op_1364_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1364_groups_0 = const()[name = string("op_1364_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_4_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91337984))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92124480))))[name = string("layers_4_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1364_cast_fp16 = conv(dilations = var_1364_dilations_0, groups = var_1364_groups_0, pad = var_1364_pad_0, pad_type = var_1364_pad_type_0, strides = var_1364_strides_0, weight = layers_4_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = string("op_1364_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_109_cast_fp16 = add(x = x_105_cast_fp16, y = var_1364_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<int32, [1]> var_1380_axes_0 = const()[name = string("op_1380_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_4_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_4_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1380_cast_fp16 = layer_norm(axes = var_1380_axes_0, epsilon = layers_4_norm_feed_forward2_eps_scaled_to_fp16, x = x_109_cast_fp16)[name = string("op_1380_cast_fp16")];
            tensor<fp16, [1024]> input_143_gamma_0_to_fp16 = const()[name = string("input_143_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92132736)))];
            tensor<fp16, [1024]> input_143_beta_0_to_fp16 = const()[name = string("input_143_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92134848)))];
            fp16 input_143_epsilon_0_to_fp16 = const()[name = string("input_143_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_143_cast_fp16 = batch_norm(beta = input_143_beta_0_to_fp16, epsilon = input_143_epsilon_0_to_fp16, gamma = input_143_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1380_cast_fp16)[name = string("input_143_cast_fp16")];
            string input_145_pad_type_0 = const()[name = string("input_145_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_145_strides_0 = const()[name = string("input_145_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_145_pad_0 = const()[name = string("input_145_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_145_dilations_0 = const()[name = string("input_145_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_145_groups_0 = const()[name = string("input_145_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_4_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92136960))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95282752))))[name = string("layers_4_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_145_cast_fp16 = conv(dilations = input_145_dilations_0, groups = input_145_groups_0, pad = input_145_pad_0, pad_type = input_145_pad_type_0, strides = input_145_strides_0, weight = layers_4_feed_forward2_linear1_weight_to_fp16_palettized, x = input_143_cast_fp16)[name = string("input_145_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_147_cast_fp16 = silu(x = input_145_cast_fp16)[name = string("input_147_cast_fp16")];
            string var_1397_pad_type_0 = const()[name = string("op_1397_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1397_strides_0 = const()[name = string("op_1397_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1397_pad_0 = const()[name = string("op_1397_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1397_dilations_0 = const()[name = string("op_1397_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1397_groups_0 = const()[name = string("op_1397_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_1398_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95315584))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98461376))))[name = string("op_1398_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1398_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_1397_dilations_0, groups = var_1397_groups_0, pad = var_1397_pad_0, pad_type = var_1397_pad_type_0, strides = var_1397_strides_0, weight = op_1398_weight_0_to_fp16_palettized, x = input_147_cast_fp16)[name = string("op_1398_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_111_cast_fp16 = add(x = x_109_cast_fp16, y = var_1398_cast_fp16)[name = string("x_111_cast_fp16")];
            tensor<int32, [1]> var_1414_axes_0 = const()[name = string("op_1414_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_4_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_4_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1414_cast_fp16 = layer_norm(axes = var_1414_axes_0, epsilon = layers_4_norm_out_eps_scaled_to_fp16, x = x_111_cast_fp16)[name = string("op_1414_cast_fp16")];
            tensor<fp16, [1024]> x_113_gamma_0_to_fp16 = const()[name = string("x_113_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98469632)))];
            tensor<fp16, [1024]> x_113_beta_0_to_fp16 = const()[name = string("x_113_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98471744)))];
            fp16 x_113_epsilon_0_to_fp16 = const()[name = string("x_113_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_113_cast_fp16 = batch_norm(beta = x_113_beta_0_to_fp16, epsilon = x_113_epsilon_0_to_fp16, gamma = x_113_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1414_cast_fp16)[name = string("x_113_cast_fp16")];
            int32 var_1433 = const()[name = string("op_1433"), val = int32(1)];
            tensor<int32, [1]> var_1460_axes_0 = const()[name = string("op_1460_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_5_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_5_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1460_cast_fp16 = layer_norm(axes = var_1460_axes_0, epsilon = layers_5_norm_feed_forward1_eps_scaled_to_fp16, x = x_113_cast_fp16)[name = string("op_1460_cast_fp16")];
            tensor<fp16, [1024]> input_149_gamma_0_to_fp16 = const()[name = string("input_149_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98473856)))];
            tensor<fp16, [1024]> input_149_beta_0_to_fp16 = const()[name = string("input_149_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98475968)))];
            fp16 input_149_epsilon_0_to_fp16 = const()[name = string("input_149_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_149_cast_fp16 = batch_norm(beta = input_149_beta_0_to_fp16, epsilon = input_149_epsilon_0_to_fp16, gamma = input_149_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1460_cast_fp16)[name = string("input_149_cast_fp16")];
            string input_151_pad_type_0 = const()[name = string("input_151_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_151_strides_0 = const()[name = string("input_151_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_151_pad_0 = const()[name = string("input_151_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_151_dilations_0 = const()[name = string("input_151_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_151_groups_0 = const()[name = string("input_151_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_5_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(98478080))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101623872))))[name = string("layers_5_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_151_cast_fp16 = conv(dilations = input_151_dilations_0, groups = input_151_groups_0, pad = input_151_pad_0, pad_type = input_151_pad_type_0, strides = input_151_strides_0, weight = layers_5_feed_forward1_linear1_weight_to_fp16_palettized, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_153_cast_fp16 = silu(x = input_151_cast_fp16)[name = string("input_153_cast_fp16")];
            string var_1477_pad_type_0 = const()[name = string("op_1477_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1477_strides_0 = const()[name = string("op_1477_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1477_pad_0 = const()[name = string("op_1477_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1477_dilations_0 = const()[name = string("op_1477_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1477_groups_0 = const()[name = string("op_1477_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_1478_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101656704))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104802496))))[name = string("op_1478_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1478_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_1477_dilations_0, groups = var_1477_groups_0, pad = var_1477_pad_0, pad_type = var_1477_pad_type_0, strides = var_1477_strides_0, weight = op_1478_weight_0_to_fp16_palettized, x = input_153_cast_fp16)[name = string("op_1478_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_115_cast_fp16 = add(x = x_113_cast_fp16, y = var_1478_cast_fp16)[name = string("x_115_cast_fp16")];
            tensor<int32, [1]> var_1494_axes_0 = const()[name = string("op_1494_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_5_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_5_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1494_cast_fp16 = layer_norm(axes = var_1494_axes_0, epsilon = layers_5_norm_self_att_eps_scaled_to_fp16, x = x_115_cast_fp16)[name = string("op_1494_cast_fp16")];
            tensor<fp16, [1024]> x_117_gamma_0_to_fp16 = const()[name = string("x_117_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104810752)))];
            tensor<fp16, [1024]> x_117_beta_0_to_fp16 = const()[name = string("x_117_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104812864)))];
            fp16 x_117_epsilon_0_to_fp16 = const()[name = string("x_117_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_117_cast_fp16 = batch_norm(beta = x_117_beta_0_to_fp16, epsilon = x_117_epsilon_0_to_fp16, gamma = x_117_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1494_cast_fp16)[name = string("x_117_cast_fp16")];
            string q_11_pad_type_0 = const()[name = string("q_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_11_strides_0 = const()[name = string("q_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_11_pad_0 = const()[name = string("q_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_11_dilations_0 = const()[name = string("q_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_11_groups_0 = const()[name = string("q_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104814976))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105601472))))[name = string("layers_5_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_11_cast_fp16 = conv(dilations = q_11_dilations_0, groups = q_11_groups_0, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = q_11_strides_0, weight = layers_5_self_attn_linear_q_weight_to_fp16_palettized, x = x_117_cast_fp16)[name = string("q_11_cast_fp16")];
            string k_11_pad_type_0 = const()[name = string("k_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_11_strides_0 = const()[name = string("k_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_11_pad_0 = const()[name = string("k_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_11_dilations_0 = const()[name = string("k_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_11_groups_0 = const()[name = string("k_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105609728))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106396224))))[name = string("layers_5_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_11_cast_fp16 = conv(dilations = k_11_dilations_0, groups = k_11_groups_0, pad = k_11_pad_0, pad_type = k_11_pad_type_0, strides = k_11_strides_0, weight = layers_5_self_attn_linear_k_weight_to_fp16_palettized, x = x_117_cast_fp16)[name = string("k_11_cast_fp16")];
            string v_11_pad_type_0 = const()[name = string("v_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_11_strides_0 = const()[name = string("v_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_11_pad_0 = const()[name = string("v_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_11_dilations_0 = const()[name = string("v_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_11_groups_0 = const()[name = string("v_11_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106404480))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107190976))))[name = string("layers_5_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_11_cast_fp16 = conv(dilations = v_11_dilations_0, groups = v_11_groups_0, pad = v_11_pad_0, pad_type = v_11_pad_type_0, strides = v_11_strides_0, weight = layers_5_self_attn_linear_v_weight_to_fp16_palettized, x = x_117_cast_fp16)[name = string("v_11_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_11_to_fp16 = const()[name = string("bv_all_11_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107199232)))];
            tensor<fp16, [1, 1024, 1, 188]> var_1526_cast_fp16 = add(x = q_11_cast_fp16, y = bv_all_11_to_fp16)[name = string("op_1526_cast_fp16")];
            tensor<int32, [3]> var_1527 = const()[name = string("op_1527"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_11_cast_fp16 = reshape(shape = var_1527, x = var_1526_cast_fp16)[name = string("qb_11_cast_fp16")];
            bool bd_all_21_transpose_x_0 = const()[name = string("bd_all_21_transpose_x_0"), val = bool(false)];
            bool bd_all_21_transpose_y_0 = const()[name = string("bd_all_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_5_self_attn_pos_proj_to_fp16 = const()[name = string("layers_5_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107201344)))];
            tensor<fp16, [8, 375, 188]> bd_all_21_cast_fp16 = matmul(transpose_x = bd_all_21_transpose_x_0, transpose_y = bd_all_21_transpose_y_0, x = layers_5_self_attn_pos_proj_to_fp16, y = qb_11_cast_fp16)[name = string("bd_all_21_cast_fp16")];
            tensor<int32, [3]> x_119_perm_0 = const()[name = string("x_119_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_121_pad_0 = const()[name = string("x_121_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_121_mode_0 = const()[name = string("x_121_mode_0"), val = string("constant")];
            fp16 const_30_to_fp16 = const()[name = string("const_30_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_119_cast_fp16 = transpose(perm = x_119_perm_0, x = bd_all_21_cast_fp16)[name = string("transpose_113")];
            tensor<fp16, [8, 188, 376]> x_121_cast_fp16 = pad(constant_val = const_30_to_fp16, mode = x_121_mode_0, pad = x_121_pad_0, x = x_119_cast_fp16)[name = string("x_121_cast_fp16")];
            tensor<int32, [3]> var_1534 = const()[name = string("op_1534"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_123_cast_fp16 = reshape(shape = var_1534, x = x_121_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<int32, [3]> var_1537_begin_0 = const()[name = string("op_1537_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_1537_end_0 = const()[name = string("op_1537_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_1537_end_mask_0 = const()[name = string("op_1537_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_1537_cast_fp16 = slice_by_index(begin = var_1537_begin_0, end = var_1537_end_0, end_mask = var_1537_end_mask_0, x = x_123_cast_fp16)[name = string("op_1537_cast_fp16")];
            tensor<int32, [3]> var_1538 = const()[name = string("op_1538"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_125_cast_fp16 = reshape(shape = var_1538, x = var_1537_cast_fp16)[name = string("x_125_cast_fp16")];
            tensor<int32, [3]> bd_all_23_begin_0 = const()[name = string("bd_all_23_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_23_end_0 = const()[name = string("bd_all_23_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_23_end_mask_0 = const()[name = string("bd_all_23_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_23_cast_fp16 = slice_by_index(begin = bd_all_23_begin_0, end = bd_all_23_end_0, end_mask = bd_all_23_end_mask_0, x = x_125_cast_fp16)[name = string("bd_all_23_cast_fp16")];
            tensor<int32, [4]> var_1543 = const()[name = string("op_1543"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_1544_cast_fp16 = reshape(shape = var_1543, x = q_11_cast_fp16)[name = string("op_1544_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_1546_to_fp16 = const()[name = string("op_1546_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107969408)))];
            tensor<fp16, [8, 128, 1, 188]> var_1547_cast_fp16 = add(x = var_1544_cast_fp16, y = var_1546_to_fp16)[name = string("op_1547_cast_fp16")];
            tensor<int32, [4]> var_1548 = const()[name = string("op_1548"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_11_cast_fp16 = reshape(shape = var_1548, x = k_11_cast_fp16)[name = string("kh_11_cast_fp16")];
            tensor<int32, [4]> var_1550 = const()[name = string("op_1550"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_11_cast_fp16 = reshape(shape = var_1550, x = v_11_cast_fp16)[name = string("vh_11_cast_fp16")];
            tensor<int32, [4]> var_1552 = const()[name = string("op_1552"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_11_equation_0 = const()[name = string("ac_11_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_1553_cast_fp16 = transpose(perm = var_1552, x = kh_11_cast_fp16)[name = string("transpose_112")];
            tensor<fp16, [8, 188, 1, 188]> ac_11_cast_fp16 = einsum(equation = ac_11_equation_0, values = (var_1553_cast_fp16, var_1547_cast_fp16))[name = string("ac_11_cast_fp16")];
            tensor<int32, [3]> var_1556_perm_0 = const()[name = string("op_1556_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1557_axes_0 = const()[name = string("op_1557_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_1556_cast_fp16 = transpose(perm = var_1556_perm_0, x = bd_all_23_cast_fp16)[name = string("transpose_111")];
            tensor<fp16, [8, 188, 1, 188]> var_1557_cast_fp16 = expand_dims(axes = var_1557_axes_0, x = var_1556_cast_fp16)[name = string("op_1557_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_1558_cast_fp16 = add(x = ac_11_cast_fp16, y = var_1557_cast_fp16)[name = string("op_1558_cast_fp16")];
            fp16 var_1559_to_fp16 = const()[name = string("op_1559_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_21_cast_fp16 = mul(x = var_1558_cast_fp16, y = var_1559_to_fp16)[name = string("scores_21_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_23_cast_fp16 = add(x = scores_21_cast_fp16, y = key_bias)[name = string("scores_23_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_1562_cast_fp16 = softmax(axis = var_1433, x = scores_23_cast_fp16)[name = string("op_1562_cast_fp16")];
            tensor<int32, [4]> transpose_53_perm_0 = const()[name = string("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_10_perm_0 = const()[name = string("transpose_10_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_54 = const()[name = string("concat_54"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_10_cast_fp16 = transpose(perm = transpose_10_perm_0, x = var_1562_cast_fp16)[name = string("transpose_110")];
            tensor<fp16, [8, 188, 188]> reshape_15_cast_fp16 = reshape(shape = concat_54, x = transpose_10_cast_fp16)[name = string("reshape_15_cast_fp16")];
            tensor<int32, [3]> concat_55 = const()[name = string("concat_55"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_53_cast_fp16 = transpose(perm = transpose_53_perm_0, x = vh_11_cast_fp16)[name = string("transpose_109")];
            tensor<fp16, [8, 188, 128]> reshape_16_cast_fp16 = reshape(shape = concat_55, x = transpose_53_cast_fp16)[name = string("reshape_16_cast_fp16")];
            bool matmul_5_transpose_x_0 = const()[name = string("matmul_5_transpose_x_0"), val = bool(false)];
            bool matmul_5_transpose_y_0 = const()[name = string("matmul_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = reshape_15_cast_fp16, y = reshape_16_cast_fp16)[name = string("matmul_5_cast_fp16")];
            tensor<int32, [4]> concat_59 = const()[name = string("concat_59"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_17_cast_fp16 = reshape(shape = concat_59, x = matmul_5_cast_fp16)[name = string("reshape_17_cast_fp16")];
            tensor<int32, [4]> ctx_11_perm_0 = const()[name = string("ctx_11_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_1567 = const()[name = string("op_1567"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_11_cast_fp16 = transpose(perm = ctx_11_perm_0, x = reshape_17_cast_fp16)[name = string("transpose_108")];
            tensor<fp16, [1, 1024, 1, 188]> input_155_cast_fp16 = reshape(shape = var_1567, x = ctx_11_cast_fp16)[name = string("input_155_cast_fp16")];
            string var_1574_pad_type_0 = const()[name = string("op_1574_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1574_strides_0 = const()[name = string("op_1574_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1574_pad_0 = const()[name = string("op_1574_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1574_dilations_0 = const()[name = string("op_1574_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1574_groups_0 = const()[name = string("op_1574_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107971520))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108758016))))[name = string("layers_5_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1574_cast_fp16 = conv(dilations = var_1574_dilations_0, groups = var_1574_groups_0, pad = var_1574_pad_0, pad_type = var_1574_pad_type_0, strides = var_1574_strides_0, weight = layers_5_self_attn_linear_out_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = string("op_1574_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_127_cast_fp16 = add(x = x_115_cast_fp16, y = var_1574_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<int32, [1]> var_1590_axes_0 = const()[name = string("op_1590_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_5_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_5_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1590_cast_fp16 = layer_norm(axes = var_1590_axes_0, epsilon = layers_5_norm_conv_eps_scaled_to_fp16, x = x_127_cast_fp16)[name = string("op_1590_cast_fp16")];
            tensor<fp16, [1024]> input_157_gamma_0_to_fp16 = const()[name = string("input_157_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108766272)))];
            tensor<fp16, [1024]> input_157_beta_0_to_fp16 = const()[name = string("input_157_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108768384)))];
            fp16 input_157_epsilon_0_to_fp16 = const()[name = string("input_157_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_157_cast_fp16 = batch_norm(beta = input_157_beta_0_to_fp16, epsilon = input_157_epsilon_0_to_fp16, gamma = input_157_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1590_cast_fp16)[name = string("input_157_cast_fp16")];
            string input_159_pad_type_0 = const()[name = string("input_159_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_159_strides_0 = const()[name = string("input_159_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_159_pad_0 = const()[name = string("input_159_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_159_dilations_0 = const()[name = string("input_159_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_159_groups_0 = const()[name = string("input_159_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_5_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108770496))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110343424))))[name = string("layers_5_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_159_cast_fp16 = conv(dilations = input_159_dilations_0, groups = input_159_groups_0, pad = input_159_pad_0, pad_type = input_159_pad_type_0, strides = input_159_strides_0, weight = layers_5_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_157_cast_fp16)[name = string("input_159_cast_fp16")];
            int32 x_129_split_num_splits_0 = const()[name = string("x_129_split_num_splits_0"), val = int32(2)];
            int32 x_129_split_axis_0 = const()[name = string("x_129_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_129_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_159_cast_fp16)[name = string("x_129_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_129_split_1_sigmoid_cast_fp16 = sigmoid(x = x_129_split_cast_fp16_1)[name = string("x_129_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_129_cast_fp16 = mul(x = x_129_split_cast_fp16_0, y = x_129_split_1_sigmoid_cast_fp16)[name = string("x_129_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_161_cast_fp16 = mul(x = x_129_cast_fp16, y = pad_mask)[name = string("input_161_cast_fp16")];
            string input_163_pad_type_0 = const()[name = string("input_163_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_163_pad_0 = const()[name = string("input_163_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_163_groups_0 = const()[name = string("input_163_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_163_strides_0 = const()[name = string("input_163_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_163_dilations_0 = const()[name = string("input_163_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_113_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110359872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110366848))))[name = string("const_113_to_fp16_palettized")];
            tensor<fp16, [1024]> const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110375104)))];
            tensor<fp16, [1, 1024, 1, 188]> input_165_cast_fp16 = conv(bias = const_114_to_fp16, dilations = input_163_dilations_0, groups = input_163_groups_0, pad = input_163_pad_0, pad_type = input_163_pad_type_0, strides = input_163_strides_0, weight = const_113_to_fp16_palettized, x = input_161_cast_fp16)[name = string("input_165_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_167_cast_fp16 = silu(x = input_165_cast_fp16)[name = string("input_167_cast_fp16")];
            string var_1622_pad_type_0 = const()[name = string("op_1622_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1622_strides_0 = const()[name = string("op_1622_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1622_pad_0 = const()[name = string("op_1622_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1622_dilations_0 = const()[name = string("op_1622_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1622_groups_0 = const()[name = string("op_1622_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_5_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110377216))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111163712))))[name = string("layers_5_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1622_cast_fp16 = conv(dilations = var_1622_dilations_0, groups = var_1622_groups_0, pad = var_1622_pad_0, pad_type = var_1622_pad_type_0, strides = var_1622_strides_0, weight = layers_5_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = string("op_1622_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_131_cast_fp16 = add(x = x_127_cast_fp16, y = var_1622_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<int32, [1]> var_1638_axes_0 = const()[name = string("op_1638_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_5_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_5_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1638_cast_fp16 = layer_norm(axes = var_1638_axes_0, epsilon = layers_5_norm_feed_forward2_eps_scaled_to_fp16, x = x_131_cast_fp16)[name = string("op_1638_cast_fp16")];
            tensor<fp16, [1024]> input_169_gamma_0_to_fp16 = const()[name = string("input_169_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111171968)))];
            tensor<fp16, [1024]> input_169_beta_0_to_fp16 = const()[name = string("input_169_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111174080)))];
            fp16 input_169_epsilon_0_to_fp16 = const()[name = string("input_169_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_169_cast_fp16 = batch_norm(beta = input_169_beta_0_to_fp16, epsilon = input_169_epsilon_0_to_fp16, gamma = input_169_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1638_cast_fp16)[name = string("input_169_cast_fp16")];
            string input_171_pad_type_0 = const()[name = string("input_171_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_171_strides_0 = const()[name = string("input_171_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_171_pad_0 = const()[name = string("input_171_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_171_dilations_0 = const()[name = string("input_171_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_171_groups_0 = const()[name = string("input_171_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_5_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111176192))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114321984))))[name = string("layers_5_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_171_cast_fp16 = conv(dilations = input_171_dilations_0, groups = input_171_groups_0, pad = input_171_pad_0, pad_type = input_171_pad_type_0, strides = input_171_strides_0, weight = layers_5_feed_forward2_linear1_weight_to_fp16_palettized, x = input_169_cast_fp16)[name = string("input_171_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_173_cast_fp16 = silu(x = input_171_cast_fp16)[name = string("input_173_cast_fp16")];
            string var_1655_pad_type_0 = const()[name = string("op_1655_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1655_strides_0 = const()[name = string("op_1655_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1655_pad_0 = const()[name = string("op_1655_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1655_dilations_0 = const()[name = string("op_1655_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1655_groups_0 = const()[name = string("op_1655_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_1656_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114354816))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117500608))))[name = string("op_1656_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1656_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_1655_dilations_0, groups = var_1655_groups_0, pad = var_1655_pad_0, pad_type = var_1655_pad_type_0, strides = var_1655_strides_0, weight = op_1656_weight_0_to_fp16_palettized, x = input_173_cast_fp16)[name = string("op_1656_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_133_cast_fp16 = add(x = x_131_cast_fp16, y = var_1656_cast_fp16)[name = string("x_133_cast_fp16")];
            tensor<int32, [1]> var_1672_axes_0 = const()[name = string("op_1672_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_5_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_5_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1672_cast_fp16 = layer_norm(axes = var_1672_axes_0, epsilon = layers_5_norm_out_eps_scaled_to_fp16, x = x_133_cast_fp16)[name = string("op_1672_cast_fp16")];
            tensor<fp16, [1024]> x_135_gamma_0_to_fp16 = const()[name = string("x_135_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117508864)))];
            tensor<fp16, [1024]> x_135_beta_0_to_fp16 = const()[name = string("x_135_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117510976)))];
            fp16 x_135_epsilon_0_to_fp16 = const()[name = string("x_135_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_135_cast_fp16 = batch_norm(beta = x_135_beta_0_to_fp16, epsilon = x_135_epsilon_0_to_fp16, gamma = x_135_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1672_cast_fp16)[name = string("x_135_cast_fp16")];
            int32 var_1691 = const()[name = string("op_1691"), val = int32(1)];
            tensor<int32, [1]> var_1718_axes_0 = const()[name = string("op_1718_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_6_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_6_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1718_cast_fp16 = layer_norm(axes = var_1718_axes_0, epsilon = layers_6_norm_feed_forward1_eps_scaled_to_fp16, x = x_135_cast_fp16)[name = string("op_1718_cast_fp16")];
            tensor<fp16, [1024]> input_175_gamma_0_to_fp16 = const()[name = string("input_175_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117513088)))];
            tensor<fp16, [1024]> input_175_beta_0_to_fp16 = const()[name = string("input_175_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117515200)))];
            fp16 input_175_epsilon_0_to_fp16 = const()[name = string("input_175_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_175_cast_fp16 = batch_norm(beta = input_175_beta_0_to_fp16, epsilon = input_175_epsilon_0_to_fp16, gamma = input_175_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1718_cast_fp16)[name = string("input_175_cast_fp16")];
            string input_177_pad_type_0 = const()[name = string("input_177_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_177_strides_0 = const()[name = string("input_177_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_177_pad_0 = const()[name = string("input_177_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_177_dilations_0 = const()[name = string("input_177_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_177_groups_0 = const()[name = string("input_177_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_6_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117517312))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120663104))))[name = string("layers_6_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_177_cast_fp16 = conv(dilations = input_177_dilations_0, groups = input_177_groups_0, pad = input_177_pad_0, pad_type = input_177_pad_type_0, strides = input_177_strides_0, weight = layers_6_feed_forward1_linear1_weight_to_fp16_palettized, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_179_cast_fp16 = silu(x = input_177_cast_fp16)[name = string("input_179_cast_fp16")];
            string var_1735_pad_type_0 = const()[name = string("op_1735_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1735_strides_0 = const()[name = string("op_1735_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1735_pad_0 = const()[name = string("op_1735_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1735_dilations_0 = const()[name = string("op_1735_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1735_groups_0 = const()[name = string("op_1735_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_1736_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120695936))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123841728))))[name = string("op_1736_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1736_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_1735_dilations_0, groups = var_1735_groups_0, pad = var_1735_pad_0, pad_type = var_1735_pad_type_0, strides = var_1735_strides_0, weight = op_1736_weight_0_to_fp16_palettized, x = input_179_cast_fp16)[name = string("op_1736_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_137_cast_fp16 = add(x = x_135_cast_fp16, y = var_1736_cast_fp16)[name = string("x_137_cast_fp16")];
            tensor<int32, [1]> var_1752_axes_0 = const()[name = string("op_1752_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_6_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_6_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1752_cast_fp16 = layer_norm(axes = var_1752_axes_0, epsilon = layers_6_norm_self_att_eps_scaled_to_fp16, x = x_137_cast_fp16)[name = string("op_1752_cast_fp16")];
            tensor<fp16, [1024]> x_139_gamma_0_to_fp16 = const()[name = string("x_139_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123849984)))];
            tensor<fp16, [1024]> x_139_beta_0_to_fp16 = const()[name = string("x_139_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123852096)))];
            fp16 x_139_epsilon_0_to_fp16 = const()[name = string("x_139_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_139_cast_fp16 = batch_norm(beta = x_139_beta_0_to_fp16, epsilon = x_139_epsilon_0_to_fp16, gamma = x_139_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1752_cast_fp16)[name = string("x_139_cast_fp16")];
            string q_13_pad_type_0 = const()[name = string("q_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_13_strides_0 = const()[name = string("q_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_13_pad_0 = const()[name = string("q_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_13_dilations_0 = const()[name = string("q_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_13_groups_0 = const()[name = string("q_13_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123854208))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124640704))))[name = string("layers_6_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_13_cast_fp16 = conv(dilations = q_13_dilations_0, groups = q_13_groups_0, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = q_13_strides_0, weight = layers_6_self_attn_linear_q_weight_to_fp16_palettized, x = x_139_cast_fp16)[name = string("q_13_cast_fp16")];
            string k_13_pad_type_0 = const()[name = string("k_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_13_strides_0 = const()[name = string("k_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_13_pad_0 = const()[name = string("k_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_13_dilations_0 = const()[name = string("k_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_13_groups_0 = const()[name = string("k_13_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124648960))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125435456))))[name = string("layers_6_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_13_cast_fp16 = conv(dilations = k_13_dilations_0, groups = k_13_groups_0, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = k_13_strides_0, weight = layers_6_self_attn_linear_k_weight_to_fp16_palettized, x = x_139_cast_fp16)[name = string("k_13_cast_fp16")];
            string v_13_pad_type_0 = const()[name = string("v_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_13_strides_0 = const()[name = string("v_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_13_pad_0 = const()[name = string("v_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_13_dilations_0 = const()[name = string("v_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_13_groups_0 = const()[name = string("v_13_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(125443712))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126230208))))[name = string("layers_6_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_13_cast_fp16 = conv(dilations = v_13_dilations_0, groups = v_13_groups_0, pad = v_13_pad_0, pad_type = v_13_pad_type_0, strides = v_13_strides_0, weight = layers_6_self_attn_linear_v_weight_to_fp16_palettized, x = x_139_cast_fp16)[name = string("v_13_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_13_to_fp16 = const()[name = string("bv_all_13_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126238464)))];
            tensor<fp16, [1, 1024, 1, 188]> var_1784_cast_fp16 = add(x = q_13_cast_fp16, y = bv_all_13_to_fp16)[name = string("op_1784_cast_fp16")];
            tensor<int32, [3]> var_1785 = const()[name = string("op_1785"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_13_cast_fp16 = reshape(shape = var_1785, x = var_1784_cast_fp16)[name = string("qb_13_cast_fp16")];
            bool bd_all_25_transpose_x_0 = const()[name = string("bd_all_25_transpose_x_0"), val = bool(false)];
            bool bd_all_25_transpose_y_0 = const()[name = string("bd_all_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_6_self_attn_pos_proj_to_fp16 = const()[name = string("layers_6_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126240576)))];
            tensor<fp16, [8, 375, 188]> bd_all_25_cast_fp16 = matmul(transpose_x = bd_all_25_transpose_x_0, transpose_y = bd_all_25_transpose_y_0, x = layers_6_self_attn_pos_proj_to_fp16, y = qb_13_cast_fp16)[name = string("bd_all_25_cast_fp16")];
            tensor<int32, [3]> x_141_perm_0 = const()[name = string("x_141_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_143_pad_0 = const()[name = string("x_143_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_143_mode_0 = const()[name = string("x_143_mode_0"), val = string("constant")];
            fp16 const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_141_cast_fp16 = transpose(perm = x_141_perm_0, x = bd_all_25_cast_fp16)[name = string("transpose_107")];
            tensor<fp16, [8, 188, 376]> x_143_cast_fp16 = pad(constant_val = const_34_to_fp16, mode = x_143_mode_0, pad = x_143_pad_0, x = x_141_cast_fp16)[name = string("x_143_cast_fp16")];
            tensor<int32, [3]> var_1792 = const()[name = string("op_1792"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_145_cast_fp16 = reshape(shape = var_1792, x = x_143_cast_fp16)[name = string("x_145_cast_fp16")];
            tensor<int32, [3]> var_1795_begin_0 = const()[name = string("op_1795_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_1795_end_0 = const()[name = string("op_1795_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_1795_end_mask_0 = const()[name = string("op_1795_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_1795_cast_fp16 = slice_by_index(begin = var_1795_begin_0, end = var_1795_end_0, end_mask = var_1795_end_mask_0, x = x_145_cast_fp16)[name = string("op_1795_cast_fp16")];
            tensor<int32, [3]> var_1796 = const()[name = string("op_1796"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_147_cast_fp16 = reshape(shape = var_1796, x = var_1795_cast_fp16)[name = string("x_147_cast_fp16")];
            tensor<int32, [3]> bd_all_27_begin_0 = const()[name = string("bd_all_27_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_27_end_0 = const()[name = string("bd_all_27_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_27_end_mask_0 = const()[name = string("bd_all_27_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_27_cast_fp16 = slice_by_index(begin = bd_all_27_begin_0, end = bd_all_27_end_0, end_mask = bd_all_27_end_mask_0, x = x_147_cast_fp16)[name = string("bd_all_27_cast_fp16")];
            tensor<int32, [4]> var_1801 = const()[name = string("op_1801"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_1802_cast_fp16 = reshape(shape = var_1801, x = q_13_cast_fp16)[name = string("op_1802_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_1804_to_fp16 = const()[name = string("op_1804_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127008640)))];
            tensor<fp16, [8, 128, 1, 188]> var_1805_cast_fp16 = add(x = var_1802_cast_fp16, y = var_1804_to_fp16)[name = string("op_1805_cast_fp16")];
            tensor<int32, [4]> var_1806 = const()[name = string("op_1806"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_13_cast_fp16 = reshape(shape = var_1806, x = k_13_cast_fp16)[name = string("kh_13_cast_fp16")];
            tensor<int32, [4]> var_1808 = const()[name = string("op_1808"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_13_cast_fp16 = reshape(shape = var_1808, x = v_13_cast_fp16)[name = string("vh_13_cast_fp16")];
            tensor<int32, [4]> var_1810 = const()[name = string("op_1810"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_13_equation_0 = const()[name = string("ac_13_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_1811_cast_fp16 = transpose(perm = var_1810, x = kh_13_cast_fp16)[name = string("transpose_106")];
            tensor<fp16, [8, 188, 1, 188]> ac_13_cast_fp16 = einsum(equation = ac_13_equation_0, values = (var_1811_cast_fp16, var_1805_cast_fp16))[name = string("ac_13_cast_fp16")];
            tensor<int32, [3]> var_1814_perm_0 = const()[name = string("op_1814_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_1815_axes_0 = const()[name = string("op_1815_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_1814_cast_fp16 = transpose(perm = var_1814_perm_0, x = bd_all_27_cast_fp16)[name = string("transpose_105")];
            tensor<fp16, [8, 188, 1, 188]> var_1815_cast_fp16 = expand_dims(axes = var_1815_axes_0, x = var_1814_cast_fp16)[name = string("op_1815_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_1816_cast_fp16 = add(x = ac_13_cast_fp16, y = var_1815_cast_fp16)[name = string("op_1816_cast_fp16")];
            fp16 var_1817_to_fp16 = const()[name = string("op_1817_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_25_cast_fp16 = mul(x = var_1816_cast_fp16, y = var_1817_to_fp16)[name = string("scores_25_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_27_cast_fp16 = add(x = scores_25_cast_fp16, y = key_bias)[name = string("scores_27_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_1820_cast_fp16 = softmax(axis = var_1691, x = scores_27_cast_fp16)[name = string("op_1820_cast_fp16")];
            tensor<int32, [4]> transpose_54_perm_0 = const()[name = string("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_12_perm_0 = const()[name = string("transpose_12_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_64 = const()[name = string("concat_64"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_12_cast_fp16 = transpose(perm = transpose_12_perm_0, x = var_1820_cast_fp16)[name = string("transpose_104")];
            tensor<fp16, [8, 188, 188]> reshape_18_cast_fp16 = reshape(shape = concat_64, x = transpose_12_cast_fp16)[name = string("reshape_18_cast_fp16")];
            tensor<int32, [3]> concat_65 = const()[name = string("concat_65"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_54_cast_fp16 = transpose(perm = transpose_54_perm_0, x = vh_13_cast_fp16)[name = string("transpose_103")];
            tensor<fp16, [8, 188, 128]> reshape_19_cast_fp16 = reshape(shape = concat_65, x = transpose_54_cast_fp16)[name = string("reshape_19_cast_fp16")];
            bool matmul_6_transpose_x_0 = const()[name = string("matmul_6_transpose_x_0"), val = bool(false)];
            bool matmul_6_transpose_y_0 = const()[name = string("matmul_6_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = reshape_18_cast_fp16, y = reshape_19_cast_fp16)[name = string("matmul_6_cast_fp16")];
            tensor<int32, [4]> concat_69 = const()[name = string("concat_69"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_20_cast_fp16 = reshape(shape = concat_69, x = matmul_6_cast_fp16)[name = string("reshape_20_cast_fp16")];
            tensor<int32, [4]> ctx_13_perm_0 = const()[name = string("ctx_13_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_1825 = const()[name = string("op_1825"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_13_cast_fp16 = transpose(perm = ctx_13_perm_0, x = reshape_20_cast_fp16)[name = string("transpose_102")];
            tensor<fp16, [1, 1024, 1, 188]> input_181_cast_fp16 = reshape(shape = var_1825, x = ctx_13_cast_fp16)[name = string("input_181_cast_fp16")];
            string var_1832_pad_type_0 = const()[name = string("op_1832_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1832_strides_0 = const()[name = string("op_1832_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1832_pad_0 = const()[name = string("op_1832_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1832_dilations_0 = const()[name = string("op_1832_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1832_groups_0 = const()[name = string("op_1832_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127010752))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127797248))))[name = string("layers_6_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1832_cast_fp16 = conv(dilations = var_1832_dilations_0, groups = var_1832_groups_0, pad = var_1832_pad_0, pad_type = var_1832_pad_type_0, strides = var_1832_strides_0, weight = layers_6_self_attn_linear_out_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = string("op_1832_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_149_cast_fp16 = add(x = x_137_cast_fp16, y = var_1832_cast_fp16)[name = string("x_149_cast_fp16")];
            tensor<int32, [1]> var_1848_axes_0 = const()[name = string("op_1848_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_6_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_6_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1848_cast_fp16 = layer_norm(axes = var_1848_axes_0, epsilon = layers_6_norm_conv_eps_scaled_to_fp16, x = x_149_cast_fp16)[name = string("op_1848_cast_fp16")];
            tensor<fp16, [1024]> input_183_gamma_0_to_fp16 = const()[name = string("input_183_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127805504)))];
            tensor<fp16, [1024]> input_183_beta_0_to_fp16 = const()[name = string("input_183_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127807616)))];
            fp16 input_183_epsilon_0_to_fp16 = const()[name = string("input_183_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_183_cast_fp16 = batch_norm(beta = input_183_beta_0_to_fp16, epsilon = input_183_epsilon_0_to_fp16, gamma = input_183_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1848_cast_fp16)[name = string("input_183_cast_fp16")];
            string input_185_pad_type_0 = const()[name = string("input_185_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_185_strides_0 = const()[name = string("input_185_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_185_pad_0 = const()[name = string("input_185_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_185_dilations_0 = const()[name = string("input_185_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_185_groups_0 = const()[name = string("input_185_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_6_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127809728))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129382656))))[name = string("layers_6_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_185_cast_fp16 = conv(dilations = input_185_dilations_0, groups = input_185_groups_0, pad = input_185_pad_0, pad_type = input_185_pad_type_0, strides = input_185_strides_0, weight = layers_6_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_183_cast_fp16)[name = string("input_185_cast_fp16")];
            int32 x_151_split_num_splits_0 = const()[name = string("x_151_split_num_splits_0"), val = int32(2)];
            int32 x_151_split_axis_0 = const()[name = string("x_151_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_151_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_185_cast_fp16)[name = string("x_151_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_151_split_1_sigmoid_cast_fp16 = sigmoid(x = x_151_split_cast_fp16_1)[name = string("x_151_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_151_cast_fp16 = mul(x = x_151_split_cast_fp16_0, y = x_151_split_1_sigmoid_cast_fp16)[name = string("x_151_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_187_cast_fp16 = mul(x = x_151_cast_fp16, y = pad_mask)[name = string("input_187_cast_fp16")];
            string input_189_pad_type_0 = const()[name = string("input_189_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_189_pad_0 = const()[name = string("input_189_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_189_groups_0 = const()[name = string("input_189_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_189_strides_0 = const()[name = string("input_189_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_189_dilations_0 = const()[name = string("input_189_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_115_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129399104))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129406080))))[name = string("const_115_to_fp16_palettized")];
            tensor<fp16, [1024]> const_116_to_fp16 = const()[name = string("const_116_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129414336)))];
            tensor<fp16, [1, 1024, 1, 188]> input_191_cast_fp16 = conv(bias = const_116_to_fp16, dilations = input_189_dilations_0, groups = input_189_groups_0, pad = input_189_pad_0, pad_type = input_189_pad_type_0, strides = input_189_strides_0, weight = const_115_to_fp16_palettized, x = input_187_cast_fp16)[name = string("input_191_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_193_cast_fp16 = silu(x = input_191_cast_fp16)[name = string("input_193_cast_fp16")];
            string var_1880_pad_type_0 = const()[name = string("op_1880_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1880_strides_0 = const()[name = string("op_1880_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1880_pad_0 = const()[name = string("op_1880_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1880_dilations_0 = const()[name = string("op_1880_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1880_groups_0 = const()[name = string("op_1880_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_6_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129416448))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130202944))))[name = string("layers_6_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1880_cast_fp16 = conv(dilations = var_1880_dilations_0, groups = var_1880_groups_0, pad = var_1880_pad_0, pad_type = var_1880_pad_type_0, strides = var_1880_strides_0, weight = layers_6_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = string("op_1880_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_153_cast_fp16 = add(x = x_149_cast_fp16, y = var_1880_cast_fp16)[name = string("x_153_cast_fp16")];
            tensor<int32, [1]> var_1896_axes_0 = const()[name = string("op_1896_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_6_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_6_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1896_cast_fp16 = layer_norm(axes = var_1896_axes_0, epsilon = layers_6_norm_feed_forward2_eps_scaled_to_fp16, x = x_153_cast_fp16)[name = string("op_1896_cast_fp16")];
            tensor<fp16, [1024]> input_195_gamma_0_to_fp16 = const()[name = string("input_195_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130211200)))];
            tensor<fp16, [1024]> input_195_beta_0_to_fp16 = const()[name = string("input_195_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130213312)))];
            fp16 input_195_epsilon_0_to_fp16 = const()[name = string("input_195_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_195_cast_fp16 = batch_norm(beta = input_195_beta_0_to_fp16, epsilon = input_195_epsilon_0_to_fp16, gamma = input_195_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1896_cast_fp16)[name = string("input_195_cast_fp16")];
            string input_197_pad_type_0 = const()[name = string("input_197_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_197_strides_0 = const()[name = string("input_197_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_197_pad_0 = const()[name = string("input_197_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_197_dilations_0 = const()[name = string("input_197_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_197_groups_0 = const()[name = string("input_197_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_6_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(130215424))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133361216))))[name = string("layers_6_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_197_cast_fp16 = conv(dilations = input_197_dilations_0, groups = input_197_groups_0, pad = input_197_pad_0, pad_type = input_197_pad_type_0, strides = input_197_strides_0, weight = layers_6_feed_forward2_linear1_weight_to_fp16_palettized, x = input_195_cast_fp16)[name = string("input_197_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_199_cast_fp16 = silu(x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
            string var_1913_pad_type_0 = const()[name = string("op_1913_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1913_strides_0 = const()[name = string("op_1913_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1913_pad_0 = const()[name = string("op_1913_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1913_dilations_0 = const()[name = string("op_1913_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1913_groups_0 = const()[name = string("op_1913_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_1914_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133394048))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136539840))))[name = string("op_1914_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1914_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_1913_dilations_0, groups = var_1913_groups_0, pad = var_1913_pad_0, pad_type = var_1913_pad_type_0, strides = var_1913_strides_0, weight = op_1914_weight_0_to_fp16_palettized, x = input_199_cast_fp16)[name = string("op_1914_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_155_cast_fp16 = add(x = x_153_cast_fp16, y = var_1914_cast_fp16)[name = string("x_155_cast_fp16")];
            tensor<int32, [1]> var_1930_axes_0 = const()[name = string("op_1930_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_6_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_6_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1930_cast_fp16 = layer_norm(axes = var_1930_axes_0, epsilon = layers_6_norm_out_eps_scaled_to_fp16, x = x_155_cast_fp16)[name = string("op_1930_cast_fp16")];
            tensor<fp16, [1024]> x_157_gamma_0_to_fp16 = const()[name = string("x_157_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136548096)))];
            tensor<fp16, [1024]> x_157_beta_0_to_fp16 = const()[name = string("x_157_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136550208)))];
            fp16 x_157_epsilon_0_to_fp16 = const()[name = string("x_157_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_157_cast_fp16 = batch_norm(beta = x_157_beta_0_to_fp16, epsilon = x_157_epsilon_0_to_fp16, gamma = x_157_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1930_cast_fp16)[name = string("x_157_cast_fp16")];
            int32 var_1949 = const()[name = string("op_1949"), val = int32(1)];
            tensor<int32, [1]> var_1976_axes_0 = const()[name = string("op_1976_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_7_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_7_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_1976_cast_fp16 = layer_norm(axes = var_1976_axes_0, epsilon = layers_7_norm_feed_forward1_eps_scaled_to_fp16, x = x_157_cast_fp16)[name = string("op_1976_cast_fp16")];
            tensor<fp16, [1024]> input_201_gamma_0_to_fp16 = const()[name = string("input_201_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136552320)))];
            tensor<fp16, [1024]> input_201_beta_0_to_fp16 = const()[name = string("input_201_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136554432)))];
            fp16 input_201_epsilon_0_to_fp16 = const()[name = string("input_201_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_201_cast_fp16 = batch_norm(beta = input_201_beta_0_to_fp16, epsilon = input_201_epsilon_0_to_fp16, gamma = input_201_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_1976_cast_fp16)[name = string("input_201_cast_fp16")];
            string input_203_pad_type_0 = const()[name = string("input_203_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_203_strides_0 = const()[name = string("input_203_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_203_pad_0 = const()[name = string("input_203_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_203_dilations_0 = const()[name = string("input_203_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_203_groups_0 = const()[name = string("input_203_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_7_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136556544))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139702336))))[name = string("layers_7_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_203_cast_fp16 = conv(dilations = input_203_dilations_0, groups = input_203_groups_0, pad = input_203_pad_0, pad_type = input_203_pad_type_0, strides = input_203_strides_0, weight = layers_7_feed_forward1_linear1_weight_to_fp16_palettized, x = input_201_cast_fp16)[name = string("input_203_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_205_cast_fp16 = silu(x = input_203_cast_fp16)[name = string("input_205_cast_fp16")];
            string var_1993_pad_type_0 = const()[name = string("op_1993_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_1993_strides_0 = const()[name = string("op_1993_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_1993_pad_0 = const()[name = string("op_1993_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_1993_dilations_0 = const()[name = string("op_1993_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_1993_groups_0 = const()[name = string("op_1993_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_1994_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139735168))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142880960))))[name = string("op_1994_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_1994_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_1993_dilations_0, groups = var_1993_groups_0, pad = var_1993_pad_0, pad_type = var_1993_pad_type_0, strides = var_1993_strides_0, weight = op_1994_weight_0_to_fp16_palettized, x = input_205_cast_fp16)[name = string("op_1994_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_159_cast_fp16 = add(x = x_157_cast_fp16, y = var_1994_cast_fp16)[name = string("x_159_cast_fp16")];
            tensor<int32, [1]> var_2010_axes_0 = const()[name = string("op_2010_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_7_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_7_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2010_cast_fp16 = layer_norm(axes = var_2010_axes_0, epsilon = layers_7_norm_self_att_eps_scaled_to_fp16, x = x_159_cast_fp16)[name = string("op_2010_cast_fp16")];
            tensor<fp16, [1024]> x_161_gamma_0_to_fp16 = const()[name = string("x_161_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142889216)))];
            tensor<fp16, [1024]> x_161_beta_0_to_fp16 = const()[name = string("x_161_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142891328)))];
            fp16 x_161_epsilon_0_to_fp16 = const()[name = string("x_161_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_161_cast_fp16 = batch_norm(beta = x_161_beta_0_to_fp16, epsilon = x_161_epsilon_0_to_fp16, gamma = x_161_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2010_cast_fp16)[name = string("x_161_cast_fp16")];
            string q_15_pad_type_0 = const()[name = string("q_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_15_strides_0 = const()[name = string("q_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_15_pad_0 = const()[name = string("q_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_15_dilations_0 = const()[name = string("q_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_15_groups_0 = const()[name = string("q_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142893440))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143679936))))[name = string("layers_7_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_15_cast_fp16 = conv(dilations = q_15_dilations_0, groups = q_15_groups_0, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = q_15_strides_0, weight = layers_7_self_attn_linear_q_weight_to_fp16_palettized, x = x_161_cast_fp16)[name = string("q_15_cast_fp16")];
            string k_15_pad_type_0 = const()[name = string("k_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_15_strides_0 = const()[name = string("k_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_15_pad_0 = const()[name = string("k_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_15_dilations_0 = const()[name = string("k_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_15_groups_0 = const()[name = string("k_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143688192))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144474688))))[name = string("layers_7_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_15_cast_fp16 = conv(dilations = k_15_dilations_0, groups = k_15_groups_0, pad = k_15_pad_0, pad_type = k_15_pad_type_0, strides = k_15_strides_0, weight = layers_7_self_attn_linear_k_weight_to_fp16_palettized, x = x_161_cast_fp16)[name = string("k_15_cast_fp16")];
            string v_15_pad_type_0 = const()[name = string("v_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_15_strides_0 = const()[name = string("v_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_15_pad_0 = const()[name = string("v_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_15_dilations_0 = const()[name = string("v_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_15_groups_0 = const()[name = string("v_15_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144482944))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145269440))))[name = string("layers_7_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_15_cast_fp16 = conv(dilations = v_15_dilations_0, groups = v_15_groups_0, pad = v_15_pad_0, pad_type = v_15_pad_type_0, strides = v_15_strides_0, weight = layers_7_self_attn_linear_v_weight_to_fp16_palettized, x = x_161_cast_fp16)[name = string("v_15_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_15_to_fp16 = const()[name = string("bv_all_15_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145277696)))];
            tensor<fp16, [1, 1024, 1, 188]> var_2042_cast_fp16 = add(x = q_15_cast_fp16, y = bv_all_15_to_fp16)[name = string("op_2042_cast_fp16")];
            tensor<int32, [3]> var_2043 = const()[name = string("op_2043"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_15_cast_fp16 = reshape(shape = var_2043, x = var_2042_cast_fp16)[name = string("qb_15_cast_fp16")];
            bool bd_all_29_transpose_x_0 = const()[name = string("bd_all_29_transpose_x_0"), val = bool(false)];
            bool bd_all_29_transpose_y_0 = const()[name = string("bd_all_29_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_7_self_attn_pos_proj_to_fp16 = const()[name = string("layers_7_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145279808)))];
            tensor<fp16, [8, 375, 188]> bd_all_29_cast_fp16 = matmul(transpose_x = bd_all_29_transpose_x_0, transpose_y = bd_all_29_transpose_y_0, x = layers_7_self_attn_pos_proj_to_fp16, y = qb_15_cast_fp16)[name = string("bd_all_29_cast_fp16")];
            tensor<int32, [3]> x_163_perm_0 = const()[name = string("x_163_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_165_pad_0 = const()[name = string("x_165_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_165_mode_0 = const()[name = string("x_165_mode_0"), val = string("constant")];
            fp16 const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_163_cast_fp16 = transpose(perm = x_163_perm_0, x = bd_all_29_cast_fp16)[name = string("transpose_101")];
            tensor<fp16, [8, 188, 376]> x_165_cast_fp16 = pad(constant_val = const_38_to_fp16, mode = x_165_mode_0, pad = x_165_pad_0, x = x_163_cast_fp16)[name = string("x_165_cast_fp16")];
            tensor<int32, [3]> var_2050 = const()[name = string("op_2050"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_167_cast_fp16 = reshape(shape = var_2050, x = x_165_cast_fp16)[name = string("x_167_cast_fp16")];
            tensor<int32, [3]> var_2053_begin_0 = const()[name = string("op_2053_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_2053_end_0 = const()[name = string("op_2053_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_2053_end_mask_0 = const()[name = string("op_2053_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_2053_cast_fp16 = slice_by_index(begin = var_2053_begin_0, end = var_2053_end_0, end_mask = var_2053_end_mask_0, x = x_167_cast_fp16)[name = string("op_2053_cast_fp16")];
            tensor<int32, [3]> var_2054 = const()[name = string("op_2054"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_169_cast_fp16 = reshape(shape = var_2054, x = var_2053_cast_fp16)[name = string("x_169_cast_fp16")];
            tensor<int32, [3]> bd_all_31_begin_0 = const()[name = string("bd_all_31_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_31_end_0 = const()[name = string("bd_all_31_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_31_end_mask_0 = const()[name = string("bd_all_31_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_31_cast_fp16 = slice_by_index(begin = bd_all_31_begin_0, end = bd_all_31_end_0, end_mask = bd_all_31_end_mask_0, x = x_169_cast_fp16)[name = string("bd_all_31_cast_fp16")];
            tensor<int32, [4]> var_2059 = const()[name = string("op_2059"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_2060_cast_fp16 = reshape(shape = var_2059, x = q_15_cast_fp16)[name = string("op_2060_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_2062_to_fp16 = const()[name = string("op_2062_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146047872)))];
            tensor<fp16, [8, 128, 1, 188]> var_2063_cast_fp16 = add(x = var_2060_cast_fp16, y = var_2062_to_fp16)[name = string("op_2063_cast_fp16")];
            tensor<int32, [4]> var_2064 = const()[name = string("op_2064"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_15_cast_fp16 = reshape(shape = var_2064, x = k_15_cast_fp16)[name = string("kh_15_cast_fp16")];
            tensor<int32, [4]> var_2066 = const()[name = string("op_2066"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_15_cast_fp16 = reshape(shape = var_2066, x = v_15_cast_fp16)[name = string("vh_15_cast_fp16")];
            tensor<int32, [4]> var_2068 = const()[name = string("op_2068"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_15_equation_0 = const()[name = string("ac_15_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_2069_cast_fp16 = transpose(perm = var_2068, x = kh_15_cast_fp16)[name = string("transpose_100")];
            tensor<fp16, [8, 188, 1, 188]> ac_15_cast_fp16 = einsum(equation = ac_15_equation_0, values = (var_2069_cast_fp16, var_2063_cast_fp16))[name = string("ac_15_cast_fp16")];
            tensor<int32, [3]> var_2072_perm_0 = const()[name = string("op_2072_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2073_axes_0 = const()[name = string("op_2073_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_2072_cast_fp16 = transpose(perm = var_2072_perm_0, x = bd_all_31_cast_fp16)[name = string("transpose_99")];
            tensor<fp16, [8, 188, 1, 188]> var_2073_cast_fp16 = expand_dims(axes = var_2073_axes_0, x = var_2072_cast_fp16)[name = string("op_2073_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_2074_cast_fp16 = add(x = ac_15_cast_fp16, y = var_2073_cast_fp16)[name = string("op_2074_cast_fp16")];
            fp16 var_2075_to_fp16 = const()[name = string("op_2075_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_29_cast_fp16 = mul(x = var_2074_cast_fp16, y = var_2075_to_fp16)[name = string("scores_29_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_31_cast_fp16 = add(x = scores_29_cast_fp16, y = key_bias)[name = string("scores_31_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_2078_cast_fp16 = softmax(axis = var_1949, x = scores_31_cast_fp16)[name = string("op_2078_cast_fp16")];
            tensor<int32, [4]> transpose_55_perm_0 = const()[name = string("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_14_perm_0 = const()[name = string("transpose_14_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_74 = const()[name = string("concat_74"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_14_cast_fp16 = transpose(perm = transpose_14_perm_0, x = var_2078_cast_fp16)[name = string("transpose_98")];
            tensor<fp16, [8, 188, 188]> reshape_21_cast_fp16 = reshape(shape = concat_74, x = transpose_14_cast_fp16)[name = string("reshape_21_cast_fp16")];
            tensor<int32, [3]> concat_75 = const()[name = string("concat_75"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_55_cast_fp16 = transpose(perm = transpose_55_perm_0, x = vh_15_cast_fp16)[name = string("transpose_97")];
            tensor<fp16, [8, 188, 128]> reshape_22_cast_fp16 = reshape(shape = concat_75, x = transpose_55_cast_fp16)[name = string("reshape_22_cast_fp16")];
            bool matmul_7_transpose_x_0 = const()[name = string("matmul_7_transpose_x_0"), val = bool(false)];
            bool matmul_7_transpose_y_0 = const()[name = string("matmul_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = reshape_21_cast_fp16, y = reshape_22_cast_fp16)[name = string("matmul_7_cast_fp16")];
            tensor<int32, [4]> concat_79 = const()[name = string("concat_79"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_23_cast_fp16 = reshape(shape = concat_79, x = matmul_7_cast_fp16)[name = string("reshape_23_cast_fp16")];
            tensor<int32, [4]> ctx_15_perm_0 = const()[name = string("ctx_15_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_2083 = const()[name = string("op_2083"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_15_cast_fp16 = transpose(perm = ctx_15_perm_0, x = reshape_23_cast_fp16)[name = string("transpose_96")];
            tensor<fp16, [1, 1024, 1, 188]> input_207_cast_fp16 = reshape(shape = var_2083, x = ctx_15_cast_fp16)[name = string("input_207_cast_fp16")];
            string var_2090_pad_type_0 = const()[name = string("op_2090_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2090_strides_0 = const()[name = string("op_2090_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2090_pad_0 = const()[name = string("op_2090_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2090_dilations_0 = const()[name = string("op_2090_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2090_groups_0 = const()[name = string("op_2090_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146049984))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146836480))))[name = string("layers_7_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2090_cast_fp16 = conv(dilations = var_2090_dilations_0, groups = var_2090_groups_0, pad = var_2090_pad_0, pad_type = var_2090_pad_type_0, strides = var_2090_strides_0, weight = layers_7_self_attn_linear_out_weight_to_fp16_palettized, x = input_207_cast_fp16)[name = string("op_2090_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_171_cast_fp16 = add(x = x_159_cast_fp16, y = var_2090_cast_fp16)[name = string("x_171_cast_fp16")];
            tensor<int32, [1]> var_2106_axes_0 = const()[name = string("op_2106_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_7_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_7_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2106_cast_fp16 = layer_norm(axes = var_2106_axes_0, epsilon = layers_7_norm_conv_eps_scaled_to_fp16, x = x_171_cast_fp16)[name = string("op_2106_cast_fp16")];
            tensor<fp16, [1024]> input_209_gamma_0_to_fp16 = const()[name = string("input_209_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146844736)))];
            tensor<fp16, [1024]> input_209_beta_0_to_fp16 = const()[name = string("input_209_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146846848)))];
            fp16 input_209_epsilon_0_to_fp16 = const()[name = string("input_209_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_209_cast_fp16 = batch_norm(beta = input_209_beta_0_to_fp16, epsilon = input_209_epsilon_0_to_fp16, gamma = input_209_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2106_cast_fp16)[name = string("input_209_cast_fp16")];
            string input_211_pad_type_0 = const()[name = string("input_211_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_211_strides_0 = const()[name = string("input_211_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_211_pad_0 = const()[name = string("input_211_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_211_dilations_0 = const()[name = string("input_211_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_211_groups_0 = const()[name = string("input_211_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_7_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146848960))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148421888))))[name = string("layers_7_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_211_cast_fp16 = conv(dilations = input_211_dilations_0, groups = input_211_groups_0, pad = input_211_pad_0, pad_type = input_211_pad_type_0, strides = input_211_strides_0, weight = layers_7_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_209_cast_fp16)[name = string("input_211_cast_fp16")];
            int32 x_173_split_num_splits_0 = const()[name = string("x_173_split_num_splits_0"), val = int32(2)];
            int32 x_173_split_axis_0 = const()[name = string("x_173_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_173_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_211_cast_fp16)[name = string("x_173_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_173_split_1_sigmoid_cast_fp16 = sigmoid(x = x_173_split_cast_fp16_1)[name = string("x_173_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_173_cast_fp16 = mul(x = x_173_split_cast_fp16_0, y = x_173_split_1_sigmoid_cast_fp16)[name = string("x_173_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_213_cast_fp16 = mul(x = x_173_cast_fp16, y = pad_mask)[name = string("input_213_cast_fp16")];
            string input_215_pad_type_0 = const()[name = string("input_215_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_215_pad_0 = const()[name = string("input_215_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_215_groups_0 = const()[name = string("input_215_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_215_strides_0 = const()[name = string("input_215_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_215_dilations_0 = const()[name = string("input_215_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_117_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148438336))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148445312))))[name = string("const_117_to_fp16_palettized")];
            tensor<fp16, [1024]> const_118_to_fp16 = const()[name = string("const_118_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148453568)))];
            tensor<fp16, [1, 1024, 1, 188]> input_217_cast_fp16 = conv(bias = const_118_to_fp16, dilations = input_215_dilations_0, groups = input_215_groups_0, pad = input_215_pad_0, pad_type = input_215_pad_type_0, strides = input_215_strides_0, weight = const_117_to_fp16_palettized, x = input_213_cast_fp16)[name = string("input_217_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_219_cast_fp16 = silu(x = input_217_cast_fp16)[name = string("input_219_cast_fp16")];
            string var_2138_pad_type_0 = const()[name = string("op_2138_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2138_strides_0 = const()[name = string("op_2138_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2138_pad_0 = const()[name = string("op_2138_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2138_dilations_0 = const()[name = string("op_2138_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2138_groups_0 = const()[name = string("op_2138_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_7_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148455680))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149242176))))[name = string("layers_7_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2138_cast_fp16 = conv(dilations = var_2138_dilations_0, groups = var_2138_groups_0, pad = var_2138_pad_0, pad_type = var_2138_pad_type_0, strides = var_2138_strides_0, weight = layers_7_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = string("op_2138_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_175_cast_fp16 = add(x = x_171_cast_fp16, y = var_2138_cast_fp16)[name = string("x_175_cast_fp16")];
            tensor<int32, [1]> var_2154_axes_0 = const()[name = string("op_2154_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_7_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_7_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2154_cast_fp16 = layer_norm(axes = var_2154_axes_0, epsilon = layers_7_norm_feed_forward2_eps_scaled_to_fp16, x = x_175_cast_fp16)[name = string("op_2154_cast_fp16")];
            tensor<fp16, [1024]> input_221_gamma_0_to_fp16 = const()[name = string("input_221_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149250432)))];
            tensor<fp16, [1024]> input_221_beta_0_to_fp16 = const()[name = string("input_221_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149252544)))];
            fp16 input_221_epsilon_0_to_fp16 = const()[name = string("input_221_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_221_cast_fp16 = batch_norm(beta = input_221_beta_0_to_fp16, epsilon = input_221_epsilon_0_to_fp16, gamma = input_221_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2154_cast_fp16)[name = string("input_221_cast_fp16")];
            string input_223_pad_type_0 = const()[name = string("input_223_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_223_strides_0 = const()[name = string("input_223_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_223_pad_0 = const()[name = string("input_223_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_223_dilations_0 = const()[name = string("input_223_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_223_groups_0 = const()[name = string("input_223_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_7_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149254656))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152400448))))[name = string("layers_7_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_223_cast_fp16 = conv(dilations = input_223_dilations_0, groups = input_223_groups_0, pad = input_223_pad_0, pad_type = input_223_pad_type_0, strides = input_223_strides_0, weight = layers_7_feed_forward2_linear1_weight_to_fp16_palettized, x = input_221_cast_fp16)[name = string("input_223_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_225_cast_fp16 = silu(x = input_223_cast_fp16)[name = string("input_225_cast_fp16")];
            string var_2171_pad_type_0 = const()[name = string("op_2171_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2171_strides_0 = const()[name = string("op_2171_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2171_pad_0 = const()[name = string("op_2171_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2171_dilations_0 = const()[name = string("op_2171_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2171_groups_0 = const()[name = string("op_2171_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_2172_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152433280))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155579072))))[name = string("op_2172_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2172_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_2171_dilations_0, groups = var_2171_groups_0, pad = var_2171_pad_0, pad_type = var_2171_pad_type_0, strides = var_2171_strides_0, weight = op_2172_weight_0_to_fp16_palettized, x = input_225_cast_fp16)[name = string("op_2172_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_177_cast_fp16 = add(x = x_175_cast_fp16, y = var_2172_cast_fp16)[name = string("x_177_cast_fp16")];
            tensor<int32, [1]> var_2188_axes_0 = const()[name = string("op_2188_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_7_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_7_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2188_cast_fp16 = layer_norm(axes = var_2188_axes_0, epsilon = layers_7_norm_out_eps_scaled_to_fp16, x = x_177_cast_fp16)[name = string("op_2188_cast_fp16")];
            tensor<fp16, [1024]> x_179_gamma_0_to_fp16 = const()[name = string("x_179_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155587328)))];
            tensor<fp16, [1024]> x_179_beta_0_to_fp16 = const()[name = string("x_179_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155589440)))];
            fp16 x_179_epsilon_0_to_fp16 = const()[name = string("x_179_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_179_cast_fp16 = batch_norm(beta = x_179_beta_0_to_fp16, epsilon = x_179_epsilon_0_to_fp16, gamma = x_179_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2188_cast_fp16)[name = string("x_179_cast_fp16")];
            int32 var_2207 = const()[name = string("op_2207"), val = int32(1)];
            tensor<int32, [1]> var_2234_axes_0 = const()[name = string("op_2234_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_8_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_8_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2234_cast_fp16 = layer_norm(axes = var_2234_axes_0, epsilon = layers_8_norm_feed_forward1_eps_scaled_to_fp16, x = x_179_cast_fp16)[name = string("op_2234_cast_fp16")];
            tensor<fp16, [1024]> input_227_gamma_0_to_fp16 = const()[name = string("input_227_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155591552)))];
            tensor<fp16, [1024]> input_227_beta_0_to_fp16 = const()[name = string("input_227_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155593664)))];
            fp16 input_227_epsilon_0_to_fp16 = const()[name = string("input_227_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_227_cast_fp16 = batch_norm(beta = input_227_beta_0_to_fp16, epsilon = input_227_epsilon_0_to_fp16, gamma = input_227_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2234_cast_fp16)[name = string("input_227_cast_fp16")];
            string input_229_pad_type_0 = const()[name = string("input_229_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_229_strides_0 = const()[name = string("input_229_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_229_pad_0 = const()[name = string("input_229_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_229_dilations_0 = const()[name = string("input_229_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_229_groups_0 = const()[name = string("input_229_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_8_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(155595776))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158741568))))[name = string("layers_8_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_229_cast_fp16 = conv(dilations = input_229_dilations_0, groups = input_229_groups_0, pad = input_229_pad_0, pad_type = input_229_pad_type_0, strides = input_229_strides_0, weight = layers_8_feed_forward1_linear1_weight_to_fp16_palettized, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_231_cast_fp16 = silu(x = input_229_cast_fp16)[name = string("input_231_cast_fp16")];
            string var_2251_pad_type_0 = const()[name = string("op_2251_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2251_strides_0 = const()[name = string("op_2251_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2251_pad_0 = const()[name = string("op_2251_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2251_dilations_0 = const()[name = string("op_2251_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2251_groups_0 = const()[name = string("op_2251_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_2252_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158774400))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161920192))))[name = string("op_2252_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2252_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_2251_dilations_0, groups = var_2251_groups_0, pad = var_2251_pad_0, pad_type = var_2251_pad_type_0, strides = var_2251_strides_0, weight = op_2252_weight_0_to_fp16_palettized, x = input_231_cast_fp16)[name = string("op_2252_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_181_cast_fp16 = add(x = x_179_cast_fp16, y = var_2252_cast_fp16)[name = string("x_181_cast_fp16")];
            tensor<int32, [1]> var_2268_axes_0 = const()[name = string("op_2268_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_8_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_8_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2268_cast_fp16 = layer_norm(axes = var_2268_axes_0, epsilon = layers_8_norm_self_att_eps_scaled_to_fp16, x = x_181_cast_fp16)[name = string("op_2268_cast_fp16")];
            tensor<fp16, [1024]> x_183_gamma_0_to_fp16 = const()[name = string("x_183_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161928448)))];
            tensor<fp16, [1024]> x_183_beta_0_to_fp16 = const()[name = string("x_183_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161930560)))];
            fp16 x_183_epsilon_0_to_fp16 = const()[name = string("x_183_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_183_cast_fp16 = batch_norm(beta = x_183_beta_0_to_fp16, epsilon = x_183_epsilon_0_to_fp16, gamma = x_183_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2268_cast_fp16)[name = string("x_183_cast_fp16")];
            string q_17_pad_type_0 = const()[name = string("q_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_17_strides_0 = const()[name = string("q_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_17_pad_0 = const()[name = string("q_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_17_dilations_0 = const()[name = string("q_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_17_groups_0 = const()[name = string("q_17_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161932672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162719168))))[name = string("layers_8_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_17_cast_fp16 = conv(dilations = q_17_dilations_0, groups = q_17_groups_0, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = q_17_strides_0, weight = layers_8_self_attn_linear_q_weight_to_fp16_palettized, x = x_183_cast_fp16)[name = string("q_17_cast_fp16")];
            string k_17_pad_type_0 = const()[name = string("k_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_17_strides_0 = const()[name = string("k_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_17_pad_0 = const()[name = string("k_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_17_dilations_0 = const()[name = string("k_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_17_groups_0 = const()[name = string("k_17_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162727424))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163513920))))[name = string("layers_8_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_17_cast_fp16 = conv(dilations = k_17_dilations_0, groups = k_17_groups_0, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = k_17_strides_0, weight = layers_8_self_attn_linear_k_weight_to_fp16_palettized, x = x_183_cast_fp16)[name = string("k_17_cast_fp16")];
            string v_17_pad_type_0 = const()[name = string("v_17_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_17_strides_0 = const()[name = string("v_17_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_17_pad_0 = const()[name = string("v_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_17_dilations_0 = const()[name = string("v_17_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_17_groups_0 = const()[name = string("v_17_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163522176))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164308672))))[name = string("layers_8_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_17_cast_fp16 = conv(dilations = v_17_dilations_0, groups = v_17_groups_0, pad = v_17_pad_0, pad_type = v_17_pad_type_0, strides = v_17_strides_0, weight = layers_8_self_attn_linear_v_weight_to_fp16_palettized, x = x_183_cast_fp16)[name = string("v_17_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_17_to_fp16 = const()[name = string("bv_all_17_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164316928)))];
            tensor<fp16, [1, 1024, 1, 188]> var_2300_cast_fp16 = add(x = q_17_cast_fp16, y = bv_all_17_to_fp16)[name = string("op_2300_cast_fp16")];
            tensor<int32, [3]> var_2301 = const()[name = string("op_2301"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_17_cast_fp16 = reshape(shape = var_2301, x = var_2300_cast_fp16)[name = string("qb_17_cast_fp16")];
            bool bd_all_33_transpose_x_0 = const()[name = string("bd_all_33_transpose_x_0"), val = bool(false)];
            bool bd_all_33_transpose_y_0 = const()[name = string("bd_all_33_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_8_self_attn_pos_proj_to_fp16 = const()[name = string("layers_8_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164319040)))];
            tensor<fp16, [8, 375, 188]> bd_all_33_cast_fp16 = matmul(transpose_x = bd_all_33_transpose_x_0, transpose_y = bd_all_33_transpose_y_0, x = layers_8_self_attn_pos_proj_to_fp16, y = qb_17_cast_fp16)[name = string("bd_all_33_cast_fp16")];
            tensor<int32, [3]> x_185_perm_0 = const()[name = string("x_185_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_187_pad_0 = const()[name = string("x_187_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_187_mode_0 = const()[name = string("x_187_mode_0"), val = string("constant")];
            fp16 const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_185_cast_fp16 = transpose(perm = x_185_perm_0, x = bd_all_33_cast_fp16)[name = string("transpose_95")];
            tensor<fp16, [8, 188, 376]> x_187_cast_fp16 = pad(constant_val = const_42_to_fp16, mode = x_187_mode_0, pad = x_187_pad_0, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
            tensor<int32, [3]> var_2308 = const()[name = string("op_2308"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_189_cast_fp16 = reshape(shape = var_2308, x = x_187_cast_fp16)[name = string("x_189_cast_fp16")];
            tensor<int32, [3]> var_2311_begin_0 = const()[name = string("op_2311_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_2311_end_0 = const()[name = string("op_2311_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_2311_end_mask_0 = const()[name = string("op_2311_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_2311_cast_fp16 = slice_by_index(begin = var_2311_begin_0, end = var_2311_end_0, end_mask = var_2311_end_mask_0, x = x_189_cast_fp16)[name = string("op_2311_cast_fp16")];
            tensor<int32, [3]> var_2312 = const()[name = string("op_2312"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_191_cast_fp16 = reshape(shape = var_2312, x = var_2311_cast_fp16)[name = string("x_191_cast_fp16")];
            tensor<int32, [3]> bd_all_35_begin_0 = const()[name = string("bd_all_35_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_35_end_0 = const()[name = string("bd_all_35_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_35_end_mask_0 = const()[name = string("bd_all_35_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_35_cast_fp16 = slice_by_index(begin = bd_all_35_begin_0, end = bd_all_35_end_0, end_mask = bd_all_35_end_mask_0, x = x_191_cast_fp16)[name = string("bd_all_35_cast_fp16")];
            tensor<int32, [4]> var_2317 = const()[name = string("op_2317"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_2318_cast_fp16 = reshape(shape = var_2317, x = q_17_cast_fp16)[name = string("op_2318_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_2320_to_fp16 = const()[name = string("op_2320_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165087104)))];
            tensor<fp16, [8, 128, 1, 188]> var_2321_cast_fp16 = add(x = var_2318_cast_fp16, y = var_2320_to_fp16)[name = string("op_2321_cast_fp16")];
            tensor<int32, [4]> var_2322 = const()[name = string("op_2322"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_17_cast_fp16 = reshape(shape = var_2322, x = k_17_cast_fp16)[name = string("kh_17_cast_fp16")];
            tensor<int32, [4]> var_2324 = const()[name = string("op_2324"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_17_cast_fp16 = reshape(shape = var_2324, x = v_17_cast_fp16)[name = string("vh_17_cast_fp16")];
            tensor<int32, [4]> var_2326 = const()[name = string("op_2326"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_17_equation_0 = const()[name = string("ac_17_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_2327_cast_fp16 = transpose(perm = var_2326, x = kh_17_cast_fp16)[name = string("transpose_94")];
            tensor<fp16, [8, 188, 1, 188]> ac_17_cast_fp16 = einsum(equation = ac_17_equation_0, values = (var_2327_cast_fp16, var_2321_cast_fp16))[name = string("ac_17_cast_fp16")];
            tensor<int32, [3]> var_2330_perm_0 = const()[name = string("op_2330_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2331_axes_0 = const()[name = string("op_2331_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_2330_cast_fp16 = transpose(perm = var_2330_perm_0, x = bd_all_35_cast_fp16)[name = string("transpose_93")];
            tensor<fp16, [8, 188, 1, 188]> var_2331_cast_fp16 = expand_dims(axes = var_2331_axes_0, x = var_2330_cast_fp16)[name = string("op_2331_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_2332_cast_fp16 = add(x = ac_17_cast_fp16, y = var_2331_cast_fp16)[name = string("op_2332_cast_fp16")];
            fp16 var_2333_to_fp16 = const()[name = string("op_2333_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_33_cast_fp16 = mul(x = var_2332_cast_fp16, y = var_2333_to_fp16)[name = string("scores_33_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_35_cast_fp16 = add(x = scores_33_cast_fp16, y = key_bias)[name = string("scores_35_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_2336_cast_fp16 = softmax(axis = var_2207, x = scores_35_cast_fp16)[name = string("op_2336_cast_fp16")];
            tensor<int32, [4]> transpose_56_perm_0 = const()[name = string("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_16_perm_0 = const()[name = string("transpose_16_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_84 = const()[name = string("concat_84"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_16_cast_fp16 = transpose(perm = transpose_16_perm_0, x = var_2336_cast_fp16)[name = string("transpose_92")];
            tensor<fp16, [8, 188, 188]> reshape_24_cast_fp16 = reshape(shape = concat_84, x = transpose_16_cast_fp16)[name = string("reshape_24_cast_fp16")];
            tensor<int32, [3]> concat_85 = const()[name = string("concat_85"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_56_cast_fp16 = transpose(perm = transpose_56_perm_0, x = vh_17_cast_fp16)[name = string("transpose_91")];
            tensor<fp16, [8, 188, 128]> reshape_25_cast_fp16 = reshape(shape = concat_85, x = transpose_56_cast_fp16)[name = string("reshape_25_cast_fp16")];
            bool matmul_8_transpose_x_0 = const()[name = string("matmul_8_transpose_x_0"), val = bool(false)];
            bool matmul_8_transpose_y_0 = const()[name = string("matmul_8_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = reshape_24_cast_fp16, y = reshape_25_cast_fp16)[name = string("matmul_8_cast_fp16")];
            tensor<int32, [4]> concat_89 = const()[name = string("concat_89"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_26_cast_fp16 = reshape(shape = concat_89, x = matmul_8_cast_fp16)[name = string("reshape_26_cast_fp16")];
            tensor<int32, [4]> ctx_17_perm_0 = const()[name = string("ctx_17_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_2341 = const()[name = string("op_2341"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_17_cast_fp16 = transpose(perm = ctx_17_perm_0, x = reshape_26_cast_fp16)[name = string("transpose_90")];
            tensor<fp16, [1, 1024, 1, 188]> input_233_cast_fp16 = reshape(shape = var_2341, x = ctx_17_cast_fp16)[name = string("input_233_cast_fp16")];
            string var_2348_pad_type_0 = const()[name = string("op_2348_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2348_strides_0 = const()[name = string("op_2348_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2348_pad_0 = const()[name = string("op_2348_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2348_dilations_0 = const()[name = string("op_2348_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2348_groups_0 = const()[name = string("op_2348_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165089216))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165875712))))[name = string("layers_8_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2348_cast_fp16 = conv(dilations = var_2348_dilations_0, groups = var_2348_groups_0, pad = var_2348_pad_0, pad_type = var_2348_pad_type_0, strides = var_2348_strides_0, weight = layers_8_self_attn_linear_out_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = string("op_2348_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_193_cast_fp16 = add(x = x_181_cast_fp16, y = var_2348_cast_fp16)[name = string("x_193_cast_fp16")];
            tensor<int32, [1]> var_2364_axes_0 = const()[name = string("op_2364_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_8_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_8_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2364_cast_fp16 = layer_norm(axes = var_2364_axes_0, epsilon = layers_8_norm_conv_eps_scaled_to_fp16, x = x_193_cast_fp16)[name = string("op_2364_cast_fp16")];
            tensor<fp16, [1024]> input_235_gamma_0_to_fp16 = const()[name = string("input_235_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165883968)))];
            tensor<fp16, [1024]> input_235_beta_0_to_fp16 = const()[name = string("input_235_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165886080)))];
            fp16 input_235_epsilon_0_to_fp16 = const()[name = string("input_235_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_235_cast_fp16 = batch_norm(beta = input_235_beta_0_to_fp16, epsilon = input_235_epsilon_0_to_fp16, gamma = input_235_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2364_cast_fp16)[name = string("input_235_cast_fp16")];
            string input_237_pad_type_0 = const()[name = string("input_237_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_237_strides_0 = const()[name = string("input_237_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_237_pad_0 = const()[name = string("input_237_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_237_dilations_0 = const()[name = string("input_237_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_237_groups_0 = const()[name = string("input_237_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_8_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165888192))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167461120))))[name = string("layers_8_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_237_cast_fp16 = conv(dilations = input_237_dilations_0, groups = input_237_groups_0, pad = input_237_pad_0, pad_type = input_237_pad_type_0, strides = input_237_strides_0, weight = layers_8_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_235_cast_fp16)[name = string("input_237_cast_fp16")];
            int32 x_195_split_num_splits_0 = const()[name = string("x_195_split_num_splits_0"), val = int32(2)];
            int32 x_195_split_axis_0 = const()[name = string("x_195_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_195_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_237_cast_fp16)[name = string("x_195_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_195_split_1_sigmoid_cast_fp16 = sigmoid(x = x_195_split_cast_fp16_1)[name = string("x_195_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_195_cast_fp16 = mul(x = x_195_split_cast_fp16_0, y = x_195_split_1_sigmoid_cast_fp16)[name = string("x_195_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_239_cast_fp16 = mul(x = x_195_cast_fp16, y = pad_mask)[name = string("input_239_cast_fp16")];
            string input_241_pad_type_0 = const()[name = string("input_241_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_241_pad_0 = const()[name = string("input_241_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_241_groups_0 = const()[name = string("input_241_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_241_strides_0 = const()[name = string("input_241_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_241_dilations_0 = const()[name = string("input_241_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_119_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167477568))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167484544))))[name = string("const_119_to_fp16_palettized")];
            tensor<fp16, [1024]> const_120_to_fp16 = const()[name = string("const_120_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167492800)))];
            tensor<fp16, [1, 1024, 1, 188]> input_243_cast_fp16 = conv(bias = const_120_to_fp16, dilations = input_241_dilations_0, groups = input_241_groups_0, pad = input_241_pad_0, pad_type = input_241_pad_type_0, strides = input_241_strides_0, weight = const_119_to_fp16_palettized, x = input_239_cast_fp16)[name = string("input_243_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_245_cast_fp16 = silu(x = input_243_cast_fp16)[name = string("input_245_cast_fp16")];
            string var_2396_pad_type_0 = const()[name = string("op_2396_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2396_strides_0 = const()[name = string("op_2396_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2396_pad_0 = const()[name = string("op_2396_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2396_dilations_0 = const()[name = string("op_2396_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2396_groups_0 = const()[name = string("op_2396_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_8_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167494912))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168281408))))[name = string("layers_8_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2396_cast_fp16 = conv(dilations = var_2396_dilations_0, groups = var_2396_groups_0, pad = var_2396_pad_0, pad_type = var_2396_pad_type_0, strides = var_2396_strides_0, weight = layers_8_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_245_cast_fp16)[name = string("op_2396_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_197_cast_fp16 = add(x = x_193_cast_fp16, y = var_2396_cast_fp16)[name = string("x_197_cast_fp16")];
            tensor<int32, [1]> var_2412_axes_0 = const()[name = string("op_2412_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_8_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_8_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2412_cast_fp16 = layer_norm(axes = var_2412_axes_0, epsilon = layers_8_norm_feed_forward2_eps_scaled_to_fp16, x = x_197_cast_fp16)[name = string("op_2412_cast_fp16")];
            tensor<fp16, [1024]> input_247_gamma_0_to_fp16 = const()[name = string("input_247_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168289664)))];
            tensor<fp16, [1024]> input_247_beta_0_to_fp16 = const()[name = string("input_247_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168291776)))];
            fp16 input_247_epsilon_0_to_fp16 = const()[name = string("input_247_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_247_cast_fp16 = batch_norm(beta = input_247_beta_0_to_fp16, epsilon = input_247_epsilon_0_to_fp16, gamma = input_247_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2412_cast_fp16)[name = string("input_247_cast_fp16")];
            string input_249_pad_type_0 = const()[name = string("input_249_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_249_strides_0 = const()[name = string("input_249_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_249_pad_0 = const()[name = string("input_249_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_249_dilations_0 = const()[name = string("input_249_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_249_groups_0 = const()[name = string("input_249_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_8_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168293888))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171439680))))[name = string("layers_8_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_249_cast_fp16 = conv(dilations = input_249_dilations_0, groups = input_249_groups_0, pad = input_249_pad_0, pad_type = input_249_pad_type_0, strides = input_249_strides_0, weight = layers_8_feed_forward2_linear1_weight_to_fp16_palettized, x = input_247_cast_fp16)[name = string("input_249_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_251_cast_fp16 = silu(x = input_249_cast_fp16)[name = string("input_251_cast_fp16")];
            string var_2429_pad_type_0 = const()[name = string("op_2429_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2429_strides_0 = const()[name = string("op_2429_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2429_pad_0 = const()[name = string("op_2429_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2429_dilations_0 = const()[name = string("op_2429_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2429_groups_0 = const()[name = string("op_2429_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_2430_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171472512))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174618304))))[name = string("op_2430_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2430_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_2429_dilations_0, groups = var_2429_groups_0, pad = var_2429_pad_0, pad_type = var_2429_pad_type_0, strides = var_2429_strides_0, weight = op_2430_weight_0_to_fp16_palettized, x = input_251_cast_fp16)[name = string("op_2430_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_199_cast_fp16 = add(x = x_197_cast_fp16, y = var_2430_cast_fp16)[name = string("x_199_cast_fp16")];
            tensor<int32, [1]> var_2446_axes_0 = const()[name = string("op_2446_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_8_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_8_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2446_cast_fp16 = layer_norm(axes = var_2446_axes_0, epsilon = layers_8_norm_out_eps_scaled_to_fp16, x = x_199_cast_fp16)[name = string("op_2446_cast_fp16")];
            tensor<fp16, [1024]> x_201_gamma_0_to_fp16 = const()[name = string("x_201_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174626560)))];
            tensor<fp16, [1024]> x_201_beta_0_to_fp16 = const()[name = string("x_201_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174628672)))];
            fp16 x_201_epsilon_0_to_fp16 = const()[name = string("x_201_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_201_cast_fp16 = batch_norm(beta = x_201_beta_0_to_fp16, epsilon = x_201_epsilon_0_to_fp16, gamma = x_201_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2446_cast_fp16)[name = string("x_201_cast_fp16")];
            int32 var_2465 = const()[name = string("op_2465"), val = int32(1)];
            tensor<int32, [1]> var_2492_axes_0 = const()[name = string("op_2492_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_9_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_9_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2492_cast_fp16 = layer_norm(axes = var_2492_axes_0, epsilon = layers_9_norm_feed_forward1_eps_scaled_to_fp16, x = x_201_cast_fp16)[name = string("op_2492_cast_fp16")];
            tensor<fp16, [1024]> input_253_gamma_0_to_fp16 = const()[name = string("input_253_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174630784)))];
            tensor<fp16, [1024]> input_253_beta_0_to_fp16 = const()[name = string("input_253_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174632896)))];
            fp16 input_253_epsilon_0_to_fp16 = const()[name = string("input_253_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_253_cast_fp16 = batch_norm(beta = input_253_beta_0_to_fp16, epsilon = input_253_epsilon_0_to_fp16, gamma = input_253_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2492_cast_fp16)[name = string("input_253_cast_fp16")];
            string input_255_pad_type_0 = const()[name = string("input_255_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_255_strides_0 = const()[name = string("input_255_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_255_pad_0 = const()[name = string("input_255_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_255_dilations_0 = const()[name = string("input_255_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_255_groups_0 = const()[name = string("input_255_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_9_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174635008))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177780800))))[name = string("layers_9_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_255_cast_fp16 = conv(dilations = input_255_dilations_0, groups = input_255_groups_0, pad = input_255_pad_0, pad_type = input_255_pad_type_0, strides = input_255_strides_0, weight = layers_9_feed_forward1_linear1_weight_to_fp16_palettized, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_257_cast_fp16 = silu(x = input_255_cast_fp16)[name = string("input_257_cast_fp16")];
            string var_2509_pad_type_0 = const()[name = string("op_2509_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2509_strides_0 = const()[name = string("op_2509_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2509_pad_0 = const()[name = string("op_2509_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2509_dilations_0 = const()[name = string("op_2509_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2509_groups_0 = const()[name = string("op_2509_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_2510_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177813632))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180959424))))[name = string("op_2510_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2510_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_2509_dilations_0, groups = var_2509_groups_0, pad = var_2509_pad_0, pad_type = var_2509_pad_type_0, strides = var_2509_strides_0, weight = op_2510_weight_0_to_fp16_palettized, x = input_257_cast_fp16)[name = string("op_2510_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_203_cast_fp16 = add(x = x_201_cast_fp16, y = var_2510_cast_fp16)[name = string("x_203_cast_fp16")];
            tensor<int32, [1]> var_2526_axes_0 = const()[name = string("op_2526_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_9_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_9_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2526_cast_fp16 = layer_norm(axes = var_2526_axes_0, epsilon = layers_9_norm_self_att_eps_scaled_to_fp16, x = x_203_cast_fp16)[name = string("op_2526_cast_fp16")];
            tensor<fp16, [1024]> x_205_gamma_0_to_fp16 = const()[name = string("x_205_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180967680)))];
            tensor<fp16, [1024]> x_205_beta_0_to_fp16 = const()[name = string("x_205_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180969792)))];
            fp16 x_205_epsilon_0_to_fp16 = const()[name = string("x_205_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_205_cast_fp16 = batch_norm(beta = x_205_beta_0_to_fp16, epsilon = x_205_epsilon_0_to_fp16, gamma = x_205_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2526_cast_fp16)[name = string("x_205_cast_fp16")];
            string q_19_pad_type_0 = const()[name = string("q_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_19_strides_0 = const()[name = string("q_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_19_pad_0 = const()[name = string("q_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_19_dilations_0 = const()[name = string("q_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_19_groups_0 = const()[name = string("q_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180971904))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181758400))))[name = string("layers_9_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_19_cast_fp16 = conv(dilations = q_19_dilations_0, groups = q_19_groups_0, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = q_19_strides_0, weight = layers_9_self_attn_linear_q_weight_to_fp16_palettized, x = x_205_cast_fp16)[name = string("q_19_cast_fp16")];
            string k_19_pad_type_0 = const()[name = string("k_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_19_strides_0 = const()[name = string("k_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_19_pad_0 = const()[name = string("k_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_19_dilations_0 = const()[name = string("k_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_19_groups_0 = const()[name = string("k_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181766656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182553152))))[name = string("layers_9_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_19_cast_fp16 = conv(dilations = k_19_dilations_0, groups = k_19_groups_0, pad = k_19_pad_0, pad_type = k_19_pad_type_0, strides = k_19_strides_0, weight = layers_9_self_attn_linear_k_weight_to_fp16_palettized, x = x_205_cast_fp16)[name = string("k_19_cast_fp16")];
            string v_19_pad_type_0 = const()[name = string("v_19_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_19_strides_0 = const()[name = string("v_19_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_19_pad_0 = const()[name = string("v_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_19_dilations_0 = const()[name = string("v_19_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_19_groups_0 = const()[name = string("v_19_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182561408))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183347904))))[name = string("layers_9_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_19_cast_fp16 = conv(dilations = v_19_dilations_0, groups = v_19_groups_0, pad = v_19_pad_0, pad_type = v_19_pad_type_0, strides = v_19_strides_0, weight = layers_9_self_attn_linear_v_weight_to_fp16_palettized, x = x_205_cast_fp16)[name = string("v_19_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_19_to_fp16 = const()[name = string("bv_all_19_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183356160)))];
            tensor<fp16, [1, 1024, 1, 188]> var_2558_cast_fp16 = add(x = q_19_cast_fp16, y = bv_all_19_to_fp16)[name = string("op_2558_cast_fp16")];
            tensor<int32, [3]> var_2559 = const()[name = string("op_2559"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_19_cast_fp16 = reshape(shape = var_2559, x = var_2558_cast_fp16)[name = string("qb_19_cast_fp16")];
            bool bd_all_37_transpose_x_0 = const()[name = string("bd_all_37_transpose_x_0"), val = bool(false)];
            bool bd_all_37_transpose_y_0 = const()[name = string("bd_all_37_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_9_self_attn_pos_proj_to_fp16 = const()[name = string("layers_9_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(183358272)))];
            tensor<fp16, [8, 375, 188]> bd_all_37_cast_fp16 = matmul(transpose_x = bd_all_37_transpose_x_0, transpose_y = bd_all_37_transpose_y_0, x = layers_9_self_attn_pos_proj_to_fp16, y = qb_19_cast_fp16)[name = string("bd_all_37_cast_fp16")];
            tensor<int32, [3]> x_207_perm_0 = const()[name = string("x_207_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_209_pad_0 = const()[name = string("x_209_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_209_mode_0 = const()[name = string("x_209_mode_0"), val = string("constant")];
            fp16 const_46_to_fp16 = const()[name = string("const_46_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_207_cast_fp16 = transpose(perm = x_207_perm_0, x = bd_all_37_cast_fp16)[name = string("transpose_89")];
            tensor<fp16, [8, 188, 376]> x_209_cast_fp16 = pad(constant_val = const_46_to_fp16, mode = x_209_mode_0, pad = x_209_pad_0, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
            tensor<int32, [3]> var_2566 = const()[name = string("op_2566"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_211_cast_fp16 = reshape(shape = var_2566, x = x_209_cast_fp16)[name = string("x_211_cast_fp16")];
            tensor<int32, [3]> var_2569_begin_0 = const()[name = string("op_2569_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_2569_end_0 = const()[name = string("op_2569_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_2569_end_mask_0 = const()[name = string("op_2569_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_2569_cast_fp16 = slice_by_index(begin = var_2569_begin_0, end = var_2569_end_0, end_mask = var_2569_end_mask_0, x = x_211_cast_fp16)[name = string("op_2569_cast_fp16")];
            tensor<int32, [3]> var_2570 = const()[name = string("op_2570"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_213_cast_fp16 = reshape(shape = var_2570, x = var_2569_cast_fp16)[name = string("x_213_cast_fp16")];
            tensor<int32, [3]> bd_all_39_begin_0 = const()[name = string("bd_all_39_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_39_end_0 = const()[name = string("bd_all_39_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_39_end_mask_0 = const()[name = string("bd_all_39_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_39_cast_fp16 = slice_by_index(begin = bd_all_39_begin_0, end = bd_all_39_end_0, end_mask = bd_all_39_end_mask_0, x = x_213_cast_fp16)[name = string("bd_all_39_cast_fp16")];
            tensor<int32, [4]> var_2575 = const()[name = string("op_2575"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_2576_cast_fp16 = reshape(shape = var_2575, x = q_19_cast_fp16)[name = string("op_2576_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_2578_to_fp16 = const()[name = string("op_2578_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184126336)))];
            tensor<fp16, [8, 128, 1, 188]> var_2579_cast_fp16 = add(x = var_2576_cast_fp16, y = var_2578_to_fp16)[name = string("op_2579_cast_fp16")];
            tensor<int32, [4]> var_2580 = const()[name = string("op_2580"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_19_cast_fp16 = reshape(shape = var_2580, x = k_19_cast_fp16)[name = string("kh_19_cast_fp16")];
            tensor<int32, [4]> var_2582 = const()[name = string("op_2582"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_19_cast_fp16 = reshape(shape = var_2582, x = v_19_cast_fp16)[name = string("vh_19_cast_fp16")];
            tensor<int32, [4]> var_2584 = const()[name = string("op_2584"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_19_equation_0 = const()[name = string("ac_19_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_2585_cast_fp16 = transpose(perm = var_2584, x = kh_19_cast_fp16)[name = string("transpose_88")];
            tensor<fp16, [8, 188, 1, 188]> ac_19_cast_fp16 = einsum(equation = ac_19_equation_0, values = (var_2585_cast_fp16, var_2579_cast_fp16))[name = string("ac_19_cast_fp16")];
            tensor<int32, [3]> var_2588_perm_0 = const()[name = string("op_2588_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2589_axes_0 = const()[name = string("op_2589_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_2588_cast_fp16 = transpose(perm = var_2588_perm_0, x = bd_all_39_cast_fp16)[name = string("transpose_87")];
            tensor<fp16, [8, 188, 1, 188]> var_2589_cast_fp16 = expand_dims(axes = var_2589_axes_0, x = var_2588_cast_fp16)[name = string("op_2589_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_2590_cast_fp16 = add(x = ac_19_cast_fp16, y = var_2589_cast_fp16)[name = string("op_2590_cast_fp16")];
            fp16 var_2591_to_fp16 = const()[name = string("op_2591_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_37_cast_fp16 = mul(x = var_2590_cast_fp16, y = var_2591_to_fp16)[name = string("scores_37_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_39_cast_fp16 = add(x = scores_37_cast_fp16, y = key_bias)[name = string("scores_39_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_2594_cast_fp16 = softmax(axis = var_2465, x = scores_39_cast_fp16)[name = string("op_2594_cast_fp16")];
            tensor<int32, [4]> transpose_57_perm_0 = const()[name = string("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_18_perm_0 = const()[name = string("transpose_18_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_94 = const()[name = string("concat_94"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_18_cast_fp16 = transpose(perm = transpose_18_perm_0, x = var_2594_cast_fp16)[name = string("transpose_86")];
            tensor<fp16, [8, 188, 188]> reshape_27_cast_fp16 = reshape(shape = concat_94, x = transpose_18_cast_fp16)[name = string("reshape_27_cast_fp16")];
            tensor<int32, [3]> concat_95 = const()[name = string("concat_95"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_57_cast_fp16 = transpose(perm = transpose_57_perm_0, x = vh_19_cast_fp16)[name = string("transpose_85")];
            tensor<fp16, [8, 188, 128]> reshape_28_cast_fp16 = reshape(shape = concat_95, x = transpose_57_cast_fp16)[name = string("reshape_28_cast_fp16")];
            bool matmul_9_transpose_x_0 = const()[name = string("matmul_9_transpose_x_0"), val = bool(false)];
            bool matmul_9_transpose_y_0 = const()[name = string("matmul_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = reshape_27_cast_fp16, y = reshape_28_cast_fp16)[name = string("matmul_9_cast_fp16")];
            tensor<int32, [4]> concat_99 = const()[name = string("concat_99"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_29_cast_fp16 = reshape(shape = concat_99, x = matmul_9_cast_fp16)[name = string("reshape_29_cast_fp16")];
            tensor<int32, [4]> ctx_19_perm_0 = const()[name = string("ctx_19_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_2599 = const()[name = string("op_2599"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_19_cast_fp16 = transpose(perm = ctx_19_perm_0, x = reshape_29_cast_fp16)[name = string("transpose_84")];
            tensor<fp16, [1, 1024, 1, 188]> input_259_cast_fp16 = reshape(shape = var_2599, x = ctx_19_cast_fp16)[name = string("input_259_cast_fp16")];
            string var_2606_pad_type_0 = const()[name = string("op_2606_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2606_strides_0 = const()[name = string("op_2606_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2606_pad_0 = const()[name = string("op_2606_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2606_dilations_0 = const()[name = string("op_2606_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2606_groups_0 = const()[name = string("op_2606_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184128448))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184914944))))[name = string("layers_9_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2606_cast_fp16 = conv(dilations = var_2606_dilations_0, groups = var_2606_groups_0, pad = var_2606_pad_0, pad_type = var_2606_pad_type_0, strides = var_2606_strides_0, weight = layers_9_self_attn_linear_out_weight_to_fp16_palettized, x = input_259_cast_fp16)[name = string("op_2606_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_215_cast_fp16 = add(x = x_203_cast_fp16, y = var_2606_cast_fp16)[name = string("x_215_cast_fp16")];
            tensor<int32, [1]> var_2622_axes_0 = const()[name = string("op_2622_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_9_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_9_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2622_cast_fp16 = layer_norm(axes = var_2622_axes_0, epsilon = layers_9_norm_conv_eps_scaled_to_fp16, x = x_215_cast_fp16)[name = string("op_2622_cast_fp16")];
            tensor<fp16, [1024]> input_261_gamma_0_to_fp16 = const()[name = string("input_261_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184923200)))];
            tensor<fp16, [1024]> input_261_beta_0_to_fp16 = const()[name = string("input_261_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184925312)))];
            fp16 input_261_epsilon_0_to_fp16 = const()[name = string("input_261_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_261_cast_fp16 = batch_norm(beta = input_261_beta_0_to_fp16, epsilon = input_261_epsilon_0_to_fp16, gamma = input_261_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2622_cast_fp16)[name = string("input_261_cast_fp16")];
            string input_263_pad_type_0 = const()[name = string("input_263_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_263_strides_0 = const()[name = string("input_263_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_263_pad_0 = const()[name = string("input_263_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_263_dilations_0 = const()[name = string("input_263_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_263_groups_0 = const()[name = string("input_263_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_9_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184927424))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186500352))))[name = string("layers_9_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_263_cast_fp16 = conv(dilations = input_263_dilations_0, groups = input_263_groups_0, pad = input_263_pad_0, pad_type = input_263_pad_type_0, strides = input_263_strides_0, weight = layers_9_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_261_cast_fp16)[name = string("input_263_cast_fp16")];
            int32 x_217_split_num_splits_0 = const()[name = string("x_217_split_num_splits_0"), val = int32(2)];
            int32 x_217_split_axis_0 = const()[name = string("x_217_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_217_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_263_cast_fp16)[name = string("x_217_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_217_split_1_sigmoid_cast_fp16 = sigmoid(x = x_217_split_cast_fp16_1)[name = string("x_217_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_217_cast_fp16 = mul(x = x_217_split_cast_fp16_0, y = x_217_split_1_sigmoid_cast_fp16)[name = string("x_217_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_265_cast_fp16 = mul(x = x_217_cast_fp16, y = pad_mask)[name = string("input_265_cast_fp16")];
            string input_267_pad_type_0 = const()[name = string("input_267_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_267_pad_0 = const()[name = string("input_267_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_267_groups_0 = const()[name = string("input_267_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_267_strides_0 = const()[name = string("input_267_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_267_dilations_0 = const()[name = string("input_267_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_121_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186516800))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186523776))))[name = string("const_121_to_fp16_palettized")];
            tensor<fp16, [1024]> const_122_to_fp16 = const()[name = string("const_122_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186532032)))];
            tensor<fp16, [1, 1024, 1, 188]> input_269_cast_fp16 = conv(bias = const_122_to_fp16, dilations = input_267_dilations_0, groups = input_267_groups_0, pad = input_267_pad_0, pad_type = input_267_pad_type_0, strides = input_267_strides_0, weight = const_121_to_fp16_palettized, x = input_265_cast_fp16)[name = string("input_269_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_271_cast_fp16 = silu(x = input_269_cast_fp16)[name = string("input_271_cast_fp16")];
            string var_2654_pad_type_0 = const()[name = string("op_2654_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2654_strides_0 = const()[name = string("op_2654_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2654_pad_0 = const()[name = string("op_2654_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2654_dilations_0 = const()[name = string("op_2654_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2654_groups_0 = const()[name = string("op_2654_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_9_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186534144))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187320640))))[name = string("layers_9_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2654_cast_fp16 = conv(dilations = var_2654_dilations_0, groups = var_2654_groups_0, pad = var_2654_pad_0, pad_type = var_2654_pad_type_0, strides = var_2654_strides_0, weight = layers_9_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_271_cast_fp16)[name = string("op_2654_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_219_cast_fp16 = add(x = x_215_cast_fp16, y = var_2654_cast_fp16)[name = string("x_219_cast_fp16")];
            tensor<int32, [1]> var_2670_axes_0 = const()[name = string("op_2670_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_9_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_9_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2670_cast_fp16 = layer_norm(axes = var_2670_axes_0, epsilon = layers_9_norm_feed_forward2_eps_scaled_to_fp16, x = x_219_cast_fp16)[name = string("op_2670_cast_fp16")];
            tensor<fp16, [1024]> input_273_gamma_0_to_fp16 = const()[name = string("input_273_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187328896)))];
            tensor<fp16, [1024]> input_273_beta_0_to_fp16 = const()[name = string("input_273_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187331008)))];
            fp16 input_273_epsilon_0_to_fp16 = const()[name = string("input_273_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_273_cast_fp16 = batch_norm(beta = input_273_beta_0_to_fp16, epsilon = input_273_epsilon_0_to_fp16, gamma = input_273_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2670_cast_fp16)[name = string("input_273_cast_fp16")];
            string input_275_pad_type_0 = const()[name = string("input_275_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_275_strides_0 = const()[name = string("input_275_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_275_pad_0 = const()[name = string("input_275_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_275_dilations_0 = const()[name = string("input_275_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_275_groups_0 = const()[name = string("input_275_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_9_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187333120))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190478912))))[name = string("layers_9_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_275_cast_fp16 = conv(dilations = input_275_dilations_0, groups = input_275_groups_0, pad = input_275_pad_0, pad_type = input_275_pad_type_0, strides = input_275_strides_0, weight = layers_9_feed_forward2_linear1_weight_to_fp16_palettized, x = input_273_cast_fp16)[name = string("input_275_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_277_cast_fp16 = silu(x = input_275_cast_fp16)[name = string("input_277_cast_fp16")];
            string var_2687_pad_type_0 = const()[name = string("op_2687_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2687_strides_0 = const()[name = string("op_2687_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2687_pad_0 = const()[name = string("op_2687_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2687_dilations_0 = const()[name = string("op_2687_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2687_groups_0 = const()[name = string("op_2687_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_2688_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190511744))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193657536))))[name = string("op_2688_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2688_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_2687_dilations_0, groups = var_2687_groups_0, pad = var_2687_pad_0, pad_type = var_2687_pad_type_0, strides = var_2687_strides_0, weight = op_2688_weight_0_to_fp16_palettized, x = input_277_cast_fp16)[name = string("op_2688_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_221_cast_fp16 = add(x = x_219_cast_fp16, y = var_2688_cast_fp16)[name = string("x_221_cast_fp16")];
            tensor<int32, [1]> var_2704_axes_0 = const()[name = string("op_2704_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_9_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_9_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2704_cast_fp16 = layer_norm(axes = var_2704_axes_0, epsilon = layers_9_norm_out_eps_scaled_to_fp16, x = x_221_cast_fp16)[name = string("op_2704_cast_fp16")];
            tensor<fp16, [1024]> x_223_gamma_0_to_fp16 = const()[name = string("x_223_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193665792)))];
            tensor<fp16, [1024]> x_223_beta_0_to_fp16 = const()[name = string("x_223_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193667904)))];
            fp16 x_223_epsilon_0_to_fp16 = const()[name = string("x_223_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_223_cast_fp16 = batch_norm(beta = x_223_beta_0_to_fp16, epsilon = x_223_epsilon_0_to_fp16, gamma = x_223_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2704_cast_fp16)[name = string("x_223_cast_fp16")];
            int32 var_2723 = const()[name = string("op_2723"), val = int32(1)];
            tensor<int32, [1]> var_2750_axes_0 = const()[name = string("op_2750_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_10_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_10_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2750_cast_fp16 = layer_norm(axes = var_2750_axes_0, epsilon = layers_10_norm_feed_forward1_eps_scaled_to_fp16, x = x_223_cast_fp16)[name = string("op_2750_cast_fp16")];
            tensor<fp16, [1024]> input_279_gamma_0_to_fp16 = const()[name = string("input_279_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193670016)))];
            tensor<fp16, [1024]> input_279_beta_0_to_fp16 = const()[name = string("input_279_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193672128)))];
            fp16 input_279_epsilon_0_to_fp16 = const()[name = string("input_279_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_279_cast_fp16 = batch_norm(beta = input_279_beta_0_to_fp16, epsilon = input_279_epsilon_0_to_fp16, gamma = input_279_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2750_cast_fp16)[name = string("input_279_cast_fp16")];
            string input_281_pad_type_0 = const()[name = string("input_281_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_281_strides_0 = const()[name = string("input_281_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_281_pad_0 = const()[name = string("input_281_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_281_dilations_0 = const()[name = string("input_281_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_281_groups_0 = const()[name = string("input_281_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_10_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193674240))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196820032))))[name = string("layers_10_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_281_cast_fp16 = conv(dilations = input_281_dilations_0, groups = input_281_groups_0, pad = input_281_pad_0, pad_type = input_281_pad_type_0, strides = input_281_strides_0, weight = layers_10_feed_forward1_linear1_weight_to_fp16_palettized, x = input_279_cast_fp16)[name = string("input_281_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_283_cast_fp16 = silu(x = input_281_cast_fp16)[name = string("input_283_cast_fp16")];
            string var_2767_pad_type_0 = const()[name = string("op_2767_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2767_strides_0 = const()[name = string("op_2767_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2767_pad_0 = const()[name = string("op_2767_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2767_dilations_0 = const()[name = string("op_2767_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2767_groups_0 = const()[name = string("op_2767_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_2768_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196852864))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(199998656))))[name = string("op_2768_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2768_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_2767_dilations_0, groups = var_2767_groups_0, pad = var_2767_pad_0, pad_type = var_2767_pad_type_0, strides = var_2767_strides_0, weight = op_2768_weight_0_to_fp16_palettized, x = input_283_cast_fp16)[name = string("op_2768_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_225_cast_fp16 = add(x = x_223_cast_fp16, y = var_2768_cast_fp16)[name = string("x_225_cast_fp16")];
            tensor<int32, [1]> var_2784_axes_0 = const()[name = string("op_2784_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_10_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_10_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2784_cast_fp16 = layer_norm(axes = var_2784_axes_0, epsilon = layers_10_norm_self_att_eps_scaled_to_fp16, x = x_225_cast_fp16)[name = string("op_2784_cast_fp16")];
            tensor<fp16, [1024]> x_227_gamma_0_to_fp16 = const()[name = string("x_227_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200006912)))];
            tensor<fp16, [1024]> x_227_beta_0_to_fp16 = const()[name = string("x_227_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200009024)))];
            fp16 x_227_epsilon_0_to_fp16 = const()[name = string("x_227_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_227_cast_fp16 = batch_norm(beta = x_227_beta_0_to_fp16, epsilon = x_227_epsilon_0_to_fp16, gamma = x_227_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2784_cast_fp16)[name = string("x_227_cast_fp16")];
            string q_21_pad_type_0 = const()[name = string("q_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_21_strides_0 = const()[name = string("q_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_21_pad_0 = const()[name = string("q_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_21_dilations_0 = const()[name = string("q_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_21_groups_0 = const()[name = string("q_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200011136))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200797632))))[name = string("layers_10_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_21_cast_fp16 = conv(dilations = q_21_dilations_0, groups = q_21_groups_0, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = q_21_strides_0, weight = layers_10_self_attn_linear_q_weight_to_fp16_palettized, x = x_227_cast_fp16)[name = string("q_21_cast_fp16")];
            string k_21_pad_type_0 = const()[name = string("k_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_21_strides_0 = const()[name = string("k_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_21_pad_0 = const()[name = string("k_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_21_dilations_0 = const()[name = string("k_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_21_groups_0 = const()[name = string("k_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200805888))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201592384))))[name = string("layers_10_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_21_cast_fp16 = conv(dilations = k_21_dilations_0, groups = k_21_groups_0, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = k_21_strides_0, weight = layers_10_self_attn_linear_k_weight_to_fp16_palettized, x = x_227_cast_fp16)[name = string("k_21_cast_fp16")];
            string v_21_pad_type_0 = const()[name = string("v_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_21_strides_0 = const()[name = string("v_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_21_pad_0 = const()[name = string("v_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_21_dilations_0 = const()[name = string("v_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_21_groups_0 = const()[name = string("v_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201600640))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202387136))))[name = string("layers_10_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_21_cast_fp16 = conv(dilations = v_21_dilations_0, groups = v_21_groups_0, pad = v_21_pad_0, pad_type = v_21_pad_type_0, strides = v_21_strides_0, weight = layers_10_self_attn_linear_v_weight_to_fp16_palettized, x = x_227_cast_fp16)[name = string("v_21_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_21_to_fp16 = const()[name = string("bv_all_21_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202395392)))];
            tensor<fp16, [1, 1024, 1, 188]> var_2816_cast_fp16 = add(x = q_21_cast_fp16, y = bv_all_21_to_fp16)[name = string("op_2816_cast_fp16")];
            tensor<int32, [3]> var_2817 = const()[name = string("op_2817"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_21_cast_fp16 = reshape(shape = var_2817, x = var_2816_cast_fp16)[name = string("qb_21_cast_fp16")];
            bool bd_all_41_transpose_x_0 = const()[name = string("bd_all_41_transpose_x_0"), val = bool(false)];
            bool bd_all_41_transpose_y_0 = const()[name = string("bd_all_41_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_10_self_attn_pos_proj_to_fp16 = const()[name = string("layers_10_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202397504)))];
            tensor<fp16, [8, 375, 188]> bd_all_41_cast_fp16 = matmul(transpose_x = bd_all_41_transpose_x_0, transpose_y = bd_all_41_transpose_y_0, x = layers_10_self_attn_pos_proj_to_fp16, y = qb_21_cast_fp16)[name = string("bd_all_41_cast_fp16")];
            tensor<int32, [3]> x_229_perm_0 = const()[name = string("x_229_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_231_pad_0 = const()[name = string("x_231_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_231_mode_0 = const()[name = string("x_231_mode_0"), val = string("constant")];
            fp16 const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_229_cast_fp16 = transpose(perm = x_229_perm_0, x = bd_all_41_cast_fp16)[name = string("transpose_83")];
            tensor<fp16, [8, 188, 376]> x_231_cast_fp16 = pad(constant_val = const_50_to_fp16, mode = x_231_mode_0, pad = x_231_pad_0, x = x_229_cast_fp16)[name = string("x_231_cast_fp16")];
            tensor<int32, [3]> var_2824 = const()[name = string("op_2824"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_233_cast_fp16 = reshape(shape = var_2824, x = x_231_cast_fp16)[name = string("x_233_cast_fp16")];
            tensor<int32, [3]> var_2827_begin_0 = const()[name = string("op_2827_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_2827_end_0 = const()[name = string("op_2827_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_2827_end_mask_0 = const()[name = string("op_2827_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_2827_cast_fp16 = slice_by_index(begin = var_2827_begin_0, end = var_2827_end_0, end_mask = var_2827_end_mask_0, x = x_233_cast_fp16)[name = string("op_2827_cast_fp16")];
            tensor<int32, [3]> var_2828 = const()[name = string("op_2828"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_235_cast_fp16 = reshape(shape = var_2828, x = var_2827_cast_fp16)[name = string("x_235_cast_fp16")];
            tensor<int32, [3]> bd_all_43_begin_0 = const()[name = string("bd_all_43_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_43_end_0 = const()[name = string("bd_all_43_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_43_end_mask_0 = const()[name = string("bd_all_43_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_43_cast_fp16 = slice_by_index(begin = bd_all_43_begin_0, end = bd_all_43_end_0, end_mask = bd_all_43_end_mask_0, x = x_235_cast_fp16)[name = string("bd_all_43_cast_fp16")];
            tensor<int32, [4]> var_2833 = const()[name = string("op_2833"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_2834_cast_fp16 = reshape(shape = var_2833, x = q_21_cast_fp16)[name = string("op_2834_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_2836_to_fp16 = const()[name = string("op_2836_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203165568)))];
            tensor<fp16, [8, 128, 1, 188]> var_2837_cast_fp16 = add(x = var_2834_cast_fp16, y = var_2836_to_fp16)[name = string("op_2837_cast_fp16")];
            tensor<int32, [4]> var_2838 = const()[name = string("op_2838"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_21_cast_fp16 = reshape(shape = var_2838, x = k_21_cast_fp16)[name = string("kh_21_cast_fp16")];
            tensor<int32, [4]> var_2840 = const()[name = string("op_2840"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_21_cast_fp16 = reshape(shape = var_2840, x = v_21_cast_fp16)[name = string("vh_21_cast_fp16")];
            tensor<int32, [4]> var_2842 = const()[name = string("op_2842"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_21_equation_0 = const()[name = string("ac_21_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_2843_cast_fp16 = transpose(perm = var_2842, x = kh_21_cast_fp16)[name = string("transpose_82")];
            tensor<fp16, [8, 188, 1, 188]> ac_21_cast_fp16 = einsum(equation = ac_21_equation_0, values = (var_2843_cast_fp16, var_2837_cast_fp16))[name = string("ac_21_cast_fp16")];
            tensor<int32, [3]> var_2846_perm_0 = const()[name = string("op_2846_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_2847_axes_0 = const()[name = string("op_2847_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_2846_cast_fp16 = transpose(perm = var_2846_perm_0, x = bd_all_43_cast_fp16)[name = string("transpose_81")];
            tensor<fp16, [8, 188, 1, 188]> var_2847_cast_fp16 = expand_dims(axes = var_2847_axes_0, x = var_2846_cast_fp16)[name = string("op_2847_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_2848_cast_fp16 = add(x = ac_21_cast_fp16, y = var_2847_cast_fp16)[name = string("op_2848_cast_fp16")];
            fp16 var_2849_to_fp16 = const()[name = string("op_2849_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_41_cast_fp16 = mul(x = var_2848_cast_fp16, y = var_2849_to_fp16)[name = string("scores_41_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_43_cast_fp16 = add(x = scores_41_cast_fp16, y = key_bias)[name = string("scores_43_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_2852_cast_fp16 = softmax(axis = var_2723, x = scores_43_cast_fp16)[name = string("op_2852_cast_fp16")];
            tensor<int32, [4]> transpose_58_perm_0 = const()[name = string("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_20_perm_0 = const()[name = string("transpose_20_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_104 = const()[name = string("concat_104"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_20_cast_fp16 = transpose(perm = transpose_20_perm_0, x = var_2852_cast_fp16)[name = string("transpose_80")];
            tensor<fp16, [8, 188, 188]> reshape_30_cast_fp16 = reshape(shape = concat_104, x = transpose_20_cast_fp16)[name = string("reshape_30_cast_fp16")];
            tensor<int32, [3]> concat_105 = const()[name = string("concat_105"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_58_cast_fp16 = transpose(perm = transpose_58_perm_0, x = vh_21_cast_fp16)[name = string("transpose_79")];
            tensor<fp16, [8, 188, 128]> reshape_31_cast_fp16 = reshape(shape = concat_105, x = transpose_58_cast_fp16)[name = string("reshape_31_cast_fp16")];
            bool matmul_10_transpose_x_0 = const()[name = string("matmul_10_transpose_x_0"), val = bool(false)];
            bool matmul_10_transpose_y_0 = const()[name = string("matmul_10_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = reshape_30_cast_fp16, y = reshape_31_cast_fp16)[name = string("matmul_10_cast_fp16")];
            tensor<int32, [4]> concat_109 = const()[name = string("concat_109"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_32_cast_fp16 = reshape(shape = concat_109, x = matmul_10_cast_fp16)[name = string("reshape_32_cast_fp16")];
            tensor<int32, [4]> ctx_21_perm_0 = const()[name = string("ctx_21_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_2857 = const()[name = string("op_2857"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_21_cast_fp16 = transpose(perm = ctx_21_perm_0, x = reshape_32_cast_fp16)[name = string("transpose_78")];
            tensor<fp16, [1, 1024, 1, 188]> input_285_cast_fp16 = reshape(shape = var_2857, x = ctx_21_cast_fp16)[name = string("input_285_cast_fp16")];
            string var_2864_pad_type_0 = const()[name = string("op_2864_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2864_strides_0 = const()[name = string("op_2864_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2864_pad_0 = const()[name = string("op_2864_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2864_dilations_0 = const()[name = string("op_2864_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2864_groups_0 = const()[name = string("op_2864_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203167680))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203954176))))[name = string("layers_10_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2864_cast_fp16 = conv(dilations = var_2864_dilations_0, groups = var_2864_groups_0, pad = var_2864_pad_0, pad_type = var_2864_pad_type_0, strides = var_2864_strides_0, weight = layers_10_self_attn_linear_out_weight_to_fp16_palettized, x = input_285_cast_fp16)[name = string("op_2864_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_237_cast_fp16 = add(x = x_225_cast_fp16, y = var_2864_cast_fp16)[name = string("x_237_cast_fp16")];
            tensor<int32, [1]> var_2880_axes_0 = const()[name = string("op_2880_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_10_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_10_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2880_cast_fp16 = layer_norm(axes = var_2880_axes_0, epsilon = layers_10_norm_conv_eps_scaled_to_fp16, x = x_237_cast_fp16)[name = string("op_2880_cast_fp16")];
            tensor<fp16, [1024]> input_287_gamma_0_to_fp16 = const()[name = string("input_287_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203962432)))];
            tensor<fp16, [1024]> input_287_beta_0_to_fp16 = const()[name = string("input_287_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203964544)))];
            fp16 input_287_epsilon_0_to_fp16 = const()[name = string("input_287_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_287_cast_fp16 = batch_norm(beta = input_287_beta_0_to_fp16, epsilon = input_287_epsilon_0_to_fp16, gamma = input_287_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2880_cast_fp16)[name = string("input_287_cast_fp16")];
            string input_289_pad_type_0 = const()[name = string("input_289_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_289_strides_0 = const()[name = string("input_289_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_289_pad_0 = const()[name = string("input_289_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_289_dilations_0 = const()[name = string("input_289_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_289_groups_0 = const()[name = string("input_289_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_10_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203966656))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205539584))))[name = string("layers_10_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_289_cast_fp16 = conv(dilations = input_289_dilations_0, groups = input_289_groups_0, pad = input_289_pad_0, pad_type = input_289_pad_type_0, strides = input_289_strides_0, weight = layers_10_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_287_cast_fp16)[name = string("input_289_cast_fp16")];
            int32 x_239_split_num_splits_0 = const()[name = string("x_239_split_num_splits_0"), val = int32(2)];
            int32 x_239_split_axis_0 = const()[name = string("x_239_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_239_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_289_cast_fp16)[name = string("x_239_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_239_split_1_sigmoid_cast_fp16 = sigmoid(x = x_239_split_cast_fp16_1)[name = string("x_239_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_239_cast_fp16 = mul(x = x_239_split_cast_fp16_0, y = x_239_split_1_sigmoid_cast_fp16)[name = string("x_239_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_291_cast_fp16 = mul(x = x_239_cast_fp16, y = pad_mask)[name = string("input_291_cast_fp16")];
            string input_293_pad_type_0 = const()[name = string("input_293_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_293_pad_0 = const()[name = string("input_293_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_293_groups_0 = const()[name = string("input_293_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_293_strides_0 = const()[name = string("input_293_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_293_dilations_0 = const()[name = string("input_293_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_123_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205556032))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205563008))))[name = string("const_123_to_fp16_palettized")];
            tensor<fp16, [1024]> const_124_to_fp16 = const()[name = string("const_124_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205571264)))];
            tensor<fp16, [1, 1024, 1, 188]> input_295_cast_fp16 = conv(bias = const_124_to_fp16, dilations = input_293_dilations_0, groups = input_293_groups_0, pad = input_293_pad_0, pad_type = input_293_pad_type_0, strides = input_293_strides_0, weight = const_123_to_fp16_palettized, x = input_291_cast_fp16)[name = string("input_295_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_297_cast_fp16 = silu(x = input_295_cast_fp16)[name = string("input_297_cast_fp16")];
            string var_2912_pad_type_0 = const()[name = string("op_2912_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2912_strides_0 = const()[name = string("op_2912_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2912_pad_0 = const()[name = string("op_2912_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2912_dilations_0 = const()[name = string("op_2912_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2912_groups_0 = const()[name = string("op_2912_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_10_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205573376))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206359872))))[name = string("layers_10_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2912_cast_fp16 = conv(dilations = var_2912_dilations_0, groups = var_2912_groups_0, pad = var_2912_pad_0, pad_type = var_2912_pad_type_0, strides = var_2912_strides_0, weight = layers_10_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_297_cast_fp16)[name = string("op_2912_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_241_cast_fp16 = add(x = x_237_cast_fp16, y = var_2912_cast_fp16)[name = string("x_241_cast_fp16")];
            tensor<int32, [1]> var_2928_axes_0 = const()[name = string("op_2928_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_10_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_10_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2928_cast_fp16 = layer_norm(axes = var_2928_axes_0, epsilon = layers_10_norm_feed_forward2_eps_scaled_to_fp16, x = x_241_cast_fp16)[name = string("op_2928_cast_fp16")];
            tensor<fp16, [1024]> input_299_gamma_0_to_fp16 = const()[name = string("input_299_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206368128)))];
            tensor<fp16, [1024]> input_299_beta_0_to_fp16 = const()[name = string("input_299_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206370240)))];
            fp16 input_299_epsilon_0_to_fp16 = const()[name = string("input_299_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_299_cast_fp16 = batch_norm(beta = input_299_beta_0_to_fp16, epsilon = input_299_epsilon_0_to_fp16, gamma = input_299_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2928_cast_fp16)[name = string("input_299_cast_fp16")];
            string input_301_pad_type_0 = const()[name = string("input_301_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_301_strides_0 = const()[name = string("input_301_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_301_pad_0 = const()[name = string("input_301_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_301_dilations_0 = const()[name = string("input_301_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_301_groups_0 = const()[name = string("input_301_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_10_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(206372352))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209518144))))[name = string("layers_10_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_301_cast_fp16 = conv(dilations = input_301_dilations_0, groups = input_301_groups_0, pad = input_301_pad_0, pad_type = input_301_pad_type_0, strides = input_301_strides_0, weight = layers_10_feed_forward2_linear1_weight_to_fp16_palettized, x = input_299_cast_fp16)[name = string("input_301_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_303_cast_fp16 = silu(x = input_301_cast_fp16)[name = string("input_303_cast_fp16")];
            string var_2945_pad_type_0 = const()[name = string("op_2945_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2945_strides_0 = const()[name = string("op_2945_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2945_pad_0 = const()[name = string("op_2945_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2945_dilations_0 = const()[name = string("op_2945_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2945_groups_0 = const()[name = string("op_2945_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_2946_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209550976))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212696768))))[name = string("op_2946_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_2946_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_2945_dilations_0, groups = var_2945_groups_0, pad = var_2945_pad_0, pad_type = var_2945_pad_type_0, strides = var_2945_strides_0, weight = op_2946_weight_0_to_fp16_palettized, x = input_303_cast_fp16)[name = string("op_2946_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_243_cast_fp16 = add(x = x_241_cast_fp16, y = var_2946_cast_fp16)[name = string("x_243_cast_fp16")];
            tensor<int32, [1]> var_2962_axes_0 = const()[name = string("op_2962_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_10_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_10_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_2962_cast_fp16 = layer_norm(axes = var_2962_axes_0, epsilon = layers_10_norm_out_eps_scaled_to_fp16, x = x_243_cast_fp16)[name = string("op_2962_cast_fp16")];
            tensor<fp16, [1024]> x_245_gamma_0_to_fp16 = const()[name = string("x_245_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212705024)))];
            tensor<fp16, [1024]> x_245_beta_0_to_fp16 = const()[name = string("x_245_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212707136)))];
            fp16 x_245_epsilon_0_to_fp16 = const()[name = string("x_245_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_245_cast_fp16 = batch_norm(beta = x_245_beta_0_to_fp16, epsilon = x_245_epsilon_0_to_fp16, gamma = x_245_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_2962_cast_fp16)[name = string("x_245_cast_fp16")];
            int32 var_2981 = const()[name = string("op_2981"), val = int32(1)];
            tensor<int32, [1]> var_3008_axes_0 = const()[name = string("op_3008_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_11_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_11_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3008_cast_fp16 = layer_norm(axes = var_3008_axes_0, epsilon = layers_11_norm_feed_forward1_eps_scaled_to_fp16, x = x_245_cast_fp16)[name = string("op_3008_cast_fp16")];
            tensor<fp16, [1024]> input_305_gamma_0_to_fp16 = const()[name = string("input_305_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212709248)))];
            tensor<fp16, [1024]> input_305_beta_0_to_fp16 = const()[name = string("input_305_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212711360)))];
            fp16 input_305_epsilon_0_to_fp16 = const()[name = string("input_305_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_305_cast_fp16 = batch_norm(beta = input_305_beta_0_to_fp16, epsilon = input_305_epsilon_0_to_fp16, gamma = input_305_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3008_cast_fp16)[name = string("input_305_cast_fp16")];
            string input_307_pad_type_0 = const()[name = string("input_307_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_307_strides_0 = const()[name = string("input_307_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_307_pad_0 = const()[name = string("input_307_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_307_dilations_0 = const()[name = string("input_307_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_307_groups_0 = const()[name = string("input_307_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_11_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212713472))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215859264))))[name = string("layers_11_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_307_cast_fp16 = conv(dilations = input_307_dilations_0, groups = input_307_groups_0, pad = input_307_pad_0, pad_type = input_307_pad_type_0, strides = input_307_strides_0, weight = layers_11_feed_forward1_linear1_weight_to_fp16_palettized, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_309_cast_fp16 = silu(x = input_307_cast_fp16)[name = string("input_309_cast_fp16")];
            string var_3025_pad_type_0 = const()[name = string("op_3025_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3025_strides_0 = const()[name = string("op_3025_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3025_pad_0 = const()[name = string("op_3025_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3025_dilations_0 = const()[name = string("op_3025_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3025_groups_0 = const()[name = string("op_3025_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_3026_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215892096))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219037888))))[name = string("op_3026_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3026_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_3025_dilations_0, groups = var_3025_groups_0, pad = var_3025_pad_0, pad_type = var_3025_pad_type_0, strides = var_3025_strides_0, weight = op_3026_weight_0_to_fp16_palettized, x = input_309_cast_fp16)[name = string("op_3026_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_247_cast_fp16 = add(x = x_245_cast_fp16, y = var_3026_cast_fp16)[name = string("x_247_cast_fp16")];
            tensor<int32, [1]> var_3042_axes_0 = const()[name = string("op_3042_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_11_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_11_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3042_cast_fp16 = layer_norm(axes = var_3042_axes_0, epsilon = layers_11_norm_self_att_eps_scaled_to_fp16, x = x_247_cast_fp16)[name = string("op_3042_cast_fp16")];
            tensor<fp16, [1024]> x_249_gamma_0_to_fp16 = const()[name = string("x_249_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219046144)))];
            tensor<fp16, [1024]> x_249_beta_0_to_fp16 = const()[name = string("x_249_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219048256)))];
            fp16 x_249_epsilon_0_to_fp16 = const()[name = string("x_249_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_249_cast_fp16 = batch_norm(beta = x_249_beta_0_to_fp16, epsilon = x_249_epsilon_0_to_fp16, gamma = x_249_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3042_cast_fp16)[name = string("x_249_cast_fp16")];
            string q_23_pad_type_0 = const()[name = string("q_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_23_strides_0 = const()[name = string("q_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_23_pad_0 = const()[name = string("q_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_23_dilations_0 = const()[name = string("q_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_23_groups_0 = const()[name = string("q_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219050368))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219836864))))[name = string("layers_11_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_23_cast_fp16 = conv(dilations = q_23_dilations_0, groups = q_23_groups_0, pad = q_23_pad_0, pad_type = q_23_pad_type_0, strides = q_23_strides_0, weight = layers_11_self_attn_linear_q_weight_to_fp16_palettized, x = x_249_cast_fp16)[name = string("q_23_cast_fp16")];
            string k_23_pad_type_0 = const()[name = string("k_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_23_strides_0 = const()[name = string("k_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_23_pad_0 = const()[name = string("k_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_23_dilations_0 = const()[name = string("k_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_23_groups_0 = const()[name = string("k_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219845120))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220631616))))[name = string("layers_11_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_23_cast_fp16 = conv(dilations = k_23_dilations_0, groups = k_23_groups_0, pad = k_23_pad_0, pad_type = k_23_pad_type_0, strides = k_23_strides_0, weight = layers_11_self_attn_linear_k_weight_to_fp16_palettized, x = x_249_cast_fp16)[name = string("k_23_cast_fp16")];
            string v_23_pad_type_0 = const()[name = string("v_23_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_23_strides_0 = const()[name = string("v_23_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_23_pad_0 = const()[name = string("v_23_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_23_dilations_0 = const()[name = string("v_23_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_23_groups_0 = const()[name = string("v_23_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220639872))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221426368))))[name = string("layers_11_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_23_cast_fp16 = conv(dilations = v_23_dilations_0, groups = v_23_groups_0, pad = v_23_pad_0, pad_type = v_23_pad_type_0, strides = v_23_strides_0, weight = layers_11_self_attn_linear_v_weight_to_fp16_palettized, x = x_249_cast_fp16)[name = string("v_23_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_23_to_fp16 = const()[name = string("bv_all_23_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221434624)))];
            tensor<fp16, [1, 1024, 1, 188]> var_3074_cast_fp16 = add(x = q_23_cast_fp16, y = bv_all_23_to_fp16)[name = string("op_3074_cast_fp16")];
            tensor<int32, [3]> var_3075 = const()[name = string("op_3075"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_23_cast_fp16 = reshape(shape = var_3075, x = var_3074_cast_fp16)[name = string("qb_23_cast_fp16")];
            bool bd_all_45_transpose_x_0 = const()[name = string("bd_all_45_transpose_x_0"), val = bool(false)];
            bool bd_all_45_transpose_y_0 = const()[name = string("bd_all_45_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_11_self_attn_pos_proj_to_fp16 = const()[name = string("layers_11_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221436736)))];
            tensor<fp16, [8, 375, 188]> bd_all_45_cast_fp16 = matmul(transpose_x = bd_all_45_transpose_x_0, transpose_y = bd_all_45_transpose_y_0, x = layers_11_self_attn_pos_proj_to_fp16, y = qb_23_cast_fp16)[name = string("bd_all_45_cast_fp16")];
            tensor<int32, [3]> x_251_perm_0 = const()[name = string("x_251_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_253_pad_0 = const()[name = string("x_253_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_253_mode_0 = const()[name = string("x_253_mode_0"), val = string("constant")];
            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_251_cast_fp16 = transpose(perm = x_251_perm_0, x = bd_all_45_cast_fp16)[name = string("transpose_77")];
            tensor<fp16, [8, 188, 376]> x_253_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = x_253_mode_0, pad = x_253_pad_0, x = x_251_cast_fp16)[name = string("x_253_cast_fp16")];
            tensor<int32, [3]> var_3082 = const()[name = string("op_3082"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_255_cast_fp16 = reshape(shape = var_3082, x = x_253_cast_fp16)[name = string("x_255_cast_fp16")];
            tensor<int32, [3]> var_3085_begin_0 = const()[name = string("op_3085_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_3085_end_0 = const()[name = string("op_3085_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_3085_end_mask_0 = const()[name = string("op_3085_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_3085_cast_fp16 = slice_by_index(begin = var_3085_begin_0, end = var_3085_end_0, end_mask = var_3085_end_mask_0, x = x_255_cast_fp16)[name = string("op_3085_cast_fp16")];
            tensor<int32, [3]> var_3086 = const()[name = string("op_3086"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_257_cast_fp16 = reshape(shape = var_3086, x = var_3085_cast_fp16)[name = string("x_257_cast_fp16")];
            tensor<int32, [3]> bd_all_47_begin_0 = const()[name = string("bd_all_47_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_47_end_0 = const()[name = string("bd_all_47_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_47_end_mask_0 = const()[name = string("bd_all_47_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_47_cast_fp16 = slice_by_index(begin = bd_all_47_begin_0, end = bd_all_47_end_0, end_mask = bd_all_47_end_mask_0, x = x_257_cast_fp16)[name = string("bd_all_47_cast_fp16")];
            tensor<int32, [4]> var_3091 = const()[name = string("op_3091"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_3092_cast_fp16 = reshape(shape = var_3091, x = q_23_cast_fp16)[name = string("op_3092_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_3094_to_fp16 = const()[name = string("op_3094_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222204800)))];
            tensor<fp16, [8, 128, 1, 188]> var_3095_cast_fp16 = add(x = var_3092_cast_fp16, y = var_3094_to_fp16)[name = string("op_3095_cast_fp16")];
            tensor<int32, [4]> var_3096 = const()[name = string("op_3096"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_23_cast_fp16 = reshape(shape = var_3096, x = k_23_cast_fp16)[name = string("kh_23_cast_fp16")];
            tensor<int32, [4]> var_3098 = const()[name = string("op_3098"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_23_cast_fp16 = reshape(shape = var_3098, x = v_23_cast_fp16)[name = string("vh_23_cast_fp16")];
            tensor<int32, [4]> var_3100 = const()[name = string("op_3100"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_23_equation_0 = const()[name = string("ac_23_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_3101_cast_fp16 = transpose(perm = var_3100, x = kh_23_cast_fp16)[name = string("transpose_76")];
            tensor<fp16, [8, 188, 1, 188]> ac_23_cast_fp16 = einsum(equation = ac_23_equation_0, values = (var_3101_cast_fp16, var_3095_cast_fp16))[name = string("ac_23_cast_fp16")];
            tensor<int32, [3]> var_3104_perm_0 = const()[name = string("op_3104_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_3105_axes_0 = const()[name = string("op_3105_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_3104_cast_fp16 = transpose(perm = var_3104_perm_0, x = bd_all_47_cast_fp16)[name = string("transpose_75")];
            tensor<fp16, [8, 188, 1, 188]> var_3105_cast_fp16 = expand_dims(axes = var_3105_axes_0, x = var_3104_cast_fp16)[name = string("op_3105_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_3106_cast_fp16 = add(x = ac_23_cast_fp16, y = var_3105_cast_fp16)[name = string("op_3106_cast_fp16")];
            fp16 var_3107_to_fp16 = const()[name = string("op_3107_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_45_cast_fp16 = mul(x = var_3106_cast_fp16, y = var_3107_to_fp16)[name = string("scores_45_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_47_cast_fp16 = add(x = scores_45_cast_fp16, y = key_bias)[name = string("scores_47_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_3110_cast_fp16 = softmax(axis = var_2981, x = scores_47_cast_fp16)[name = string("op_3110_cast_fp16")];
            tensor<int32, [4]> transpose_59_perm_0 = const()[name = string("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_22_perm_0 = const()[name = string("transpose_22_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_114 = const()[name = string("concat_114"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_22_cast_fp16 = transpose(perm = transpose_22_perm_0, x = var_3110_cast_fp16)[name = string("transpose_74")];
            tensor<fp16, [8, 188, 188]> reshape_33_cast_fp16 = reshape(shape = concat_114, x = transpose_22_cast_fp16)[name = string("reshape_33_cast_fp16")];
            tensor<int32, [3]> concat_115 = const()[name = string("concat_115"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_59_cast_fp16 = transpose(perm = transpose_59_perm_0, x = vh_23_cast_fp16)[name = string("transpose_73")];
            tensor<fp16, [8, 188, 128]> reshape_34_cast_fp16 = reshape(shape = concat_115, x = transpose_59_cast_fp16)[name = string("reshape_34_cast_fp16")];
            bool matmul_11_transpose_x_0 = const()[name = string("matmul_11_transpose_x_0"), val = bool(false)];
            bool matmul_11_transpose_y_0 = const()[name = string("matmul_11_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = reshape_33_cast_fp16, y = reshape_34_cast_fp16)[name = string("matmul_11_cast_fp16")];
            tensor<int32, [4]> concat_119 = const()[name = string("concat_119"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_35_cast_fp16 = reshape(shape = concat_119, x = matmul_11_cast_fp16)[name = string("reshape_35_cast_fp16")];
            tensor<int32, [4]> ctx_23_perm_0 = const()[name = string("ctx_23_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_3115 = const()[name = string("op_3115"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_23_cast_fp16 = transpose(perm = ctx_23_perm_0, x = reshape_35_cast_fp16)[name = string("transpose_72")];
            tensor<fp16, [1, 1024, 1, 188]> input_311_cast_fp16 = reshape(shape = var_3115, x = ctx_23_cast_fp16)[name = string("input_311_cast_fp16")];
            string var_3122_pad_type_0 = const()[name = string("op_3122_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3122_strides_0 = const()[name = string("op_3122_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3122_pad_0 = const()[name = string("op_3122_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3122_dilations_0 = const()[name = string("op_3122_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3122_groups_0 = const()[name = string("op_3122_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222206912))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(222993408))))[name = string("layers_11_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3122_cast_fp16 = conv(dilations = var_3122_dilations_0, groups = var_3122_groups_0, pad = var_3122_pad_0, pad_type = var_3122_pad_type_0, strides = var_3122_strides_0, weight = layers_11_self_attn_linear_out_weight_to_fp16_palettized, x = input_311_cast_fp16)[name = string("op_3122_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_259_cast_fp16 = add(x = x_247_cast_fp16, y = var_3122_cast_fp16)[name = string("x_259_cast_fp16")];
            tensor<int32, [1]> var_3138_axes_0 = const()[name = string("op_3138_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_11_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_11_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3138_cast_fp16 = layer_norm(axes = var_3138_axes_0, epsilon = layers_11_norm_conv_eps_scaled_to_fp16, x = x_259_cast_fp16)[name = string("op_3138_cast_fp16")];
            tensor<fp16, [1024]> input_313_gamma_0_to_fp16 = const()[name = string("input_313_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223001664)))];
            tensor<fp16, [1024]> input_313_beta_0_to_fp16 = const()[name = string("input_313_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223003776)))];
            fp16 input_313_epsilon_0_to_fp16 = const()[name = string("input_313_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_313_cast_fp16 = batch_norm(beta = input_313_beta_0_to_fp16, epsilon = input_313_epsilon_0_to_fp16, gamma = input_313_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3138_cast_fp16)[name = string("input_313_cast_fp16")];
            string input_315_pad_type_0 = const()[name = string("input_315_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_315_strides_0 = const()[name = string("input_315_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_315_pad_0 = const()[name = string("input_315_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_315_dilations_0 = const()[name = string("input_315_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_315_groups_0 = const()[name = string("input_315_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_11_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223005888))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224578816))))[name = string("layers_11_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_315_cast_fp16 = conv(dilations = input_315_dilations_0, groups = input_315_groups_0, pad = input_315_pad_0, pad_type = input_315_pad_type_0, strides = input_315_strides_0, weight = layers_11_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_313_cast_fp16)[name = string("input_315_cast_fp16")];
            int32 x_261_split_num_splits_0 = const()[name = string("x_261_split_num_splits_0"), val = int32(2)];
            int32 x_261_split_axis_0 = const()[name = string("x_261_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_261_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_315_cast_fp16)[name = string("x_261_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_261_split_1_sigmoid_cast_fp16 = sigmoid(x = x_261_split_cast_fp16_1)[name = string("x_261_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_261_cast_fp16 = mul(x = x_261_split_cast_fp16_0, y = x_261_split_1_sigmoid_cast_fp16)[name = string("x_261_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_317_cast_fp16 = mul(x = x_261_cast_fp16, y = pad_mask)[name = string("input_317_cast_fp16")];
            string input_319_pad_type_0 = const()[name = string("input_319_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_319_pad_0 = const()[name = string("input_319_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_319_groups_0 = const()[name = string("input_319_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_319_strides_0 = const()[name = string("input_319_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_319_dilations_0 = const()[name = string("input_319_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_125_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224595264))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224602240))))[name = string("const_125_to_fp16_palettized")];
            tensor<fp16, [1024]> const_126_to_fp16 = const()[name = string("const_126_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224610496)))];
            tensor<fp16, [1, 1024, 1, 188]> input_321_cast_fp16 = conv(bias = const_126_to_fp16, dilations = input_319_dilations_0, groups = input_319_groups_0, pad = input_319_pad_0, pad_type = input_319_pad_type_0, strides = input_319_strides_0, weight = const_125_to_fp16_palettized, x = input_317_cast_fp16)[name = string("input_321_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_323_cast_fp16 = silu(x = input_321_cast_fp16)[name = string("input_323_cast_fp16")];
            string var_3170_pad_type_0 = const()[name = string("op_3170_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3170_strides_0 = const()[name = string("op_3170_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3170_pad_0 = const()[name = string("op_3170_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3170_dilations_0 = const()[name = string("op_3170_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3170_groups_0 = const()[name = string("op_3170_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_11_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(224612608))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225399104))))[name = string("layers_11_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3170_cast_fp16 = conv(dilations = var_3170_dilations_0, groups = var_3170_groups_0, pad = var_3170_pad_0, pad_type = var_3170_pad_type_0, strides = var_3170_strides_0, weight = layers_11_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_323_cast_fp16)[name = string("op_3170_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_263_cast_fp16 = add(x = x_259_cast_fp16, y = var_3170_cast_fp16)[name = string("x_263_cast_fp16")];
            tensor<int32, [1]> var_3186_axes_0 = const()[name = string("op_3186_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_11_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_11_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3186_cast_fp16 = layer_norm(axes = var_3186_axes_0, epsilon = layers_11_norm_feed_forward2_eps_scaled_to_fp16, x = x_263_cast_fp16)[name = string("op_3186_cast_fp16")];
            tensor<fp16, [1024]> input_325_gamma_0_to_fp16 = const()[name = string("input_325_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225407360)))];
            tensor<fp16, [1024]> input_325_beta_0_to_fp16 = const()[name = string("input_325_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225409472)))];
            fp16 input_325_epsilon_0_to_fp16 = const()[name = string("input_325_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_325_cast_fp16 = batch_norm(beta = input_325_beta_0_to_fp16, epsilon = input_325_epsilon_0_to_fp16, gamma = input_325_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3186_cast_fp16)[name = string("input_325_cast_fp16")];
            string input_327_pad_type_0 = const()[name = string("input_327_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_327_strides_0 = const()[name = string("input_327_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_327_pad_0 = const()[name = string("input_327_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_327_dilations_0 = const()[name = string("input_327_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_327_groups_0 = const()[name = string("input_327_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_11_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(225411584))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228557376))))[name = string("layers_11_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_327_cast_fp16 = conv(dilations = input_327_dilations_0, groups = input_327_groups_0, pad = input_327_pad_0, pad_type = input_327_pad_type_0, strides = input_327_strides_0, weight = layers_11_feed_forward2_linear1_weight_to_fp16_palettized, x = input_325_cast_fp16)[name = string("input_327_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_329_cast_fp16 = silu(x = input_327_cast_fp16)[name = string("input_329_cast_fp16")];
            string var_3203_pad_type_0 = const()[name = string("op_3203_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3203_strides_0 = const()[name = string("op_3203_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3203_pad_0 = const()[name = string("op_3203_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3203_dilations_0 = const()[name = string("op_3203_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3203_groups_0 = const()[name = string("op_3203_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_3204_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228590208))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231736000))))[name = string("op_3204_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3204_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_3203_dilations_0, groups = var_3203_groups_0, pad = var_3203_pad_0, pad_type = var_3203_pad_type_0, strides = var_3203_strides_0, weight = op_3204_weight_0_to_fp16_palettized, x = input_329_cast_fp16)[name = string("op_3204_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_265_cast_fp16 = add(x = x_263_cast_fp16, y = var_3204_cast_fp16)[name = string("x_265_cast_fp16")];
            tensor<int32, [1]> var_3220_axes_0 = const()[name = string("op_3220_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_11_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_11_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3220_cast_fp16 = layer_norm(axes = var_3220_axes_0, epsilon = layers_11_norm_out_eps_scaled_to_fp16, x = x_265_cast_fp16)[name = string("op_3220_cast_fp16")];
            tensor<fp16, [1024]> x_267_gamma_0_to_fp16 = const()[name = string("x_267_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231744256)))];
            tensor<fp16, [1024]> x_267_beta_0_to_fp16 = const()[name = string("x_267_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231746368)))];
            fp16 x_267_epsilon_0_to_fp16 = const()[name = string("x_267_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_267_cast_fp16 = batch_norm(beta = x_267_beta_0_to_fp16, epsilon = x_267_epsilon_0_to_fp16, gamma = x_267_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3220_cast_fp16)[name = string("x_267_cast_fp16")];
            int32 var_3239 = const()[name = string("op_3239"), val = int32(1)];
            tensor<int32, [1]> var_3266_axes_0 = const()[name = string("op_3266_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_12_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_12_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3266_cast_fp16 = layer_norm(axes = var_3266_axes_0, epsilon = layers_12_norm_feed_forward1_eps_scaled_to_fp16, x = x_267_cast_fp16)[name = string("op_3266_cast_fp16")];
            tensor<fp16, [1024]> input_331_gamma_0_to_fp16 = const()[name = string("input_331_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231748480)))];
            tensor<fp16, [1024]> input_331_beta_0_to_fp16 = const()[name = string("input_331_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231750592)))];
            fp16 input_331_epsilon_0_to_fp16 = const()[name = string("input_331_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_331_cast_fp16 = batch_norm(beta = input_331_beta_0_to_fp16, epsilon = input_331_epsilon_0_to_fp16, gamma = input_331_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3266_cast_fp16)[name = string("input_331_cast_fp16")];
            string input_333_pad_type_0 = const()[name = string("input_333_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_333_strides_0 = const()[name = string("input_333_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_333_pad_0 = const()[name = string("input_333_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_333_dilations_0 = const()[name = string("input_333_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_333_groups_0 = const()[name = string("input_333_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_12_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231752704))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234898496))))[name = string("layers_12_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_333_cast_fp16 = conv(dilations = input_333_dilations_0, groups = input_333_groups_0, pad = input_333_pad_0, pad_type = input_333_pad_type_0, strides = input_333_strides_0, weight = layers_12_feed_forward1_linear1_weight_to_fp16_palettized, x = input_331_cast_fp16)[name = string("input_333_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_335_cast_fp16 = silu(x = input_333_cast_fp16)[name = string("input_335_cast_fp16")];
            string var_3283_pad_type_0 = const()[name = string("op_3283_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3283_strides_0 = const()[name = string("op_3283_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3283_pad_0 = const()[name = string("op_3283_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3283_dilations_0 = const()[name = string("op_3283_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3283_groups_0 = const()[name = string("op_3283_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_3284_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234931328))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238077120))))[name = string("op_3284_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3284_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_3283_dilations_0, groups = var_3283_groups_0, pad = var_3283_pad_0, pad_type = var_3283_pad_type_0, strides = var_3283_strides_0, weight = op_3284_weight_0_to_fp16_palettized, x = input_335_cast_fp16)[name = string("op_3284_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_269_cast_fp16 = add(x = x_267_cast_fp16, y = var_3284_cast_fp16)[name = string("x_269_cast_fp16")];
            tensor<int32, [1]> var_3300_axes_0 = const()[name = string("op_3300_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_12_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_12_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3300_cast_fp16 = layer_norm(axes = var_3300_axes_0, epsilon = layers_12_norm_self_att_eps_scaled_to_fp16, x = x_269_cast_fp16)[name = string("op_3300_cast_fp16")];
            tensor<fp16, [1024]> x_271_gamma_0_to_fp16 = const()[name = string("x_271_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238085376)))];
            tensor<fp16, [1024]> x_271_beta_0_to_fp16 = const()[name = string("x_271_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238087488)))];
            fp16 x_271_epsilon_0_to_fp16 = const()[name = string("x_271_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_271_cast_fp16 = batch_norm(beta = x_271_beta_0_to_fp16, epsilon = x_271_epsilon_0_to_fp16, gamma = x_271_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3300_cast_fp16)[name = string("x_271_cast_fp16")];
            string q_25_pad_type_0 = const()[name = string("q_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_25_strides_0 = const()[name = string("q_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_25_pad_0 = const()[name = string("q_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_25_dilations_0 = const()[name = string("q_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_25_groups_0 = const()[name = string("q_25_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238089600))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238876096))))[name = string("layers_12_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_25_cast_fp16 = conv(dilations = q_25_dilations_0, groups = q_25_groups_0, pad = q_25_pad_0, pad_type = q_25_pad_type_0, strides = q_25_strides_0, weight = layers_12_self_attn_linear_q_weight_to_fp16_palettized, x = x_271_cast_fp16)[name = string("q_25_cast_fp16")];
            string k_25_pad_type_0 = const()[name = string("k_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_25_strides_0 = const()[name = string("k_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_25_pad_0 = const()[name = string("k_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_25_dilations_0 = const()[name = string("k_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_25_groups_0 = const()[name = string("k_25_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238884352))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239670848))))[name = string("layers_12_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_25_cast_fp16 = conv(dilations = k_25_dilations_0, groups = k_25_groups_0, pad = k_25_pad_0, pad_type = k_25_pad_type_0, strides = k_25_strides_0, weight = layers_12_self_attn_linear_k_weight_to_fp16_palettized, x = x_271_cast_fp16)[name = string("k_25_cast_fp16")];
            string v_25_pad_type_0 = const()[name = string("v_25_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_25_strides_0 = const()[name = string("v_25_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_25_pad_0 = const()[name = string("v_25_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_25_dilations_0 = const()[name = string("v_25_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_25_groups_0 = const()[name = string("v_25_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239679104))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240465600))))[name = string("layers_12_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_25_cast_fp16 = conv(dilations = v_25_dilations_0, groups = v_25_groups_0, pad = v_25_pad_0, pad_type = v_25_pad_type_0, strides = v_25_strides_0, weight = layers_12_self_attn_linear_v_weight_to_fp16_palettized, x = x_271_cast_fp16)[name = string("v_25_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_25_to_fp16 = const()[name = string("bv_all_25_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240473856)))];
            tensor<fp16, [1, 1024, 1, 188]> var_3332_cast_fp16 = add(x = q_25_cast_fp16, y = bv_all_25_to_fp16)[name = string("op_3332_cast_fp16")];
            tensor<int32, [3]> var_3333 = const()[name = string("op_3333"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_25_cast_fp16 = reshape(shape = var_3333, x = var_3332_cast_fp16)[name = string("qb_25_cast_fp16")];
            bool bd_all_49_transpose_x_0 = const()[name = string("bd_all_49_transpose_x_0"), val = bool(false)];
            bool bd_all_49_transpose_y_0 = const()[name = string("bd_all_49_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_12_self_attn_pos_proj_to_fp16 = const()[name = string("layers_12_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240475968)))];
            tensor<fp16, [8, 375, 188]> bd_all_49_cast_fp16 = matmul(transpose_x = bd_all_49_transpose_x_0, transpose_y = bd_all_49_transpose_y_0, x = layers_12_self_attn_pos_proj_to_fp16, y = qb_25_cast_fp16)[name = string("bd_all_49_cast_fp16")];
            tensor<int32, [3]> x_273_perm_0 = const()[name = string("x_273_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_275_pad_0 = const()[name = string("x_275_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_275_mode_0 = const()[name = string("x_275_mode_0"), val = string("constant")];
            fp16 const_58_to_fp16 = const()[name = string("const_58_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_273_cast_fp16 = transpose(perm = x_273_perm_0, x = bd_all_49_cast_fp16)[name = string("transpose_71")];
            tensor<fp16, [8, 188, 376]> x_275_cast_fp16 = pad(constant_val = const_58_to_fp16, mode = x_275_mode_0, pad = x_275_pad_0, x = x_273_cast_fp16)[name = string("x_275_cast_fp16")];
            tensor<int32, [3]> var_3340 = const()[name = string("op_3340"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_277_cast_fp16 = reshape(shape = var_3340, x = x_275_cast_fp16)[name = string("x_277_cast_fp16")];
            tensor<int32, [3]> var_3343_begin_0 = const()[name = string("op_3343_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_3343_end_0 = const()[name = string("op_3343_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_3343_end_mask_0 = const()[name = string("op_3343_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_3343_cast_fp16 = slice_by_index(begin = var_3343_begin_0, end = var_3343_end_0, end_mask = var_3343_end_mask_0, x = x_277_cast_fp16)[name = string("op_3343_cast_fp16")];
            tensor<int32, [3]> var_3344 = const()[name = string("op_3344"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_279_cast_fp16 = reshape(shape = var_3344, x = var_3343_cast_fp16)[name = string("x_279_cast_fp16")];
            tensor<int32, [3]> bd_all_51_begin_0 = const()[name = string("bd_all_51_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_51_end_0 = const()[name = string("bd_all_51_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_51_end_mask_0 = const()[name = string("bd_all_51_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_51_cast_fp16 = slice_by_index(begin = bd_all_51_begin_0, end = bd_all_51_end_0, end_mask = bd_all_51_end_mask_0, x = x_279_cast_fp16)[name = string("bd_all_51_cast_fp16")];
            tensor<int32, [4]> var_3349 = const()[name = string("op_3349"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_3350_cast_fp16 = reshape(shape = var_3349, x = q_25_cast_fp16)[name = string("op_3350_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_3352_to_fp16 = const()[name = string("op_3352_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241244032)))];
            tensor<fp16, [8, 128, 1, 188]> var_3353_cast_fp16 = add(x = var_3350_cast_fp16, y = var_3352_to_fp16)[name = string("op_3353_cast_fp16")];
            tensor<int32, [4]> var_3354 = const()[name = string("op_3354"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_25_cast_fp16 = reshape(shape = var_3354, x = k_25_cast_fp16)[name = string("kh_25_cast_fp16")];
            tensor<int32, [4]> var_3356 = const()[name = string("op_3356"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_25_cast_fp16 = reshape(shape = var_3356, x = v_25_cast_fp16)[name = string("vh_25_cast_fp16")];
            tensor<int32, [4]> var_3358 = const()[name = string("op_3358"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_25_equation_0 = const()[name = string("ac_25_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_3359_cast_fp16 = transpose(perm = var_3358, x = kh_25_cast_fp16)[name = string("transpose_70")];
            tensor<fp16, [8, 188, 1, 188]> ac_25_cast_fp16 = einsum(equation = ac_25_equation_0, values = (var_3359_cast_fp16, var_3353_cast_fp16))[name = string("ac_25_cast_fp16")];
            tensor<int32, [3]> var_3362_perm_0 = const()[name = string("op_3362_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_3363_axes_0 = const()[name = string("op_3363_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_3362_cast_fp16 = transpose(perm = var_3362_perm_0, x = bd_all_51_cast_fp16)[name = string("transpose_69")];
            tensor<fp16, [8, 188, 1, 188]> var_3363_cast_fp16 = expand_dims(axes = var_3363_axes_0, x = var_3362_cast_fp16)[name = string("op_3363_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_3364_cast_fp16 = add(x = ac_25_cast_fp16, y = var_3363_cast_fp16)[name = string("op_3364_cast_fp16")];
            fp16 var_3365_to_fp16 = const()[name = string("op_3365_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_49_cast_fp16 = mul(x = var_3364_cast_fp16, y = var_3365_to_fp16)[name = string("scores_49_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_51_cast_fp16 = add(x = scores_49_cast_fp16, y = key_bias)[name = string("scores_51_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_3368_cast_fp16 = softmax(axis = var_3239, x = scores_51_cast_fp16)[name = string("op_3368_cast_fp16")];
            tensor<int32, [4]> transpose_60_perm_0 = const()[name = string("transpose_60_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_24_perm_0 = const()[name = string("transpose_24_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_124 = const()[name = string("concat_124"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_24_cast_fp16 = transpose(perm = transpose_24_perm_0, x = var_3368_cast_fp16)[name = string("transpose_68")];
            tensor<fp16, [8, 188, 188]> reshape_36_cast_fp16 = reshape(shape = concat_124, x = transpose_24_cast_fp16)[name = string("reshape_36_cast_fp16")];
            tensor<int32, [3]> concat_125 = const()[name = string("concat_125"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_60_cast_fp16 = transpose(perm = transpose_60_perm_0, x = vh_25_cast_fp16)[name = string("transpose_67")];
            tensor<fp16, [8, 188, 128]> reshape_37_cast_fp16 = reshape(shape = concat_125, x = transpose_60_cast_fp16)[name = string("reshape_37_cast_fp16")];
            bool matmul_12_transpose_x_0 = const()[name = string("matmul_12_transpose_x_0"), val = bool(false)];
            bool matmul_12_transpose_y_0 = const()[name = string("matmul_12_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_12_cast_fp16 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = reshape_36_cast_fp16, y = reshape_37_cast_fp16)[name = string("matmul_12_cast_fp16")];
            tensor<int32, [4]> concat_129 = const()[name = string("concat_129"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_38_cast_fp16 = reshape(shape = concat_129, x = matmul_12_cast_fp16)[name = string("reshape_38_cast_fp16")];
            tensor<int32, [4]> ctx_25_perm_0 = const()[name = string("ctx_25_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_3373 = const()[name = string("op_3373"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_25_cast_fp16 = transpose(perm = ctx_25_perm_0, x = reshape_38_cast_fp16)[name = string("transpose_66")];
            tensor<fp16, [1, 1024, 1, 188]> input_337_cast_fp16 = reshape(shape = var_3373, x = ctx_25_cast_fp16)[name = string("input_337_cast_fp16")];
            string var_3380_pad_type_0 = const()[name = string("op_3380_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3380_strides_0 = const()[name = string("op_3380_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3380_pad_0 = const()[name = string("op_3380_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3380_dilations_0 = const()[name = string("op_3380_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3380_groups_0 = const()[name = string("op_3380_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241246144))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242032640))))[name = string("layers_12_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3380_cast_fp16 = conv(dilations = var_3380_dilations_0, groups = var_3380_groups_0, pad = var_3380_pad_0, pad_type = var_3380_pad_type_0, strides = var_3380_strides_0, weight = layers_12_self_attn_linear_out_weight_to_fp16_palettized, x = input_337_cast_fp16)[name = string("op_3380_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_281_cast_fp16 = add(x = x_269_cast_fp16, y = var_3380_cast_fp16)[name = string("x_281_cast_fp16")];
            tensor<int32, [1]> var_3396_axes_0 = const()[name = string("op_3396_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_12_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_12_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3396_cast_fp16 = layer_norm(axes = var_3396_axes_0, epsilon = layers_12_norm_conv_eps_scaled_to_fp16, x = x_281_cast_fp16)[name = string("op_3396_cast_fp16")];
            tensor<fp16, [1024]> input_339_gamma_0_to_fp16 = const()[name = string("input_339_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242040896)))];
            tensor<fp16, [1024]> input_339_beta_0_to_fp16 = const()[name = string("input_339_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242043008)))];
            fp16 input_339_epsilon_0_to_fp16 = const()[name = string("input_339_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_339_cast_fp16 = batch_norm(beta = input_339_beta_0_to_fp16, epsilon = input_339_epsilon_0_to_fp16, gamma = input_339_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3396_cast_fp16)[name = string("input_339_cast_fp16")];
            string input_341_pad_type_0 = const()[name = string("input_341_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_341_strides_0 = const()[name = string("input_341_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_341_pad_0 = const()[name = string("input_341_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_341_dilations_0 = const()[name = string("input_341_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_341_groups_0 = const()[name = string("input_341_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_12_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242045120))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243618048))))[name = string("layers_12_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_341_cast_fp16 = conv(dilations = input_341_dilations_0, groups = input_341_groups_0, pad = input_341_pad_0, pad_type = input_341_pad_type_0, strides = input_341_strides_0, weight = layers_12_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_339_cast_fp16)[name = string("input_341_cast_fp16")];
            int32 x_283_split_num_splits_0 = const()[name = string("x_283_split_num_splits_0"), val = int32(2)];
            int32 x_283_split_axis_0 = const()[name = string("x_283_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_283_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_341_cast_fp16)[name = string("x_283_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_283_split_1_sigmoid_cast_fp16 = sigmoid(x = x_283_split_cast_fp16_1)[name = string("x_283_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_283_cast_fp16 = mul(x = x_283_split_cast_fp16_0, y = x_283_split_1_sigmoid_cast_fp16)[name = string("x_283_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_343_cast_fp16 = mul(x = x_283_cast_fp16, y = pad_mask)[name = string("input_343_cast_fp16")];
            string input_345_pad_type_0 = const()[name = string("input_345_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_345_pad_0 = const()[name = string("input_345_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_345_groups_0 = const()[name = string("input_345_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_345_strides_0 = const()[name = string("input_345_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_345_dilations_0 = const()[name = string("input_345_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_127_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243634496))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243641472))))[name = string("const_127_to_fp16_palettized")];
            tensor<fp16, [1024]> const_128_to_fp16 = const()[name = string("const_128_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243649728)))];
            tensor<fp16, [1, 1024, 1, 188]> input_347_cast_fp16 = conv(bias = const_128_to_fp16, dilations = input_345_dilations_0, groups = input_345_groups_0, pad = input_345_pad_0, pad_type = input_345_pad_type_0, strides = input_345_strides_0, weight = const_127_to_fp16_palettized, x = input_343_cast_fp16)[name = string("input_347_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_349_cast_fp16 = silu(x = input_347_cast_fp16)[name = string("input_349_cast_fp16")];
            string var_3428_pad_type_0 = const()[name = string("op_3428_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3428_strides_0 = const()[name = string("op_3428_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3428_pad_0 = const()[name = string("op_3428_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3428_dilations_0 = const()[name = string("op_3428_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3428_groups_0 = const()[name = string("op_3428_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_12_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243651840))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244438336))))[name = string("layers_12_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3428_cast_fp16 = conv(dilations = var_3428_dilations_0, groups = var_3428_groups_0, pad = var_3428_pad_0, pad_type = var_3428_pad_type_0, strides = var_3428_strides_0, weight = layers_12_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_349_cast_fp16)[name = string("op_3428_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_285_cast_fp16 = add(x = x_281_cast_fp16, y = var_3428_cast_fp16)[name = string("x_285_cast_fp16")];
            tensor<int32, [1]> var_3444_axes_0 = const()[name = string("op_3444_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_12_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_12_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3444_cast_fp16 = layer_norm(axes = var_3444_axes_0, epsilon = layers_12_norm_feed_forward2_eps_scaled_to_fp16, x = x_285_cast_fp16)[name = string("op_3444_cast_fp16")];
            tensor<fp16, [1024]> input_351_gamma_0_to_fp16 = const()[name = string("input_351_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244446592)))];
            tensor<fp16, [1024]> input_351_beta_0_to_fp16 = const()[name = string("input_351_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244448704)))];
            fp16 input_351_epsilon_0_to_fp16 = const()[name = string("input_351_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_351_cast_fp16 = batch_norm(beta = input_351_beta_0_to_fp16, epsilon = input_351_epsilon_0_to_fp16, gamma = input_351_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3444_cast_fp16)[name = string("input_351_cast_fp16")];
            string input_353_pad_type_0 = const()[name = string("input_353_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_353_strides_0 = const()[name = string("input_353_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_353_pad_0 = const()[name = string("input_353_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_353_dilations_0 = const()[name = string("input_353_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_353_groups_0 = const()[name = string("input_353_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_12_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244450816))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247596608))))[name = string("layers_12_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_353_cast_fp16 = conv(dilations = input_353_dilations_0, groups = input_353_groups_0, pad = input_353_pad_0, pad_type = input_353_pad_type_0, strides = input_353_strides_0, weight = layers_12_feed_forward2_linear1_weight_to_fp16_palettized, x = input_351_cast_fp16)[name = string("input_353_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_355_cast_fp16 = silu(x = input_353_cast_fp16)[name = string("input_355_cast_fp16")];
            string var_3461_pad_type_0 = const()[name = string("op_3461_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3461_strides_0 = const()[name = string("op_3461_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3461_pad_0 = const()[name = string("op_3461_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3461_dilations_0 = const()[name = string("op_3461_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3461_groups_0 = const()[name = string("op_3461_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_3462_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247629440))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250775232))))[name = string("op_3462_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3462_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_3461_dilations_0, groups = var_3461_groups_0, pad = var_3461_pad_0, pad_type = var_3461_pad_type_0, strides = var_3461_strides_0, weight = op_3462_weight_0_to_fp16_palettized, x = input_355_cast_fp16)[name = string("op_3462_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_287_cast_fp16 = add(x = x_285_cast_fp16, y = var_3462_cast_fp16)[name = string("x_287_cast_fp16")];
            tensor<int32, [1]> var_3478_axes_0 = const()[name = string("op_3478_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_12_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_12_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3478_cast_fp16 = layer_norm(axes = var_3478_axes_0, epsilon = layers_12_norm_out_eps_scaled_to_fp16, x = x_287_cast_fp16)[name = string("op_3478_cast_fp16")];
            tensor<fp16, [1024]> x_289_gamma_0_to_fp16 = const()[name = string("x_289_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250783488)))];
            tensor<fp16, [1024]> x_289_beta_0_to_fp16 = const()[name = string("x_289_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250785600)))];
            fp16 x_289_epsilon_0_to_fp16 = const()[name = string("x_289_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_289_cast_fp16 = batch_norm(beta = x_289_beta_0_to_fp16, epsilon = x_289_epsilon_0_to_fp16, gamma = x_289_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3478_cast_fp16)[name = string("x_289_cast_fp16")];
            int32 var_3497 = const()[name = string("op_3497"), val = int32(1)];
            tensor<int32, [1]> var_3524_axes_0 = const()[name = string("op_3524_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_13_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_13_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3524_cast_fp16 = layer_norm(axes = var_3524_axes_0, epsilon = layers_13_norm_feed_forward1_eps_scaled_to_fp16, x = x_289_cast_fp16)[name = string("op_3524_cast_fp16")];
            tensor<fp16, [1024]> input_357_gamma_0_to_fp16 = const()[name = string("input_357_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250787712)))];
            tensor<fp16, [1024]> input_357_beta_0_to_fp16 = const()[name = string("input_357_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250789824)))];
            fp16 input_357_epsilon_0_to_fp16 = const()[name = string("input_357_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_357_cast_fp16 = batch_norm(beta = input_357_beta_0_to_fp16, epsilon = input_357_epsilon_0_to_fp16, gamma = input_357_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3524_cast_fp16)[name = string("input_357_cast_fp16")];
            string input_359_pad_type_0 = const()[name = string("input_359_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_359_strides_0 = const()[name = string("input_359_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_359_pad_0 = const()[name = string("input_359_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_359_dilations_0 = const()[name = string("input_359_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_359_groups_0 = const()[name = string("input_359_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_13_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(250791936))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253937728))))[name = string("layers_13_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_359_cast_fp16 = conv(dilations = input_359_dilations_0, groups = input_359_groups_0, pad = input_359_pad_0, pad_type = input_359_pad_type_0, strides = input_359_strides_0, weight = layers_13_feed_forward1_linear1_weight_to_fp16_palettized, x = input_357_cast_fp16)[name = string("input_359_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_361_cast_fp16 = silu(x = input_359_cast_fp16)[name = string("input_361_cast_fp16")];
            string var_3541_pad_type_0 = const()[name = string("op_3541_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3541_strides_0 = const()[name = string("op_3541_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3541_pad_0 = const()[name = string("op_3541_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3541_dilations_0 = const()[name = string("op_3541_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3541_groups_0 = const()[name = string("op_3541_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_3542_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253970560))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257116352))))[name = string("op_3542_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3542_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_3541_dilations_0, groups = var_3541_groups_0, pad = var_3541_pad_0, pad_type = var_3541_pad_type_0, strides = var_3541_strides_0, weight = op_3542_weight_0_to_fp16_palettized, x = input_361_cast_fp16)[name = string("op_3542_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_291_cast_fp16 = add(x = x_289_cast_fp16, y = var_3542_cast_fp16)[name = string("x_291_cast_fp16")];
            tensor<int32, [1]> var_3558_axes_0 = const()[name = string("op_3558_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_13_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_13_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3558_cast_fp16 = layer_norm(axes = var_3558_axes_0, epsilon = layers_13_norm_self_att_eps_scaled_to_fp16, x = x_291_cast_fp16)[name = string("op_3558_cast_fp16")];
            tensor<fp16, [1024]> x_293_gamma_0_to_fp16 = const()[name = string("x_293_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257124608)))];
            tensor<fp16, [1024]> x_293_beta_0_to_fp16 = const()[name = string("x_293_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257126720)))];
            fp16 x_293_epsilon_0_to_fp16 = const()[name = string("x_293_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_293_cast_fp16 = batch_norm(beta = x_293_beta_0_to_fp16, epsilon = x_293_epsilon_0_to_fp16, gamma = x_293_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3558_cast_fp16)[name = string("x_293_cast_fp16")];
            string q_27_pad_type_0 = const()[name = string("q_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_27_strides_0 = const()[name = string("q_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_27_pad_0 = const()[name = string("q_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_27_dilations_0 = const()[name = string("q_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_27_groups_0 = const()[name = string("q_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257128832))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257915328))))[name = string("layers_13_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_27_cast_fp16 = conv(dilations = q_27_dilations_0, groups = q_27_groups_0, pad = q_27_pad_0, pad_type = q_27_pad_type_0, strides = q_27_strides_0, weight = layers_13_self_attn_linear_q_weight_to_fp16_palettized, x = x_293_cast_fp16)[name = string("q_27_cast_fp16")];
            string k_27_pad_type_0 = const()[name = string("k_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_27_strides_0 = const()[name = string("k_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_27_pad_0 = const()[name = string("k_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_27_dilations_0 = const()[name = string("k_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_27_groups_0 = const()[name = string("k_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257923584))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258710080))))[name = string("layers_13_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_27_cast_fp16 = conv(dilations = k_27_dilations_0, groups = k_27_groups_0, pad = k_27_pad_0, pad_type = k_27_pad_type_0, strides = k_27_strides_0, weight = layers_13_self_attn_linear_k_weight_to_fp16_palettized, x = x_293_cast_fp16)[name = string("k_27_cast_fp16")];
            string v_27_pad_type_0 = const()[name = string("v_27_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_27_strides_0 = const()[name = string("v_27_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_27_pad_0 = const()[name = string("v_27_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_27_dilations_0 = const()[name = string("v_27_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_27_groups_0 = const()[name = string("v_27_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258718336))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259504832))))[name = string("layers_13_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_27_cast_fp16 = conv(dilations = v_27_dilations_0, groups = v_27_groups_0, pad = v_27_pad_0, pad_type = v_27_pad_type_0, strides = v_27_strides_0, weight = layers_13_self_attn_linear_v_weight_to_fp16_palettized, x = x_293_cast_fp16)[name = string("v_27_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_27_to_fp16 = const()[name = string("bv_all_27_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259513088)))];
            tensor<fp16, [1, 1024, 1, 188]> var_3590_cast_fp16 = add(x = q_27_cast_fp16, y = bv_all_27_to_fp16)[name = string("op_3590_cast_fp16")];
            tensor<int32, [3]> var_3591 = const()[name = string("op_3591"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_27_cast_fp16 = reshape(shape = var_3591, x = var_3590_cast_fp16)[name = string("qb_27_cast_fp16")];
            bool bd_all_53_transpose_x_0 = const()[name = string("bd_all_53_transpose_x_0"), val = bool(false)];
            bool bd_all_53_transpose_y_0 = const()[name = string("bd_all_53_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_13_self_attn_pos_proj_to_fp16 = const()[name = string("layers_13_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259515200)))];
            tensor<fp16, [8, 375, 188]> bd_all_53_cast_fp16 = matmul(transpose_x = bd_all_53_transpose_x_0, transpose_y = bd_all_53_transpose_y_0, x = layers_13_self_attn_pos_proj_to_fp16, y = qb_27_cast_fp16)[name = string("bd_all_53_cast_fp16")];
            tensor<int32, [3]> x_295_perm_0 = const()[name = string("x_295_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_297_pad_0 = const()[name = string("x_297_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_297_mode_0 = const()[name = string("x_297_mode_0"), val = string("constant")];
            fp16 const_62_to_fp16 = const()[name = string("const_62_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_295_cast_fp16 = transpose(perm = x_295_perm_0, x = bd_all_53_cast_fp16)[name = string("transpose_65")];
            tensor<fp16, [8, 188, 376]> x_297_cast_fp16 = pad(constant_val = const_62_to_fp16, mode = x_297_mode_0, pad = x_297_pad_0, x = x_295_cast_fp16)[name = string("x_297_cast_fp16")];
            tensor<int32, [3]> var_3598 = const()[name = string("op_3598"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_299_cast_fp16 = reshape(shape = var_3598, x = x_297_cast_fp16)[name = string("x_299_cast_fp16")];
            tensor<int32, [3]> var_3601_begin_0 = const()[name = string("op_3601_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_3601_end_0 = const()[name = string("op_3601_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_3601_end_mask_0 = const()[name = string("op_3601_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_3601_cast_fp16 = slice_by_index(begin = var_3601_begin_0, end = var_3601_end_0, end_mask = var_3601_end_mask_0, x = x_299_cast_fp16)[name = string("op_3601_cast_fp16")];
            tensor<int32, [3]> var_3602 = const()[name = string("op_3602"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_301_cast_fp16 = reshape(shape = var_3602, x = var_3601_cast_fp16)[name = string("x_301_cast_fp16")];
            tensor<int32, [3]> bd_all_55_begin_0 = const()[name = string("bd_all_55_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_55_end_0 = const()[name = string("bd_all_55_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_55_end_mask_0 = const()[name = string("bd_all_55_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_55_cast_fp16 = slice_by_index(begin = bd_all_55_begin_0, end = bd_all_55_end_0, end_mask = bd_all_55_end_mask_0, x = x_301_cast_fp16)[name = string("bd_all_55_cast_fp16")];
            tensor<int32, [4]> var_3607 = const()[name = string("op_3607"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_3608_cast_fp16 = reshape(shape = var_3607, x = q_27_cast_fp16)[name = string("op_3608_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_3610_to_fp16 = const()[name = string("op_3610_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260283264)))];
            tensor<fp16, [8, 128, 1, 188]> var_3611_cast_fp16 = add(x = var_3608_cast_fp16, y = var_3610_to_fp16)[name = string("op_3611_cast_fp16")];
            tensor<int32, [4]> var_3612 = const()[name = string("op_3612"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_27_cast_fp16 = reshape(shape = var_3612, x = k_27_cast_fp16)[name = string("kh_27_cast_fp16")];
            tensor<int32, [4]> var_3614 = const()[name = string("op_3614"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_27_cast_fp16 = reshape(shape = var_3614, x = v_27_cast_fp16)[name = string("vh_27_cast_fp16")];
            tensor<int32, [4]> var_3616 = const()[name = string("op_3616"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_27_equation_0 = const()[name = string("ac_27_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_3617_cast_fp16 = transpose(perm = var_3616, x = kh_27_cast_fp16)[name = string("transpose_64")];
            tensor<fp16, [8, 188, 1, 188]> ac_27_cast_fp16 = einsum(equation = ac_27_equation_0, values = (var_3617_cast_fp16, var_3611_cast_fp16))[name = string("ac_27_cast_fp16")];
            tensor<int32, [3]> var_3620_perm_0 = const()[name = string("op_3620_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_3621_axes_0 = const()[name = string("op_3621_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_3620_cast_fp16 = transpose(perm = var_3620_perm_0, x = bd_all_55_cast_fp16)[name = string("transpose_63")];
            tensor<fp16, [8, 188, 1, 188]> var_3621_cast_fp16 = expand_dims(axes = var_3621_axes_0, x = var_3620_cast_fp16)[name = string("op_3621_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_3622_cast_fp16 = add(x = ac_27_cast_fp16, y = var_3621_cast_fp16)[name = string("op_3622_cast_fp16")];
            fp16 var_3623_to_fp16 = const()[name = string("op_3623_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_53_cast_fp16 = mul(x = var_3622_cast_fp16, y = var_3623_to_fp16)[name = string("scores_53_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_55_cast_fp16 = add(x = scores_53_cast_fp16, y = key_bias)[name = string("scores_55_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_3626_cast_fp16 = softmax(axis = var_3497, x = scores_55_cast_fp16)[name = string("op_3626_cast_fp16")];
            tensor<int32, [4]> transpose_61_perm_0 = const()[name = string("transpose_61_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_26_perm_0 = const()[name = string("transpose_26_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_134 = const()[name = string("concat_134"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_26_cast_fp16 = transpose(perm = transpose_26_perm_0, x = var_3626_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [8, 188, 188]> reshape_39_cast_fp16 = reshape(shape = concat_134, x = transpose_26_cast_fp16)[name = string("reshape_39_cast_fp16")];
            tensor<int32, [3]> concat_135 = const()[name = string("concat_135"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_61_cast_fp16 = transpose(perm = transpose_61_perm_0, x = vh_27_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [8, 188, 128]> reshape_40_cast_fp16 = reshape(shape = concat_135, x = transpose_61_cast_fp16)[name = string("reshape_40_cast_fp16")];
            bool matmul_13_transpose_x_0 = const()[name = string("matmul_13_transpose_x_0"), val = bool(false)];
            bool matmul_13_transpose_y_0 = const()[name = string("matmul_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_13_cast_fp16 = matmul(transpose_x = matmul_13_transpose_x_0, transpose_y = matmul_13_transpose_y_0, x = reshape_39_cast_fp16, y = reshape_40_cast_fp16)[name = string("matmul_13_cast_fp16")];
            tensor<int32, [4]> concat_139 = const()[name = string("concat_139"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_41_cast_fp16 = reshape(shape = concat_139, x = matmul_13_cast_fp16)[name = string("reshape_41_cast_fp16")];
            tensor<int32, [4]> ctx_27_perm_0 = const()[name = string("ctx_27_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_3631 = const()[name = string("op_3631"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_27_cast_fp16 = transpose(perm = ctx_27_perm_0, x = reshape_41_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [1, 1024, 1, 188]> input_363_cast_fp16 = reshape(shape = var_3631, x = ctx_27_cast_fp16)[name = string("input_363_cast_fp16")];
            string var_3638_pad_type_0 = const()[name = string("op_3638_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3638_strides_0 = const()[name = string("op_3638_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3638_pad_0 = const()[name = string("op_3638_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3638_dilations_0 = const()[name = string("op_3638_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3638_groups_0 = const()[name = string("op_3638_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260285376))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261071872))))[name = string("layers_13_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3638_cast_fp16 = conv(dilations = var_3638_dilations_0, groups = var_3638_groups_0, pad = var_3638_pad_0, pad_type = var_3638_pad_type_0, strides = var_3638_strides_0, weight = layers_13_self_attn_linear_out_weight_to_fp16_palettized, x = input_363_cast_fp16)[name = string("op_3638_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_303_cast_fp16 = add(x = x_291_cast_fp16, y = var_3638_cast_fp16)[name = string("x_303_cast_fp16")];
            tensor<int32, [1]> var_3654_axes_0 = const()[name = string("op_3654_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_13_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_13_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3654_cast_fp16 = layer_norm(axes = var_3654_axes_0, epsilon = layers_13_norm_conv_eps_scaled_to_fp16, x = x_303_cast_fp16)[name = string("op_3654_cast_fp16")];
            tensor<fp16, [1024]> input_365_gamma_0_to_fp16 = const()[name = string("input_365_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261080128)))];
            tensor<fp16, [1024]> input_365_beta_0_to_fp16 = const()[name = string("input_365_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261082240)))];
            fp16 input_365_epsilon_0_to_fp16 = const()[name = string("input_365_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_365_cast_fp16 = batch_norm(beta = input_365_beta_0_to_fp16, epsilon = input_365_epsilon_0_to_fp16, gamma = input_365_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3654_cast_fp16)[name = string("input_365_cast_fp16")];
            string input_367_pad_type_0 = const()[name = string("input_367_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_367_strides_0 = const()[name = string("input_367_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_367_pad_0 = const()[name = string("input_367_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_367_dilations_0 = const()[name = string("input_367_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_367_groups_0 = const()[name = string("input_367_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_13_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261084352))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262657280))))[name = string("layers_13_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_367_cast_fp16 = conv(dilations = input_367_dilations_0, groups = input_367_groups_0, pad = input_367_pad_0, pad_type = input_367_pad_type_0, strides = input_367_strides_0, weight = layers_13_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_365_cast_fp16)[name = string("input_367_cast_fp16")];
            int32 x_305_split_num_splits_0 = const()[name = string("x_305_split_num_splits_0"), val = int32(2)];
            int32 x_305_split_axis_0 = const()[name = string("x_305_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_305_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_367_cast_fp16)[name = string("x_305_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_305_split_1_sigmoid_cast_fp16 = sigmoid(x = x_305_split_cast_fp16_1)[name = string("x_305_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_305_cast_fp16 = mul(x = x_305_split_cast_fp16_0, y = x_305_split_1_sigmoid_cast_fp16)[name = string("x_305_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_369_cast_fp16 = mul(x = x_305_cast_fp16, y = pad_mask)[name = string("input_369_cast_fp16")];
            string input_371_pad_type_0 = const()[name = string("input_371_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_371_pad_0 = const()[name = string("input_371_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_371_groups_0 = const()[name = string("input_371_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_371_strides_0 = const()[name = string("input_371_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_371_dilations_0 = const()[name = string("input_371_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_129_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262673728))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262680704))))[name = string("const_129_to_fp16_palettized")];
            tensor<fp16, [1024]> const_130_to_fp16 = const()[name = string("const_130_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262688960)))];
            tensor<fp16, [1, 1024, 1, 188]> input_373_cast_fp16 = conv(bias = const_130_to_fp16, dilations = input_371_dilations_0, groups = input_371_groups_0, pad = input_371_pad_0, pad_type = input_371_pad_type_0, strides = input_371_strides_0, weight = const_129_to_fp16_palettized, x = input_369_cast_fp16)[name = string("input_373_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_375_cast_fp16 = silu(x = input_373_cast_fp16)[name = string("input_375_cast_fp16")];
            string var_3686_pad_type_0 = const()[name = string("op_3686_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3686_strides_0 = const()[name = string("op_3686_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3686_pad_0 = const()[name = string("op_3686_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3686_dilations_0 = const()[name = string("op_3686_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3686_groups_0 = const()[name = string("op_3686_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_13_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262691072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263477568))))[name = string("layers_13_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3686_cast_fp16 = conv(dilations = var_3686_dilations_0, groups = var_3686_groups_0, pad = var_3686_pad_0, pad_type = var_3686_pad_type_0, strides = var_3686_strides_0, weight = layers_13_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_375_cast_fp16)[name = string("op_3686_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_307_cast_fp16 = add(x = x_303_cast_fp16, y = var_3686_cast_fp16)[name = string("x_307_cast_fp16")];
            tensor<int32, [1]> var_3702_axes_0 = const()[name = string("op_3702_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_13_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_13_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3702_cast_fp16 = layer_norm(axes = var_3702_axes_0, epsilon = layers_13_norm_feed_forward2_eps_scaled_to_fp16, x = x_307_cast_fp16)[name = string("op_3702_cast_fp16")];
            tensor<fp16, [1024]> input_377_gamma_0_to_fp16 = const()[name = string("input_377_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263485824)))];
            tensor<fp16, [1024]> input_377_beta_0_to_fp16 = const()[name = string("input_377_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263487936)))];
            fp16 input_377_epsilon_0_to_fp16 = const()[name = string("input_377_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_377_cast_fp16 = batch_norm(beta = input_377_beta_0_to_fp16, epsilon = input_377_epsilon_0_to_fp16, gamma = input_377_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3702_cast_fp16)[name = string("input_377_cast_fp16")];
            string input_379_pad_type_0 = const()[name = string("input_379_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_379_strides_0 = const()[name = string("input_379_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_379_pad_0 = const()[name = string("input_379_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_379_dilations_0 = const()[name = string("input_379_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_379_groups_0 = const()[name = string("input_379_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_13_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263490048))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266635840))))[name = string("layers_13_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_379_cast_fp16 = conv(dilations = input_379_dilations_0, groups = input_379_groups_0, pad = input_379_pad_0, pad_type = input_379_pad_type_0, strides = input_379_strides_0, weight = layers_13_feed_forward2_linear1_weight_to_fp16_palettized, x = input_377_cast_fp16)[name = string("input_379_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_381_cast_fp16 = silu(x = input_379_cast_fp16)[name = string("input_381_cast_fp16")];
            string var_3719_pad_type_0 = const()[name = string("op_3719_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3719_strides_0 = const()[name = string("op_3719_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3719_pad_0 = const()[name = string("op_3719_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3719_dilations_0 = const()[name = string("op_3719_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3719_groups_0 = const()[name = string("op_3719_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_3720_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266668672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269814464))))[name = string("op_3720_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3720_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_3719_dilations_0, groups = var_3719_groups_0, pad = var_3719_pad_0, pad_type = var_3719_pad_type_0, strides = var_3719_strides_0, weight = op_3720_weight_0_to_fp16_palettized, x = input_381_cast_fp16)[name = string("op_3720_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_309_cast_fp16 = add(x = x_307_cast_fp16, y = var_3720_cast_fp16)[name = string("x_309_cast_fp16")];
            tensor<int32, [1]> var_3736_axes_0 = const()[name = string("op_3736_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_13_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_13_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3736_cast_fp16 = layer_norm(axes = var_3736_axes_0, epsilon = layers_13_norm_out_eps_scaled_to_fp16, x = x_309_cast_fp16)[name = string("op_3736_cast_fp16")];
            tensor<fp16, [1024]> x_311_gamma_0_to_fp16 = const()[name = string("x_311_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269822720)))];
            tensor<fp16, [1024]> x_311_beta_0_to_fp16 = const()[name = string("x_311_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269824832)))];
            fp16 x_311_epsilon_0_to_fp16 = const()[name = string("x_311_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_311_cast_fp16 = batch_norm(beta = x_311_beta_0_to_fp16, epsilon = x_311_epsilon_0_to_fp16, gamma = x_311_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3736_cast_fp16)[name = string("x_311_cast_fp16")];
            int32 var_3755 = const()[name = string("op_3755"), val = int32(1)];
            tensor<int32, [1]> var_3782_axes_0 = const()[name = string("op_3782_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_14_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_14_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3782_cast_fp16 = layer_norm(axes = var_3782_axes_0, epsilon = layers_14_norm_feed_forward1_eps_scaled_to_fp16, x = x_311_cast_fp16)[name = string("op_3782_cast_fp16")];
            tensor<fp16, [1024]> input_383_gamma_0_to_fp16 = const()[name = string("input_383_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269826944)))];
            tensor<fp16, [1024]> input_383_beta_0_to_fp16 = const()[name = string("input_383_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269829056)))];
            fp16 input_383_epsilon_0_to_fp16 = const()[name = string("input_383_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_383_cast_fp16 = batch_norm(beta = input_383_beta_0_to_fp16, epsilon = input_383_epsilon_0_to_fp16, gamma = input_383_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3782_cast_fp16)[name = string("input_383_cast_fp16")];
            string input_385_pad_type_0 = const()[name = string("input_385_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_385_strides_0 = const()[name = string("input_385_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_385_pad_0 = const()[name = string("input_385_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_385_dilations_0 = const()[name = string("input_385_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_385_groups_0 = const()[name = string("input_385_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_14_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269831168))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272976960))))[name = string("layers_14_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_385_cast_fp16 = conv(dilations = input_385_dilations_0, groups = input_385_groups_0, pad = input_385_pad_0, pad_type = input_385_pad_type_0, strides = input_385_strides_0, weight = layers_14_feed_forward1_linear1_weight_to_fp16_palettized, x = input_383_cast_fp16)[name = string("input_385_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_387_cast_fp16 = silu(x = input_385_cast_fp16)[name = string("input_387_cast_fp16")];
            string var_3799_pad_type_0 = const()[name = string("op_3799_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3799_strides_0 = const()[name = string("op_3799_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3799_pad_0 = const()[name = string("op_3799_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3799_dilations_0 = const()[name = string("op_3799_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3799_groups_0 = const()[name = string("op_3799_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_3800_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273009792))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276155584))))[name = string("op_3800_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3800_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_3799_dilations_0, groups = var_3799_groups_0, pad = var_3799_pad_0, pad_type = var_3799_pad_type_0, strides = var_3799_strides_0, weight = op_3800_weight_0_to_fp16_palettized, x = input_387_cast_fp16)[name = string("op_3800_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_313_cast_fp16 = add(x = x_311_cast_fp16, y = var_3800_cast_fp16)[name = string("x_313_cast_fp16")];
            tensor<int32, [1]> var_3816_axes_0 = const()[name = string("op_3816_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_14_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_14_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3816_cast_fp16 = layer_norm(axes = var_3816_axes_0, epsilon = layers_14_norm_self_att_eps_scaled_to_fp16, x = x_313_cast_fp16)[name = string("op_3816_cast_fp16")];
            tensor<fp16, [1024]> x_315_gamma_0_to_fp16 = const()[name = string("x_315_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276163840)))];
            tensor<fp16, [1024]> x_315_beta_0_to_fp16 = const()[name = string("x_315_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276165952)))];
            fp16 x_315_epsilon_0_to_fp16 = const()[name = string("x_315_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_315_cast_fp16 = batch_norm(beta = x_315_beta_0_to_fp16, epsilon = x_315_epsilon_0_to_fp16, gamma = x_315_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3816_cast_fp16)[name = string("x_315_cast_fp16")];
            string q_29_pad_type_0 = const()[name = string("q_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_29_strides_0 = const()[name = string("q_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_29_pad_0 = const()[name = string("q_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_29_dilations_0 = const()[name = string("q_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_29_groups_0 = const()[name = string("q_29_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276168064))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276954560))))[name = string("layers_14_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_29_cast_fp16 = conv(dilations = q_29_dilations_0, groups = q_29_groups_0, pad = q_29_pad_0, pad_type = q_29_pad_type_0, strides = q_29_strides_0, weight = layers_14_self_attn_linear_q_weight_to_fp16_palettized, x = x_315_cast_fp16)[name = string("q_29_cast_fp16")];
            string k_29_pad_type_0 = const()[name = string("k_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_29_strides_0 = const()[name = string("k_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_29_pad_0 = const()[name = string("k_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_29_dilations_0 = const()[name = string("k_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_29_groups_0 = const()[name = string("k_29_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276962816))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277749312))))[name = string("layers_14_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_29_cast_fp16 = conv(dilations = k_29_dilations_0, groups = k_29_groups_0, pad = k_29_pad_0, pad_type = k_29_pad_type_0, strides = k_29_strides_0, weight = layers_14_self_attn_linear_k_weight_to_fp16_palettized, x = x_315_cast_fp16)[name = string("k_29_cast_fp16")];
            string v_29_pad_type_0 = const()[name = string("v_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_29_strides_0 = const()[name = string("v_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_29_pad_0 = const()[name = string("v_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_29_dilations_0 = const()[name = string("v_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_29_groups_0 = const()[name = string("v_29_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277757568))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278544064))))[name = string("layers_14_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_29_cast_fp16 = conv(dilations = v_29_dilations_0, groups = v_29_groups_0, pad = v_29_pad_0, pad_type = v_29_pad_type_0, strides = v_29_strides_0, weight = layers_14_self_attn_linear_v_weight_to_fp16_palettized, x = x_315_cast_fp16)[name = string("v_29_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_29_to_fp16 = const()[name = string("bv_all_29_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278552320)))];
            tensor<fp16, [1, 1024, 1, 188]> var_3848_cast_fp16 = add(x = q_29_cast_fp16, y = bv_all_29_to_fp16)[name = string("op_3848_cast_fp16")];
            tensor<int32, [3]> var_3849 = const()[name = string("op_3849"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_29_cast_fp16 = reshape(shape = var_3849, x = var_3848_cast_fp16)[name = string("qb_29_cast_fp16")];
            bool bd_all_57_transpose_x_0 = const()[name = string("bd_all_57_transpose_x_0"), val = bool(false)];
            bool bd_all_57_transpose_y_0 = const()[name = string("bd_all_57_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_14_self_attn_pos_proj_to_fp16 = const()[name = string("layers_14_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278554432)))];
            tensor<fp16, [8, 375, 188]> bd_all_57_cast_fp16 = matmul(transpose_x = bd_all_57_transpose_x_0, transpose_y = bd_all_57_transpose_y_0, x = layers_14_self_attn_pos_proj_to_fp16, y = qb_29_cast_fp16)[name = string("bd_all_57_cast_fp16")];
            tensor<int32, [3]> x_317_perm_0 = const()[name = string("x_317_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_319_pad_0 = const()[name = string("x_319_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_319_mode_0 = const()[name = string("x_319_mode_0"), val = string("constant")];
            fp16 const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_317_cast_fp16 = transpose(perm = x_317_perm_0, x = bd_all_57_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [8, 188, 376]> x_319_cast_fp16 = pad(constant_val = const_66_to_fp16, mode = x_319_mode_0, pad = x_319_pad_0, x = x_317_cast_fp16)[name = string("x_319_cast_fp16")];
            tensor<int32, [3]> var_3856 = const()[name = string("op_3856"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_321_cast_fp16 = reshape(shape = var_3856, x = x_319_cast_fp16)[name = string("x_321_cast_fp16")];
            tensor<int32, [3]> var_3859_begin_0 = const()[name = string("op_3859_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_3859_end_0 = const()[name = string("op_3859_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_3859_end_mask_0 = const()[name = string("op_3859_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_3859_cast_fp16 = slice_by_index(begin = var_3859_begin_0, end = var_3859_end_0, end_mask = var_3859_end_mask_0, x = x_321_cast_fp16)[name = string("op_3859_cast_fp16")];
            tensor<int32, [3]> var_3860 = const()[name = string("op_3860"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_323_cast_fp16 = reshape(shape = var_3860, x = var_3859_cast_fp16)[name = string("x_323_cast_fp16")];
            tensor<int32, [3]> bd_all_59_begin_0 = const()[name = string("bd_all_59_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_59_end_0 = const()[name = string("bd_all_59_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_59_end_mask_0 = const()[name = string("bd_all_59_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_59_cast_fp16 = slice_by_index(begin = bd_all_59_begin_0, end = bd_all_59_end_0, end_mask = bd_all_59_end_mask_0, x = x_323_cast_fp16)[name = string("bd_all_59_cast_fp16")];
            tensor<int32, [4]> var_3865 = const()[name = string("op_3865"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_3866_cast_fp16 = reshape(shape = var_3865, x = q_29_cast_fp16)[name = string("op_3866_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_3868_to_fp16 = const()[name = string("op_3868_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279322496)))];
            tensor<fp16, [8, 128, 1, 188]> var_3869_cast_fp16 = add(x = var_3866_cast_fp16, y = var_3868_to_fp16)[name = string("op_3869_cast_fp16")];
            tensor<int32, [4]> var_3870 = const()[name = string("op_3870"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_29_cast_fp16 = reshape(shape = var_3870, x = k_29_cast_fp16)[name = string("kh_29_cast_fp16")];
            tensor<int32, [4]> var_3872 = const()[name = string("op_3872"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_29_cast_fp16 = reshape(shape = var_3872, x = v_29_cast_fp16)[name = string("vh_29_cast_fp16")];
            tensor<int32, [4]> var_3874 = const()[name = string("op_3874"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_29_equation_0 = const()[name = string("ac_29_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_3875_cast_fp16 = transpose(perm = var_3874, x = kh_29_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [8, 188, 1, 188]> ac_29_cast_fp16 = einsum(equation = ac_29_equation_0, values = (var_3875_cast_fp16, var_3869_cast_fp16))[name = string("ac_29_cast_fp16")];
            tensor<int32, [3]> var_3878_perm_0 = const()[name = string("op_3878_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_3879_axes_0 = const()[name = string("op_3879_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_3878_cast_fp16 = transpose(perm = var_3878_perm_0, x = bd_all_59_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [8, 188, 1, 188]> var_3879_cast_fp16 = expand_dims(axes = var_3879_axes_0, x = var_3878_cast_fp16)[name = string("op_3879_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_3880_cast_fp16 = add(x = ac_29_cast_fp16, y = var_3879_cast_fp16)[name = string("op_3880_cast_fp16")];
            fp16 var_3881_to_fp16 = const()[name = string("op_3881_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_57_cast_fp16 = mul(x = var_3880_cast_fp16, y = var_3881_to_fp16)[name = string("scores_57_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_59_cast_fp16 = add(x = scores_57_cast_fp16, y = key_bias)[name = string("scores_59_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_3884_cast_fp16 = softmax(axis = var_3755, x = scores_59_cast_fp16)[name = string("op_3884_cast_fp16")];
            tensor<int32, [4]> transpose_62_perm_0 = const()[name = string("transpose_62_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_28_perm_0 = const()[name = string("transpose_28_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_144 = const()[name = string("concat_144"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_28_cast_fp16 = transpose(perm = transpose_28_perm_0, x = var_3884_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [8, 188, 188]> reshape_42_cast_fp16 = reshape(shape = concat_144, x = transpose_28_cast_fp16)[name = string("reshape_42_cast_fp16")];
            tensor<int32, [3]> concat_145 = const()[name = string("concat_145"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_62_cast_fp16 = transpose(perm = transpose_62_perm_0, x = vh_29_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [8, 188, 128]> reshape_43_cast_fp16 = reshape(shape = concat_145, x = transpose_62_cast_fp16)[name = string("reshape_43_cast_fp16")];
            bool matmul_14_transpose_x_0 = const()[name = string("matmul_14_transpose_x_0"), val = bool(false)];
            bool matmul_14_transpose_y_0 = const()[name = string("matmul_14_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_14_cast_fp16 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = reshape_42_cast_fp16, y = reshape_43_cast_fp16)[name = string("matmul_14_cast_fp16")];
            tensor<int32, [4]> concat_149 = const()[name = string("concat_149"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_44_cast_fp16 = reshape(shape = concat_149, x = matmul_14_cast_fp16)[name = string("reshape_44_cast_fp16")];
            tensor<int32, [4]> ctx_29_perm_0 = const()[name = string("ctx_29_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_3889 = const()[name = string("op_3889"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_29_cast_fp16 = transpose(perm = ctx_29_perm_0, x = reshape_44_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [1, 1024, 1, 188]> input_389_cast_fp16 = reshape(shape = var_3889, x = ctx_29_cast_fp16)[name = string("input_389_cast_fp16")];
            string var_3896_pad_type_0 = const()[name = string("op_3896_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3896_strides_0 = const()[name = string("op_3896_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3896_pad_0 = const()[name = string("op_3896_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3896_dilations_0 = const()[name = string("op_3896_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3896_groups_0 = const()[name = string("op_3896_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279324608))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280111104))))[name = string("layers_14_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3896_cast_fp16 = conv(dilations = var_3896_dilations_0, groups = var_3896_groups_0, pad = var_3896_pad_0, pad_type = var_3896_pad_type_0, strides = var_3896_strides_0, weight = layers_14_self_attn_linear_out_weight_to_fp16_palettized, x = input_389_cast_fp16)[name = string("op_3896_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_325_cast_fp16 = add(x = x_313_cast_fp16, y = var_3896_cast_fp16)[name = string("x_325_cast_fp16")];
            tensor<int32, [1]> var_3912_axes_0 = const()[name = string("op_3912_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_14_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_14_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3912_cast_fp16 = layer_norm(axes = var_3912_axes_0, epsilon = layers_14_norm_conv_eps_scaled_to_fp16, x = x_325_cast_fp16)[name = string("op_3912_cast_fp16")];
            tensor<fp16, [1024]> input_391_gamma_0_to_fp16 = const()[name = string("input_391_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280119360)))];
            tensor<fp16, [1024]> input_391_beta_0_to_fp16 = const()[name = string("input_391_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280121472)))];
            fp16 input_391_epsilon_0_to_fp16 = const()[name = string("input_391_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_391_cast_fp16 = batch_norm(beta = input_391_beta_0_to_fp16, epsilon = input_391_epsilon_0_to_fp16, gamma = input_391_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3912_cast_fp16)[name = string("input_391_cast_fp16")];
            string input_393_pad_type_0 = const()[name = string("input_393_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_393_strides_0 = const()[name = string("input_393_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_393_pad_0 = const()[name = string("input_393_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_393_dilations_0 = const()[name = string("input_393_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_393_groups_0 = const()[name = string("input_393_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_14_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(280123584))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281696512))))[name = string("layers_14_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_393_cast_fp16 = conv(dilations = input_393_dilations_0, groups = input_393_groups_0, pad = input_393_pad_0, pad_type = input_393_pad_type_0, strides = input_393_strides_0, weight = layers_14_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_391_cast_fp16)[name = string("input_393_cast_fp16")];
            int32 x_327_split_num_splits_0 = const()[name = string("x_327_split_num_splits_0"), val = int32(2)];
            int32 x_327_split_axis_0 = const()[name = string("x_327_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_327_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_393_cast_fp16)[name = string("x_327_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_327_split_1_sigmoid_cast_fp16 = sigmoid(x = x_327_split_cast_fp16_1)[name = string("x_327_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_327_cast_fp16 = mul(x = x_327_split_cast_fp16_0, y = x_327_split_1_sigmoid_cast_fp16)[name = string("x_327_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_395_cast_fp16 = mul(x = x_327_cast_fp16, y = pad_mask)[name = string("input_395_cast_fp16")];
            string input_397_pad_type_0 = const()[name = string("input_397_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_397_pad_0 = const()[name = string("input_397_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_397_groups_0 = const()[name = string("input_397_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_397_strides_0 = const()[name = string("input_397_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_397_dilations_0 = const()[name = string("input_397_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_131_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281712960))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281719936))))[name = string("const_131_to_fp16_palettized")];
            tensor<fp16, [1024]> const_132_to_fp16 = const()[name = string("const_132_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281728192)))];
            tensor<fp16, [1, 1024, 1, 188]> input_399_cast_fp16 = conv(bias = const_132_to_fp16, dilations = input_397_dilations_0, groups = input_397_groups_0, pad = input_397_pad_0, pad_type = input_397_pad_type_0, strides = input_397_strides_0, weight = const_131_to_fp16_palettized, x = input_395_cast_fp16)[name = string("input_399_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_401_cast_fp16 = silu(x = input_399_cast_fp16)[name = string("input_401_cast_fp16")];
            string var_3944_pad_type_0 = const()[name = string("op_3944_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3944_strides_0 = const()[name = string("op_3944_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3944_pad_0 = const()[name = string("op_3944_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3944_dilations_0 = const()[name = string("op_3944_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3944_groups_0 = const()[name = string("op_3944_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_14_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281730304))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282516800))))[name = string("layers_14_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3944_cast_fp16 = conv(dilations = var_3944_dilations_0, groups = var_3944_groups_0, pad = var_3944_pad_0, pad_type = var_3944_pad_type_0, strides = var_3944_strides_0, weight = layers_14_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_401_cast_fp16)[name = string("op_3944_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_329_cast_fp16 = add(x = x_325_cast_fp16, y = var_3944_cast_fp16)[name = string("x_329_cast_fp16")];
            tensor<int32, [1]> var_3960_axes_0 = const()[name = string("op_3960_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_14_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_14_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3960_cast_fp16 = layer_norm(axes = var_3960_axes_0, epsilon = layers_14_norm_feed_forward2_eps_scaled_to_fp16, x = x_329_cast_fp16)[name = string("op_3960_cast_fp16")];
            tensor<fp16, [1024]> input_403_gamma_0_to_fp16 = const()[name = string("input_403_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282525056)))];
            tensor<fp16, [1024]> input_403_beta_0_to_fp16 = const()[name = string("input_403_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282527168)))];
            fp16 input_403_epsilon_0_to_fp16 = const()[name = string("input_403_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_403_cast_fp16 = batch_norm(beta = input_403_beta_0_to_fp16, epsilon = input_403_epsilon_0_to_fp16, gamma = input_403_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3960_cast_fp16)[name = string("input_403_cast_fp16")];
            string input_405_pad_type_0 = const()[name = string("input_405_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_405_strides_0 = const()[name = string("input_405_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_405_pad_0 = const()[name = string("input_405_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_405_dilations_0 = const()[name = string("input_405_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_405_groups_0 = const()[name = string("input_405_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_14_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(282529280))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285675072))))[name = string("layers_14_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_405_cast_fp16 = conv(dilations = input_405_dilations_0, groups = input_405_groups_0, pad = input_405_pad_0, pad_type = input_405_pad_type_0, strides = input_405_strides_0, weight = layers_14_feed_forward2_linear1_weight_to_fp16_palettized, x = input_403_cast_fp16)[name = string("input_405_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_407_cast_fp16 = silu(x = input_405_cast_fp16)[name = string("input_407_cast_fp16")];
            string var_3977_pad_type_0 = const()[name = string("op_3977_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_3977_strides_0 = const()[name = string("op_3977_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_3977_pad_0 = const()[name = string("op_3977_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_3977_dilations_0 = const()[name = string("op_3977_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_3977_groups_0 = const()[name = string("op_3977_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_3978_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285707904))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288853696))))[name = string("op_3978_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_3978_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_3977_dilations_0, groups = var_3977_groups_0, pad = var_3977_pad_0, pad_type = var_3977_pad_type_0, strides = var_3977_strides_0, weight = op_3978_weight_0_to_fp16_palettized, x = input_407_cast_fp16)[name = string("op_3978_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_331_cast_fp16 = add(x = x_329_cast_fp16, y = var_3978_cast_fp16)[name = string("x_331_cast_fp16")];
            tensor<int32, [1]> var_3994_axes_0 = const()[name = string("op_3994_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_14_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_14_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_3994_cast_fp16 = layer_norm(axes = var_3994_axes_0, epsilon = layers_14_norm_out_eps_scaled_to_fp16, x = x_331_cast_fp16)[name = string("op_3994_cast_fp16")];
            tensor<fp16, [1024]> x_333_gamma_0_to_fp16 = const()[name = string("x_333_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288861952)))];
            tensor<fp16, [1024]> x_333_beta_0_to_fp16 = const()[name = string("x_333_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288864064)))];
            fp16 x_333_epsilon_0_to_fp16 = const()[name = string("x_333_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_333_cast_fp16 = batch_norm(beta = x_333_beta_0_to_fp16, epsilon = x_333_epsilon_0_to_fp16, gamma = x_333_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_3994_cast_fp16)[name = string("x_333_cast_fp16")];
            int32 var_4013 = const()[name = string("op_4013"), val = int32(1)];
            tensor<int32, [1]> var_4040_axes_0 = const()[name = string("op_4040_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_15_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_15_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4040_cast_fp16 = layer_norm(axes = var_4040_axes_0, epsilon = layers_15_norm_feed_forward1_eps_scaled_to_fp16, x = x_333_cast_fp16)[name = string("op_4040_cast_fp16")];
            tensor<fp16, [1024]> input_409_gamma_0_to_fp16 = const()[name = string("input_409_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288866176)))];
            tensor<fp16, [1024]> input_409_beta_0_to_fp16 = const()[name = string("input_409_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288868288)))];
            fp16 input_409_epsilon_0_to_fp16 = const()[name = string("input_409_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_409_cast_fp16 = batch_norm(beta = input_409_beta_0_to_fp16, epsilon = input_409_epsilon_0_to_fp16, gamma = input_409_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4040_cast_fp16)[name = string("input_409_cast_fp16")];
            string input_411_pad_type_0 = const()[name = string("input_411_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_411_strides_0 = const()[name = string("input_411_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_411_pad_0 = const()[name = string("input_411_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_411_dilations_0 = const()[name = string("input_411_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_411_groups_0 = const()[name = string("input_411_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_15_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288870400))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292016192))))[name = string("layers_15_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_411_cast_fp16 = conv(dilations = input_411_dilations_0, groups = input_411_groups_0, pad = input_411_pad_0, pad_type = input_411_pad_type_0, strides = input_411_strides_0, weight = layers_15_feed_forward1_linear1_weight_to_fp16_palettized, x = input_409_cast_fp16)[name = string("input_411_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_413_cast_fp16 = silu(x = input_411_cast_fp16)[name = string("input_413_cast_fp16")];
            string var_4057_pad_type_0 = const()[name = string("op_4057_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4057_strides_0 = const()[name = string("op_4057_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4057_pad_0 = const()[name = string("op_4057_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4057_dilations_0 = const()[name = string("op_4057_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4057_groups_0 = const()[name = string("op_4057_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_4058_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292049024))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295194816))))[name = string("op_4058_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4058_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_4057_dilations_0, groups = var_4057_groups_0, pad = var_4057_pad_0, pad_type = var_4057_pad_type_0, strides = var_4057_strides_0, weight = op_4058_weight_0_to_fp16_palettized, x = input_413_cast_fp16)[name = string("op_4058_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_335_cast_fp16 = add(x = x_333_cast_fp16, y = var_4058_cast_fp16)[name = string("x_335_cast_fp16")];
            tensor<int32, [1]> var_4074_axes_0 = const()[name = string("op_4074_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_15_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_15_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4074_cast_fp16 = layer_norm(axes = var_4074_axes_0, epsilon = layers_15_norm_self_att_eps_scaled_to_fp16, x = x_335_cast_fp16)[name = string("op_4074_cast_fp16")];
            tensor<fp16, [1024]> x_337_gamma_0_to_fp16 = const()[name = string("x_337_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295203072)))];
            tensor<fp16, [1024]> x_337_beta_0_to_fp16 = const()[name = string("x_337_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295205184)))];
            fp16 x_337_epsilon_0_to_fp16 = const()[name = string("x_337_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_337_cast_fp16 = batch_norm(beta = x_337_beta_0_to_fp16, epsilon = x_337_epsilon_0_to_fp16, gamma = x_337_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4074_cast_fp16)[name = string("x_337_cast_fp16")];
            string q_31_pad_type_0 = const()[name = string("q_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_31_strides_0 = const()[name = string("q_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_31_pad_0 = const()[name = string("q_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_31_dilations_0 = const()[name = string("q_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_31_groups_0 = const()[name = string("q_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295207296))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295993792))))[name = string("layers_15_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_31_cast_fp16 = conv(dilations = q_31_dilations_0, groups = q_31_groups_0, pad = q_31_pad_0, pad_type = q_31_pad_type_0, strides = q_31_strides_0, weight = layers_15_self_attn_linear_q_weight_to_fp16_palettized, x = x_337_cast_fp16)[name = string("q_31_cast_fp16")];
            string k_31_pad_type_0 = const()[name = string("k_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_31_strides_0 = const()[name = string("k_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_31_pad_0 = const()[name = string("k_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_31_dilations_0 = const()[name = string("k_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_31_groups_0 = const()[name = string("k_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296002048))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296788544))))[name = string("layers_15_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_31_cast_fp16 = conv(dilations = k_31_dilations_0, groups = k_31_groups_0, pad = k_31_pad_0, pad_type = k_31_pad_type_0, strides = k_31_strides_0, weight = layers_15_self_attn_linear_k_weight_to_fp16_palettized, x = x_337_cast_fp16)[name = string("k_31_cast_fp16")];
            string v_31_pad_type_0 = const()[name = string("v_31_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_31_strides_0 = const()[name = string("v_31_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_31_pad_0 = const()[name = string("v_31_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_31_dilations_0 = const()[name = string("v_31_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_31_groups_0 = const()[name = string("v_31_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(296796800))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297583296))))[name = string("layers_15_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_31_cast_fp16 = conv(dilations = v_31_dilations_0, groups = v_31_groups_0, pad = v_31_pad_0, pad_type = v_31_pad_type_0, strides = v_31_strides_0, weight = layers_15_self_attn_linear_v_weight_to_fp16_palettized, x = x_337_cast_fp16)[name = string("v_31_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_31_to_fp16 = const()[name = string("bv_all_31_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297591552)))];
            tensor<fp16, [1, 1024, 1, 188]> var_4106_cast_fp16 = add(x = q_31_cast_fp16, y = bv_all_31_to_fp16)[name = string("op_4106_cast_fp16")];
            tensor<int32, [3]> var_4107 = const()[name = string("op_4107"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_31_cast_fp16 = reshape(shape = var_4107, x = var_4106_cast_fp16)[name = string("qb_31_cast_fp16")];
            bool bd_all_61_transpose_x_0 = const()[name = string("bd_all_61_transpose_x_0"), val = bool(false)];
            bool bd_all_61_transpose_y_0 = const()[name = string("bd_all_61_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_15_self_attn_pos_proj_to_fp16 = const()[name = string("layers_15_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297593664)))];
            tensor<fp16, [8, 375, 188]> bd_all_61_cast_fp16 = matmul(transpose_x = bd_all_61_transpose_x_0, transpose_y = bd_all_61_transpose_y_0, x = layers_15_self_attn_pos_proj_to_fp16, y = qb_31_cast_fp16)[name = string("bd_all_61_cast_fp16")];
            tensor<int32, [3]> x_339_perm_0 = const()[name = string("x_339_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_341_pad_0 = const()[name = string("x_341_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_341_mode_0 = const()[name = string("x_341_mode_0"), val = string("constant")];
            fp16 const_70_to_fp16 = const()[name = string("const_70_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_339_cast_fp16 = transpose(perm = x_339_perm_0, x = bd_all_61_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [8, 188, 376]> x_341_cast_fp16 = pad(constant_val = const_70_to_fp16, mode = x_341_mode_0, pad = x_341_pad_0, x = x_339_cast_fp16)[name = string("x_341_cast_fp16")];
            tensor<int32, [3]> var_4114 = const()[name = string("op_4114"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_343_cast_fp16 = reshape(shape = var_4114, x = x_341_cast_fp16)[name = string("x_343_cast_fp16")];
            tensor<int32, [3]> var_4117_begin_0 = const()[name = string("op_4117_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_4117_end_0 = const()[name = string("op_4117_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_4117_end_mask_0 = const()[name = string("op_4117_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_4117_cast_fp16 = slice_by_index(begin = var_4117_begin_0, end = var_4117_end_0, end_mask = var_4117_end_mask_0, x = x_343_cast_fp16)[name = string("op_4117_cast_fp16")];
            tensor<int32, [3]> var_4118 = const()[name = string("op_4118"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_345_cast_fp16 = reshape(shape = var_4118, x = var_4117_cast_fp16)[name = string("x_345_cast_fp16")];
            tensor<int32, [3]> bd_all_63_begin_0 = const()[name = string("bd_all_63_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_63_end_0 = const()[name = string("bd_all_63_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_63_end_mask_0 = const()[name = string("bd_all_63_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_63_cast_fp16 = slice_by_index(begin = bd_all_63_begin_0, end = bd_all_63_end_0, end_mask = bd_all_63_end_mask_0, x = x_345_cast_fp16)[name = string("bd_all_63_cast_fp16")];
            tensor<int32, [4]> var_4123 = const()[name = string("op_4123"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_4124_cast_fp16 = reshape(shape = var_4123, x = q_31_cast_fp16)[name = string("op_4124_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_4126_to_fp16 = const()[name = string("op_4126_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298361728)))];
            tensor<fp16, [8, 128, 1, 188]> var_4127_cast_fp16 = add(x = var_4124_cast_fp16, y = var_4126_to_fp16)[name = string("op_4127_cast_fp16")];
            tensor<int32, [4]> var_4128 = const()[name = string("op_4128"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_31_cast_fp16 = reshape(shape = var_4128, x = k_31_cast_fp16)[name = string("kh_31_cast_fp16")];
            tensor<int32, [4]> var_4130 = const()[name = string("op_4130"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_31_cast_fp16 = reshape(shape = var_4130, x = v_31_cast_fp16)[name = string("vh_31_cast_fp16")];
            tensor<int32, [4]> var_4132 = const()[name = string("op_4132"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_31_equation_0 = const()[name = string("ac_31_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_4133_cast_fp16 = transpose(perm = var_4132, x = kh_31_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [8, 188, 1, 188]> ac_31_cast_fp16 = einsum(equation = ac_31_equation_0, values = (var_4133_cast_fp16, var_4127_cast_fp16))[name = string("ac_31_cast_fp16")];
            tensor<int32, [3]> var_4136_perm_0 = const()[name = string("op_4136_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_4137_axes_0 = const()[name = string("op_4137_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_4136_cast_fp16 = transpose(perm = var_4136_perm_0, x = bd_all_63_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [8, 188, 1, 188]> var_4137_cast_fp16 = expand_dims(axes = var_4137_axes_0, x = var_4136_cast_fp16)[name = string("op_4137_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_4138_cast_fp16 = add(x = ac_31_cast_fp16, y = var_4137_cast_fp16)[name = string("op_4138_cast_fp16")];
            fp16 var_4139_to_fp16 = const()[name = string("op_4139_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_61_cast_fp16 = mul(x = var_4138_cast_fp16, y = var_4139_to_fp16)[name = string("scores_61_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_63_cast_fp16 = add(x = scores_61_cast_fp16, y = key_bias)[name = string("scores_63_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_4142_cast_fp16 = softmax(axis = var_4013, x = scores_63_cast_fp16)[name = string("op_4142_cast_fp16")];
            tensor<int32, [4]> transpose_63_perm_0 = const()[name = string("transpose_63_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_30_perm_0 = const()[name = string("transpose_30_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_154 = const()[name = string("concat_154"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_30_cast_fp16 = transpose(perm = transpose_30_perm_0, x = var_4142_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [8, 188, 188]> reshape_45_cast_fp16 = reshape(shape = concat_154, x = transpose_30_cast_fp16)[name = string("reshape_45_cast_fp16")];
            tensor<int32, [3]> concat_155 = const()[name = string("concat_155"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_63_cast_fp16 = transpose(perm = transpose_63_perm_0, x = vh_31_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [8, 188, 128]> reshape_46_cast_fp16 = reshape(shape = concat_155, x = transpose_63_cast_fp16)[name = string("reshape_46_cast_fp16")];
            bool matmul_15_transpose_x_0 = const()[name = string("matmul_15_transpose_x_0"), val = bool(false)];
            bool matmul_15_transpose_y_0 = const()[name = string("matmul_15_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_15_cast_fp16 = matmul(transpose_x = matmul_15_transpose_x_0, transpose_y = matmul_15_transpose_y_0, x = reshape_45_cast_fp16, y = reshape_46_cast_fp16)[name = string("matmul_15_cast_fp16")];
            tensor<int32, [4]> concat_159 = const()[name = string("concat_159"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_47_cast_fp16 = reshape(shape = concat_159, x = matmul_15_cast_fp16)[name = string("reshape_47_cast_fp16")];
            tensor<int32, [4]> ctx_31_perm_0 = const()[name = string("ctx_31_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_4147 = const()[name = string("op_4147"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_31_cast_fp16 = transpose(perm = ctx_31_perm_0, x = reshape_47_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [1, 1024, 1, 188]> input_415_cast_fp16 = reshape(shape = var_4147, x = ctx_31_cast_fp16)[name = string("input_415_cast_fp16")];
            string var_4154_pad_type_0 = const()[name = string("op_4154_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4154_strides_0 = const()[name = string("op_4154_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4154_pad_0 = const()[name = string("op_4154_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4154_dilations_0 = const()[name = string("op_4154_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4154_groups_0 = const()[name = string("op_4154_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298363840))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299150336))))[name = string("layers_15_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4154_cast_fp16 = conv(dilations = var_4154_dilations_0, groups = var_4154_groups_0, pad = var_4154_pad_0, pad_type = var_4154_pad_type_0, strides = var_4154_strides_0, weight = layers_15_self_attn_linear_out_weight_to_fp16_palettized, x = input_415_cast_fp16)[name = string("op_4154_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_347_cast_fp16 = add(x = x_335_cast_fp16, y = var_4154_cast_fp16)[name = string("x_347_cast_fp16")];
            tensor<int32, [1]> var_4170_axes_0 = const()[name = string("op_4170_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_15_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_15_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4170_cast_fp16 = layer_norm(axes = var_4170_axes_0, epsilon = layers_15_norm_conv_eps_scaled_to_fp16, x = x_347_cast_fp16)[name = string("op_4170_cast_fp16")];
            tensor<fp16, [1024]> input_417_gamma_0_to_fp16 = const()[name = string("input_417_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299158592)))];
            tensor<fp16, [1024]> input_417_beta_0_to_fp16 = const()[name = string("input_417_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299160704)))];
            fp16 input_417_epsilon_0_to_fp16 = const()[name = string("input_417_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_417_cast_fp16 = batch_norm(beta = input_417_beta_0_to_fp16, epsilon = input_417_epsilon_0_to_fp16, gamma = input_417_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4170_cast_fp16)[name = string("input_417_cast_fp16")];
            string input_419_pad_type_0 = const()[name = string("input_419_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_419_strides_0 = const()[name = string("input_419_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_419_pad_0 = const()[name = string("input_419_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_419_dilations_0 = const()[name = string("input_419_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_419_groups_0 = const()[name = string("input_419_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_15_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299162816))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300735744))))[name = string("layers_15_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_419_cast_fp16 = conv(dilations = input_419_dilations_0, groups = input_419_groups_0, pad = input_419_pad_0, pad_type = input_419_pad_type_0, strides = input_419_strides_0, weight = layers_15_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_417_cast_fp16)[name = string("input_419_cast_fp16")];
            int32 x_349_split_num_splits_0 = const()[name = string("x_349_split_num_splits_0"), val = int32(2)];
            int32 x_349_split_axis_0 = const()[name = string("x_349_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_349_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_419_cast_fp16)[name = string("x_349_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_349_split_1_sigmoid_cast_fp16 = sigmoid(x = x_349_split_cast_fp16_1)[name = string("x_349_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_349_cast_fp16 = mul(x = x_349_split_cast_fp16_0, y = x_349_split_1_sigmoid_cast_fp16)[name = string("x_349_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_421_cast_fp16 = mul(x = x_349_cast_fp16, y = pad_mask)[name = string("input_421_cast_fp16")];
            string input_423_pad_type_0 = const()[name = string("input_423_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_423_pad_0 = const()[name = string("input_423_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_423_groups_0 = const()[name = string("input_423_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_423_strides_0 = const()[name = string("input_423_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_423_dilations_0 = const()[name = string("input_423_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_133_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300752192))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300759168))))[name = string("const_133_to_fp16_palettized")];
            tensor<fp16, [1024]> const_134_to_fp16 = const()[name = string("const_134_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300767424)))];
            tensor<fp16, [1, 1024, 1, 188]> input_425_cast_fp16 = conv(bias = const_134_to_fp16, dilations = input_423_dilations_0, groups = input_423_groups_0, pad = input_423_pad_0, pad_type = input_423_pad_type_0, strides = input_423_strides_0, weight = const_133_to_fp16_palettized, x = input_421_cast_fp16)[name = string("input_425_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_427_cast_fp16 = silu(x = input_425_cast_fp16)[name = string("input_427_cast_fp16")];
            string var_4202_pad_type_0 = const()[name = string("op_4202_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4202_strides_0 = const()[name = string("op_4202_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4202_pad_0 = const()[name = string("op_4202_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4202_dilations_0 = const()[name = string("op_4202_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4202_groups_0 = const()[name = string("op_4202_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_15_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300769536))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301556032))))[name = string("layers_15_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4202_cast_fp16 = conv(dilations = var_4202_dilations_0, groups = var_4202_groups_0, pad = var_4202_pad_0, pad_type = var_4202_pad_type_0, strides = var_4202_strides_0, weight = layers_15_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_427_cast_fp16)[name = string("op_4202_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_351_cast_fp16 = add(x = x_347_cast_fp16, y = var_4202_cast_fp16)[name = string("x_351_cast_fp16")];
            tensor<int32, [1]> var_4218_axes_0 = const()[name = string("op_4218_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_15_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_15_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4218_cast_fp16 = layer_norm(axes = var_4218_axes_0, epsilon = layers_15_norm_feed_forward2_eps_scaled_to_fp16, x = x_351_cast_fp16)[name = string("op_4218_cast_fp16")];
            tensor<fp16, [1024]> input_429_gamma_0_to_fp16 = const()[name = string("input_429_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301564288)))];
            tensor<fp16, [1024]> input_429_beta_0_to_fp16 = const()[name = string("input_429_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301566400)))];
            fp16 input_429_epsilon_0_to_fp16 = const()[name = string("input_429_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_429_cast_fp16 = batch_norm(beta = input_429_beta_0_to_fp16, epsilon = input_429_epsilon_0_to_fp16, gamma = input_429_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4218_cast_fp16)[name = string("input_429_cast_fp16")];
            string input_431_pad_type_0 = const()[name = string("input_431_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_431_strides_0 = const()[name = string("input_431_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_431_pad_0 = const()[name = string("input_431_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_431_dilations_0 = const()[name = string("input_431_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_431_groups_0 = const()[name = string("input_431_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_15_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301568512))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304714304))))[name = string("layers_15_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_431_cast_fp16 = conv(dilations = input_431_dilations_0, groups = input_431_groups_0, pad = input_431_pad_0, pad_type = input_431_pad_type_0, strides = input_431_strides_0, weight = layers_15_feed_forward2_linear1_weight_to_fp16_palettized, x = input_429_cast_fp16)[name = string("input_431_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_433_cast_fp16 = silu(x = input_431_cast_fp16)[name = string("input_433_cast_fp16")];
            string var_4235_pad_type_0 = const()[name = string("op_4235_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4235_strides_0 = const()[name = string("op_4235_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4235_pad_0 = const()[name = string("op_4235_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4235_dilations_0 = const()[name = string("op_4235_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4235_groups_0 = const()[name = string("op_4235_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_4236_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304747136))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307892928))))[name = string("op_4236_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4236_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_4235_dilations_0, groups = var_4235_groups_0, pad = var_4235_pad_0, pad_type = var_4235_pad_type_0, strides = var_4235_strides_0, weight = op_4236_weight_0_to_fp16_palettized, x = input_433_cast_fp16)[name = string("op_4236_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_353_cast_fp16 = add(x = x_351_cast_fp16, y = var_4236_cast_fp16)[name = string("x_353_cast_fp16")];
            tensor<int32, [1]> var_4252_axes_0 = const()[name = string("op_4252_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_15_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_15_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4252_cast_fp16 = layer_norm(axes = var_4252_axes_0, epsilon = layers_15_norm_out_eps_scaled_to_fp16, x = x_353_cast_fp16)[name = string("op_4252_cast_fp16")];
            tensor<fp16, [1024]> x_355_gamma_0_to_fp16 = const()[name = string("x_355_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307901184)))];
            tensor<fp16, [1024]> x_355_beta_0_to_fp16 = const()[name = string("x_355_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307903296)))];
            fp16 x_355_epsilon_0_to_fp16 = const()[name = string("x_355_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_355_cast_fp16 = batch_norm(beta = x_355_beta_0_to_fp16, epsilon = x_355_epsilon_0_to_fp16, gamma = x_355_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4252_cast_fp16)[name = string("x_355_cast_fp16")];
            int32 var_4271 = const()[name = string("op_4271"), val = int32(1)];
            tensor<int32, [1]> var_4298_axes_0 = const()[name = string("op_4298_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_16_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_16_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4298_cast_fp16 = layer_norm(axes = var_4298_axes_0, epsilon = layers_16_norm_feed_forward1_eps_scaled_to_fp16, x = x_355_cast_fp16)[name = string("op_4298_cast_fp16")];
            tensor<fp16, [1024]> input_435_gamma_0_to_fp16 = const()[name = string("input_435_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307905408)))];
            tensor<fp16, [1024]> input_435_beta_0_to_fp16 = const()[name = string("input_435_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307907520)))];
            fp16 input_435_epsilon_0_to_fp16 = const()[name = string("input_435_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_435_cast_fp16 = batch_norm(beta = input_435_beta_0_to_fp16, epsilon = input_435_epsilon_0_to_fp16, gamma = input_435_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4298_cast_fp16)[name = string("input_435_cast_fp16")];
            string input_437_pad_type_0 = const()[name = string("input_437_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_437_strides_0 = const()[name = string("input_437_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_437_pad_0 = const()[name = string("input_437_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_437_dilations_0 = const()[name = string("input_437_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_437_groups_0 = const()[name = string("input_437_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_16_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(307909632))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311055424))))[name = string("layers_16_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_437_cast_fp16 = conv(dilations = input_437_dilations_0, groups = input_437_groups_0, pad = input_437_pad_0, pad_type = input_437_pad_type_0, strides = input_437_strides_0, weight = layers_16_feed_forward1_linear1_weight_to_fp16_palettized, x = input_435_cast_fp16)[name = string("input_437_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_439_cast_fp16 = silu(x = input_437_cast_fp16)[name = string("input_439_cast_fp16")];
            string var_4315_pad_type_0 = const()[name = string("op_4315_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4315_strides_0 = const()[name = string("op_4315_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4315_pad_0 = const()[name = string("op_4315_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4315_dilations_0 = const()[name = string("op_4315_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4315_groups_0 = const()[name = string("op_4315_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_4316_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311088256))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314234048))))[name = string("op_4316_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4316_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_4315_dilations_0, groups = var_4315_groups_0, pad = var_4315_pad_0, pad_type = var_4315_pad_type_0, strides = var_4315_strides_0, weight = op_4316_weight_0_to_fp16_palettized, x = input_439_cast_fp16)[name = string("op_4316_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_357_cast_fp16 = add(x = x_355_cast_fp16, y = var_4316_cast_fp16)[name = string("x_357_cast_fp16")];
            tensor<int32, [1]> var_4332_axes_0 = const()[name = string("op_4332_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_16_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_16_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4332_cast_fp16 = layer_norm(axes = var_4332_axes_0, epsilon = layers_16_norm_self_att_eps_scaled_to_fp16, x = x_357_cast_fp16)[name = string("op_4332_cast_fp16")];
            tensor<fp16, [1024]> x_359_gamma_0_to_fp16 = const()[name = string("x_359_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314242304)))];
            tensor<fp16, [1024]> x_359_beta_0_to_fp16 = const()[name = string("x_359_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314244416)))];
            fp16 x_359_epsilon_0_to_fp16 = const()[name = string("x_359_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_359_cast_fp16 = batch_norm(beta = x_359_beta_0_to_fp16, epsilon = x_359_epsilon_0_to_fp16, gamma = x_359_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4332_cast_fp16)[name = string("x_359_cast_fp16")];
            string q_33_pad_type_0 = const()[name = string("q_33_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_33_strides_0 = const()[name = string("q_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_33_pad_0 = const()[name = string("q_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_33_dilations_0 = const()[name = string("q_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_33_groups_0 = const()[name = string("q_33_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314246528))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315033024))))[name = string("layers_16_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_33_cast_fp16 = conv(dilations = q_33_dilations_0, groups = q_33_groups_0, pad = q_33_pad_0, pad_type = q_33_pad_type_0, strides = q_33_strides_0, weight = layers_16_self_attn_linear_q_weight_to_fp16_palettized, x = x_359_cast_fp16)[name = string("q_33_cast_fp16")];
            string k_33_pad_type_0 = const()[name = string("k_33_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_33_strides_0 = const()[name = string("k_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_33_pad_0 = const()[name = string("k_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_33_dilations_0 = const()[name = string("k_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_33_groups_0 = const()[name = string("k_33_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315041280))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315827776))))[name = string("layers_16_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_33_cast_fp16 = conv(dilations = k_33_dilations_0, groups = k_33_groups_0, pad = k_33_pad_0, pad_type = k_33_pad_type_0, strides = k_33_strides_0, weight = layers_16_self_attn_linear_k_weight_to_fp16_palettized, x = x_359_cast_fp16)[name = string("k_33_cast_fp16")];
            string v_33_pad_type_0 = const()[name = string("v_33_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_33_strides_0 = const()[name = string("v_33_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_33_pad_0 = const()[name = string("v_33_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_33_dilations_0 = const()[name = string("v_33_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_33_groups_0 = const()[name = string("v_33_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315836032))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316622528))))[name = string("layers_16_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_33_cast_fp16 = conv(dilations = v_33_dilations_0, groups = v_33_groups_0, pad = v_33_pad_0, pad_type = v_33_pad_type_0, strides = v_33_strides_0, weight = layers_16_self_attn_linear_v_weight_to_fp16_palettized, x = x_359_cast_fp16)[name = string("v_33_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_33_to_fp16 = const()[name = string("bv_all_33_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316630784)))];
            tensor<fp16, [1, 1024, 1, 188]> var_4364_cast_fp16 = add(x = q_33_cast_fp16, y = bv_all_33_to_fp16)[name = string("op_4364_cast_fp16")];
            tensor<int32, [3]> var_4365 = const()[name = string("op_4365"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_33_cast_fp16 = reshape(shape = var_4365, x = var_4364_cast_fp16)[name = string("qb_33_cast_fp16")];
            bool bd_all_65_transpose_x_0 = const()[name = string("bd_all_65_transpose_x_0"), val = bool(false)];
            bool bd_all_65_transpose_y_0 = const()[name = string("bd_all_65_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_16_self_attn_pos_proj_to_fp16 = const()[name = string("layers_16_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316632896)))];
            tensor<fp16, [8, 375, 188]> bd_all_65_cast_fp16 = matmul(transpose_x = bd_all_65_transpose_x_0, transpose_y = bd_all_65_transpose_y_0, x = layers_16_self_attn_pos_proj_to_fp16, y = qb_33_cast_fp16)[name = string("bd_all_65_cast_fp16")];
            tensor<int32, [3]> x_361_perm_0 = const()[name = string("x_361_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_363_pad_0 = const()[name = string("x_363_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_363_mode_0 = const()[name = string("x_363_mode_0"), val = string("constant")];
            fp16 const_74_to_fp16 = const()[name = string("const_74_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_361_cast_fp16 = transpose(perm = x_361_perm_0, x = bd_all_65_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [8, 188, 376]> x_363_cast_fp16 = pad(constant_val = const_74_to_fp16, mode = x_363_mode_0, pad = x_363_pad_0, x = x_361_cast_fp16)[name = string("x_363_cast_fp16")];
            tensor<int32, [3]> var_4372 = const()[name = string("op_4372"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_365_cast_fp16 = reshape(shape = var_4372, x = x_363_cast_fp16)[name = string("x_365_cast_fp16")];
            tensor<int32, [3]> var_4375_begin_0 = const()[name = string("op_4375_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_4375_end_0 = const()[name = string("op_4375_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_4375_end_mask_0 = const()[name = string("op_4375_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_4375_cast_fp16 = slice_by_index(begin = var_4375_begin_0, end = var_4375_end_0, end_mask = var_4375_end_mask_0, x = x_365_cast_fp16)[name = string("op_4375_cast_fp16")];
            tensor<int32, [3]> var_4376 = const()[name = string("op_4376"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_367_cast_fp16 = reshape(shape = var_4376, x = var_4375_cast_fp16)[name = string("x_367_cast_fp16")];
            tensor<int32, [3]> bd_all_67_begin_0 = const()[name = string("bd_all_67_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_67_end_0 = const()[name = string("bd_all_67_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_67_end_mask_0 = const()[name = string("bd_all_67_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_67_cast_fp16 = slice_by_index(begin = bd_all_67_begin_0, end = bd_all_67_end_0, end_mask = bd_all_67_end_mask_0, x = x_367_cast_fp16)[name = string("bd_all_67_cast_fp16")];
            tensor<int32, [4]> var_4381 = const()[name = string("op_4381"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_4382_cast_fp16 = reshape(shape = var_4381, x = q_33_cast_fp16)[name = string("op_4382_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_4384_to_fp16 = const()[name = string("op_4384_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317400960)))];
            tensor<fp16, [8, 128, 1, 188]> var_4385_cast_fp16 = add(x = var_4382_cast_fp16, y = var_4384_to_fp16)[name = string("op_4385_cast_fp16")];
            tensor<int32, [4]> var_4386 = const()[name = string("op_4386"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_33_cast_fp16 = reshape(shape = var_4386, x = k_33_cast_fp16)[name = string("kh_33_cast_fp16")];
            tensor<int32, [4]> var_4388 = const()[name = string("op_4388"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_33_cast_fp16 = reshape(shape = var_4388, x = v_33_cast_fp16)[name = string("vh_33_cast_fp16")];
            tensor<int32, [4]> var_4390 = const()[name = string("op_4390"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_33_equation_0 = const()[name = string("ac_33_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_4391_cast_fp16 = transpose(perm = var_4390, x = kh_33_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [8, 188, 1, 188]> ac_33_cast_fp16 = einsum(equation = ac_33_equation_0, values = (var_4391_cast_fp16, var_4385_cast_fp16))[name = string("ac_33_cast_fp16")];
            tensor<int32, [3]> var_4394_perm_0 = const()[name = string("op_4394_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_4395_axes_0 = const()[name = string("op_4395_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_4394_cast_fp16 = transpose(perm = var_4394_perm_0, x = bd_all_67_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [8, 188, 1, 188]> var_4395_cast_fp16 = expand_dims(axes = var_4395_axes_0, x = var_4394_cast_fp16)[name = string("op_4395_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_4396_cast_fp16 = add(x = ac_33_cast_fp16, y = var_4395_cast_fp16)[name = string("op_4396_cast_fp16")];
            fp16 var_4397_to_fp16 = const()[name = string("op_4397_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_65_cast_fp16 = mul(x = var_4396_cast_fp16, y = var_4397_to_fp16)[name = string("scores_65_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_67_cast_fp16 = add(x = scores_65_cast_fp16, y = key_bias)[name = string("scores_67_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_4400_cast_fp16 = softmax(axis = var_4271, x = scores_67_cast_fp16)[name = string("op_4400_cast_fp16")];
            tensor<int32, [4]> transpose_64_perm_0 = const()[name = string("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_32_perm_0 = const()[name = string("transpose_32_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_164 = const()[name = string("concat_164"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_32_cast_fp16 = transpose(perm = transpose_32_perm_0, x = var_4400_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [8, 188, 188]> reshape_48_cast_fp16 = reshape(shape = concat_164, x = transpose_32_cast_fp16)[name = string("reshape_48_cast_fp16")];
            tensor<int32, [3]> concat_165 = const()[name = string("concat_165"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_64_cast_fp16 = transpose(perm = transpose_64_perm_0, x = vh_33_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [8, 188, 128]> reshape_49_cast_fp16 = reshape(shape = concat_165, x = transpose_64_cast_fp16)[name = string("reshape_49_cast_fp16")];
            bool matmul_16_transpose_x_0 = const()[name = string("matmul_16_transpose_x_0"), val = bool(false)];
            bool matmul_16_transpose_y_0 = const()[name = string("matmul_16_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_16_cast_fp16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = reshape_48_cast_fp16, y = reshape_49_cast_fp16)[name = string("matmul_16_cast_fp16")];
            tensor<int32, [4]> concat_169 = const()[name = string("concat_169"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_50_cast_fp16 = reshape(shape = concat_169, x = matmul_16_cast_fp16)[name = string("reshape_50_cast_fp16")];
            tensor<int32, [4]> ctx_33_perm_0 = const()[name = string("ctx_33_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_4405 = const()[name = string("op_4405"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_33_cast_fp16 = transpose(perm = ctx_33_perm_0, x = reshape_50_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [1, 1024, 1, 188]> input_441_cast_fp16 = reshape(shape = var_4405, x = ctx_33_cast_fp16)[name = string("input_441_cast_fp16")];
            string var_4412_pad_type_0 = const()[name = string("op_4412_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4412_strides_0 = const()[name = string("op_4412_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4412_pad_0 = const()[name = string("op_4412_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4412_dilations_0 = const()[name = string("op_4412_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4412_groups_0 = const()[name = string("op_4412_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317403072))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318189568))))[name = string("layers_16_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4412_cast_fp16 = conv(dilations = var_4412_dilations_0, groups = var_4412_groups_0, pad = var_4412_pad_0, pad_type = var_4412_pad_type_0, strides = var_4412_strides_0, weight = layers_16_self_attn_linear_out_weight_to_fp16_palettized, x = input_441_cast_fp16)[name = string("op_4412_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_369_cast_fp16 = add(x = x_357_cast_fp16, y = var_4412_cast_fp16)[name = string("x_369_cast_fp16")];
            tensor<int32, [1]> var_4428_axes_0 = const()[name = string("op_4428_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_16_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_16_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4428_cast_fp16 = layer_norm(axes = var_4428_axes_0, epsilon = layers_16_norm_conv_eps_scaled_to_fp16, x = x_369_cast_fp16)[name = string("op_4428_cast_fp16")];
            tensor<fp16, [1024]> input_443_gamma_0_to_fp16 = const()[name = string("input_443_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318197824)))];
            tensor<fp16, [1024]> input_443_beta_0_to_fp16 = const()[name = string("input_443_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318199936)))];
            fp16 input_443_epsilon_0_to_fp16 = const()[name = string("input_443_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_443_cast_fp16 = batch_norm(beta = input_443_beta_0_to_fp16, epsilon = input_443_epsilon_0_to_fp16, gamma = input_443_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4428_cast_fp16)[name = string("input_443_cast_fp16")];
            string input_445_pad_type_0 = const()[name = string("input_445_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_445_strides_0 = const()[name = string("input_445_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_445_pad_0 = const()[name = string("input_445_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_445_dilations_0 = const()[name = string("input_445_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_445_groups_0 = const()[name = string("input_445_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_16_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318202048))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319774976))))[name = string("layers_16_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_445_cast_fp16 = conv(dilations = input_445_dilations_0, groups = input_445_groups_0, pad = input_445_pad_0, pad_type = input_445_pad_type_0, strides = input_445_strides_0, weight = layers_16_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_443_cast_fp16)[name = string("input_445_cast_fp16")];
            int32 x_371_split_num_splits_0 = const()[name = string("x_371_split_num_splits_0"), val = int32(2)];
            int32 x_371_split_axis_0 = const()[name = string("x_371_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_371_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_445_cast_fp16)[name = string("x_371_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_371_split_1_sigmoid_cast_fp16 = sigmoid(x = x_371_split_cast_fp16_1)[name = string("x_371_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_371_cast_fp16 = mul(x = x_371_split_cast_fp16_0, y = x_371_split_1_sigmoid_cast_fp16)[name = string("x_371_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_447_cast_fp16 = mul(x = x_371_cast_fp16, y = pad_mask)[name = string("input_447_cast_fp16")];
            string input_449_pad_type_0 = const()[name = string("input_449_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_449_pad_0 = const()[name = string("input_449_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_449_groups_0 = const()[name = string("input_449_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_449_strides_0 = const()[name = string("input_449_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_449_dilations_0 = const()[name = string("input_449_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_135_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319791424))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319798400))))[name = string("const_135_to_fp16_palettized")];
            tensor<fp16, [1024]> const_136_to_fp16 = const()[name = string("const_136_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319806656)))];
            tensor<fp16, [1, 1024, 1, 188]> input_451_cast_fp16 = conv(bias = const_136_to_fp16, dilations = input_449_dilations_0, groups = input_449_groups_0, pad = input_449_pad_0, pad_type = input_449_pad_type_0, strides = input_449_strides_0, weight = const_135_to_fp16_palettized, x = input_447_cast_fp16)[name = string("input_451_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_453_cast_fp16 = silu(x = input_451_cast_fp16)[name = string("input_453_cast_fp16")];
            string var_4460_pad_type_0 = const()[name = string("op_4460_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4460_strides_0 = const()[name = string("op_4460_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4460_pad_0 = const()[name = string("op_4460_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4460_dilations_0 = const()[name = string("op_4460_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4460_groups_0 = const()[name = string("op_4460_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_16_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(319808768))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320595264))))[name = string("layers_16_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4460_cast_fp16 = conv(dilations = var_4460_dilations_0, groups = var_4460_groups_0, pad = var_4460_pad_0, pad_type = var_4460_pad_type_0, strides = var_4460_strides_0, weight = layers_16_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_453_cast_fp16)[name = string("op_4460_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_373_cast_fp16 = add(x = x_369_cast_fp16, y = var_4460_cast_fp16)[name = string("x_373_cast_fp16")];
            tensor<int32, [1]> var_4476_axes_0 = const()[name = string("op_4476_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_16_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_16_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4476_cast_fp16 = layer_norm(axes = var_4476_axes_0, epsilon = layers_16_norm_feed_forward2_eps_scaled_to_fp16, x = x_373_cast_fp16)[name = string("op_4476_cast_fp16")];
            tensor<fp16, [1024]> input_455_gamma_0_to_fp16 = const()[name = string("input_455_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320603520)))];
            tensor<fp16, [1024]> input_455_beta_0_to_fp16 = const()[name = string("input_455_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320605632)))];
            fp16 input_455_epsilon_0_to_fp16 = const()[name = string("input_455_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_455_cast_fp16 = batch_norm(beta = input_455_beta_0_to_fp16, epsilon = input_455_epsilon_0_to_fp16, gamma = input_455_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4476_cast_fp16)[name = string("input_455_cast_fp16")];
            string input_457_pad_type_0 = const()[name = string("input_457_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_457_strides_0 = const()[name = string("input_457_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_457_pad_0 = const()[name = string("input_457_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_457_dilations_0 = const()[name = string("input_457_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_457_groups_0 = const()[name = string("input_457_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_16_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320607744))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323753536))))[name = string("layers_16_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_457_cast_fp16 = conv(dilations = input_457_dilations_0, groups = input_457_groups_0, pad = input_457_pad_0, pad_type = input_457_pad_type_0, strides = input_457_strides_0, weight = layers_16_feed_forward2_linear1_weight_to_fp16_palettized, x = input_455_cast_fp16)[name = string("input_457_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_459_cast_fp16 = silu(x = input_457_cast_fp16)[name = string("input_459_cast_fp16")];
            string var_4493_pad_type_0 = const()[name = string("op_4493_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4493_strides_0 = const()[name = string("op_4493_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4493_pad_0 = const()[name = string("op_4493_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4493_dilations_0 = const()[name = string("op_4493_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4493_groups_0 = const()[name = string("op_4493_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_4494_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323786368))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326932160))))[name = string("op_4494_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4494_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_4493_dilations_0, groups = var_4493_groups_0, pad = var_4493_pad_0, pad_type = var_4493_pad_type_0, strides = var_4493_strides_0, weight = op_4494_weight_0_to_fp16_palettized, x = input_459_cast_fp16)[name = string("op_4494_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_375_cast_fp16 = add(x = x_373_cast_fp16, y = var_4494_cast_fp16)[name = string("x_375_cast_fp16")];
            tensor<int32, [1]> var_4510_axes_0 = const()[name = string("op_4510_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_16_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_16_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4510_cast_fp16 = layer_norm(axes = var_4510_axes_0, epsilon = layers_16_norm_out_eps_scaled_to_fp16, x = x_375_cast_fp16)[name = string("op_4510_cast_fp16")];
            tensor<fp16, [1024]> x_377_gamma_0_to_fp16 = const()[name = string("x_377_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326940416)))];
            tensor<fp16, [1024]> x_377_beta_0_to_fp16 = const()[name = string("x_377_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326942528)))];
            fp16 x_377_epsilon_0_to_fp16 = const()[name = string("x_377_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_377_cast_fp16 = batch_norm(beta = x_377_beta_0_to_fp16, epsilon = x_377_epsilon_0_to_fp16, gamma = x_377_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4510_cast_fp16)[name = string("x_377_cast_fp16")];
            int32 var_4529 = const()[name = string("op_4529"), val = int32(1)];
            tensor<int32, [1]> var_4556_axes_0 = const()[name = string("op_4556_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_17_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_17_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4556_cast_fp16 = layer_norm(axes = var_4556_axes_0, epsilon = layers_17_norm_feed_forward1_eps_scaled_to_fp16, x = x_377_cast_fp16)[name = string("op_4556_cast_fp16")];
            tensor<fp16, [1024]> input_461_gamma_0_to_fp16 = const()[name = string("input_461_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326944640)))];
            tensor<fp16, [1024]> input_461_beta_0_to_fp16 = const()[name = string("input_461_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326946752)))];
            fp16 input_461_epsilon_0_to_fp16 = const()[name = string("input_461_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_461_cast_fp16 = batch_norm(beta = input_461_beta_0_to_fp16, epsilon = input_461_epsilon_0_to_fp16, gamma = input_461_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4556_cast_fp16)[name = string("input_461_cast_fp16")];
            string input_463_pad_type_0 = const()[name = string("input_463_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_463_strides_0 = const()[name = string("input_463_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_463_pad_0 = const()[name = string("input_463_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_463_dilations_0 = const()[name = string("input_463_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_463_groups_0 = const()[name = string("input_463_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_17_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326948864))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330094656))))[name = string("layers_17_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_463_cast_fp16 = conv(dilations = input_463_dilations_0, groups = input_463_groups_0, pad = input_463_pad_0, pad_type = input_463_pad_type_0, strides = input_463_strides_0, weight = layers_17_feed_forward1_linear1_weight_to_fp16_palettized, x = input_461_cast_fp16)[name = string("input_463_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_465_cast_fp16 = silu(x = input_463_cast_fp16)[name = string("input_465_cast_fp16")];
            string var_4573_pad_type_0 = const()[name = string("op_4573_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4573_strides_0 = const()[name = string("op_4573_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4573_pad_0 = const()[name = string("op_4573_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4573_dilations_0 = const()[name = string("op_4573_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4573_groups_0 = const()[name = string("op_4573_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_4574_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330127488))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333273280))))[name = string("op_4574_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4574_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_4573_dilations_0, groups = var_4573_groups_0, pad = var_4573_pad_0, pad_type = var_4573_pad_type_0, strides = var_4573_strides_0, weight = op_4574_weight_0_to_fp16_palettized, x = input_465_cast_fp16)[name = string("op_4574_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_379_cast_fp16 = add(x = x_377_cast_fp16, y = var_4574_cast_fp16)[name = string("x_379_cast_fp16")];
            tensor<int32, [1]> var_4590_axes_0 = const()[name = string("op_4590_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_17_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_17_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4590_cast_fp16 = layer_norm(axes = var_4590_axes_0, epsilon = layers_17_norm_self_att_eps_scaled_to_fp16, x = x_379_cast_fp16)[name = string("op_4590_cast_fp16")];
            tensor<fp16, [1024]> x_381_gamma_0_to_fp16 = const()[name = string("x_381_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333281536)))];
            tensor<fp16, [1024]> x_381_beta_0_to_fp16 = const()[name = string("x_381_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333283648)))];
            fp16 x_381_epsilon_0_to_fp16 = const()[name = string("x_381_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_381_cast_fp16 = batch_norm(beta = x_381_beta_0_to_fp16, epsilon = x_381_epsilon_0_to_fp16, gamma = x_381_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4590_cast_fp16)[name = string("x_381_cast_fp16")];
            string q_35_pad_type_0 = const()[name = string("q_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_35_strides_0 = const()[name = string("q_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_35_pad_0 = const()[name = string("q_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_35_dilations_0 = const()[name = string("q_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_35_groups_0 = const()[name = string("q_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333285760))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334072256))))[name = string("layers_17_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_35_cast_fp16 = conv(dilations = q_35_dilations_0, groups = q_35_groups_0, pad = q_35_pad_0, pad_type = q_35_pad_type_0, strides = q_35_strides_0, weight = layers_17_self_attn_linear_q_weight_to_fp16_palettized, x = x_381_cast_fp16)[name = string("q_35_cast_fp16")];
            string k_35_pad_type_0 = const()[name = string("k_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_35_strides_0 = const()[name = string("k_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_35_pad_0 = const()[name = string("k_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_35_dilations_0 = const()[name = string("k_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_35_groups_0 = const()[name = string("k_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334080512))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334867008))))[name = string("layers_17_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_35_cast_fp16 = conv(dilations = k_35_dilations_0, groups = k_35_groups_0, pad = k_35_pad_0, pad_type = k_35_pad_type_0, strides = k_35_strides_0, weight = layers_17_self_attn_linear_k_weight_to_fp16_palettized, x = x_381_cast_fp16)[name = string("k_35_cast_fp16")];
            string v_35_pad_type_0 = const()[name = string("v_35_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_35_strides_0 = const()[name = string("v_35_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_35_pad_0 = const()[name = string("v_35_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_35_dilations_0 = const()[name = string("v_35_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_35_groups_0 = const()[name = string("v_35_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334875264))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335661760))))[name = string("layers_17_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_35_cast_fp16 = conv(dilations = v_35_dilations_0, groups = v_35_groups_0, pad = v_35_pad_0, pad_type = v_35_pad_type_0, strides = v_35_strides_0, weight = layers_17_self_attn_linear_v_weight_to_fp16_palettized, x = x_381_cast_fp16)[name = string("v_35_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_35_to_fp16 = const()[name = string("bv_all_35_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335670016)))];
            tensor<fp16, [1, 1024, 1, 188]> var_4622_cast_fp16 = add(x = q_35_cast_fp16, y = bv_all_35_to_fp16)[name = string("op_4622_cast_fp16")];
            tensor<int32, [3]> var_4623 = const()[name = string("op_4623"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_35_cast_fp16 = reshape(shape = var_4623, x = var_4622_cast_fp16)[name = string("qb_35_cast_fp16")];
            bool bd_all_69_transpose_x_0 = const()[name = string("bd_all_69_transpose_x_0"), val = bool(false)];
            bool bd_all_69_transpose_y_0 = const()[name = string("bd_all_69_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_17_self_attn_pos_proj_to_fp16 = const()[name = string("layers_17_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335672128)))];
            tensor<fp16, [8, 375, 188]> bd_all_69_cast_fp16 = matmul(transpose_x = bd_all_69_transpose_x_0, transpose_y = bd_all_69_transpose_y_0, x = layers_17_self_attn_pos_proj_to_fp16, y = qb_35_cast_fp16)[name = string("bd_all_69_cast_fp16")];
            tensor<int32, [3]> x_383_perm_0 = const()[name = string("x_383_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_385_pad_0 = const()[name = string("x_385_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_385_mode_0 = const()[name = string("x_385_mode_0"), val = string("constant")];
            fp16 const_78_to_fp16 = const()[name = string("const_78_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_383_cast_fp16 = transpose(perm = x_383_perm_0, x = bd_all_69_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [8, 188, 376]> x_385_cast_fp16 = pad(constant_val = const_78_to_fp16, mode = x_385_mode_0, pad = x_385_pad_0, x = x_383_cast_fp16)[name = string("x_385_cast_fp16")];
            tensor<int32, [3]> var_4630 = const()[name = string("op_4630"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_387_cast_fp16 = reshape(shape = var_4630, x = x_385_cast_fp16)[name = string("x_387_cast_fp16")];
            tensor<int32, [3]> var_4633_begin_0 = const()[name = string("op_4633_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_4633_end_0 = const()[name = string("op_4633_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_4633_end_mask_0 = const()[name = string("op_4633_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_4633_cast_fp16 = slice_by_index(begin = var_4633_begin_0, end = var_4633_end_0, end_mask = var_4633_end_mask_0, x = x_387_cast_fp16)[name = string("op_4633_cast_fp16")];
            tensor<int32, [3]> var_4634 = const()[name = string("op_4634"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_389_cast_fp16 = reshape(shape = var_4634, x = var_4633_cast_fp16)[name = string("x_389_cast_fp16")];
            tensor<int32, [3]> bd_all_71_begin_0 = const()[name = string("bd_all_71_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_71_end_0 = const()[name = string("bd_all_71_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_71_end_mask_0 = const()[name = string("bd_all_71_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_71_cast_fp16 = slice_by_index(begin = bd_all_71_begin_0, end = bd_all_71_end_0, end_mask = bd_all_71_end_mask_0, x = x_389_cast_fp16)[name = string("bd_all_71_cast_fp16")];
            tensor<int32, [4]> var_4639 = const()[name = string("op_4639"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_4640_cast_fp16 = reshape(shape = var_4639, x = q_35_cast_fp16)[name = string("op_4640_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_4642_to_fp16 = const()[name = string("op_4642_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336440192)))];
            tensor<fp16, [8, 128, 1, 188]> var_4643_cast_fp16 = add(x = var_4640_cast_fp16, y = var_4642_to_fp16)[name = string("op_4643_cast_fp16")];
            tensor<int32, [4]> var_4644 = const()[name = string("op_4644"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_35_cast_fp16 = reshape(shape = var_4644, x = k_35_cast_fp16)[name = string("kh_35_cast_fp16")];
            tensor<int32, [4]> var_4646 = const()[name = string("op_4646"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_35_cast_fp16 = reshape(shape = var_4646, x = v_35_cast_fp16)[name = string("vh_35_cast_fp16")];
            tensor<int32, [4]> var_4648 = const()[name = string("op_4648"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_35_equation_0 = const()[name = string("ac_35_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_4649_cast_fp16 = transpose(perm = var_4648, x = kh_35_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [8, 188, 1, 188]> ac_35_cast_fp16 = einsum(equation = ac_35_equation_0, values = (var_4649_cast_fp16, var_4643_cast_fp16))[name = string("ac_35_cast_fp16")];
            tensor<int32, [3]> var_4652_perm_0 = const()[name = string("op_4652_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_4653_axes_0 = const()[name = string("op_4653_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_4652_cast_fp16 = transpose(perm = var_4652_perm_0, x = bd_all_71_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [8, 188, 1, 188]> var_4653_cast_fp16 = expand_dims(axes = var_4653_axes_0, x = var_4652_cast_fp16)[name = string("op_4653_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_4654_cast_fp16 = add(x = ac_35_cast_fp16, y = var_4653_cast_fp16)[name = string("op_4654_cast_fp16")];
            fp16 var_4655_to_fp16 = const()[name = string("op_4655_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_69_cast_fp16 = mul(x = var_4654_cast_fp16, y = var_4655_to_fp16)[name = string("scores_69_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_71_cast_fp16 = add(x = scores_69_cast_fp16, y = key_bias)[name = string("scores_71_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_4658_cast_fp16 = softmax(axis = var_4529, x = scores_71_cast_fp16)[name = string("op_4658_cast_fp16")];
            tensor<int32, [4]> transpose_65_perm_0 = const()[name = string("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_34_perm_0 = const()[name = string("transpose_34_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_174 = const()[name = string("concat_174"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_34_cast_fp16 = transpose(perm = transpose_34_perm_0, x = var_4658_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [8, 188, 188]> reshape_51_cast_fp16 = reshape(shape = concat_174, x = transpose_34_cast_fp16)[name = string("reshape_51_cast_fp16")];
            tensor<int32, [3]> concat_175 = const()[name = string("concat_175"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_65_cast_fp16 = transpose(perm = transpose_65_perm_0, x = vh_35_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [8, 188, 128]> reshape_52_cast_fp16 = reshape(shape = concat_175, x = transpose_65_cast_fp16)[name = string("reshape_52_cast_fp16")];
            bool matmul_17_transpose_x_0 = const()[name = string("matmul_17_transpose_x_0"), val = bool(false)];
            bool matmul_17_transpose_y_0 = const()[name = string("matmul_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_17_cast_fp16 = matmul(transpose_x = matmul_17_transpose_x_0, transpose_y = matmul_17_transpose_y_0, x = reshape_51_cast_fp16, y = reshape_52_cast_fp16)[name = string("matmul_17_cast_fp16")];
            tensor<int32, [4]> concat_179 = const()[name = string("concat_179"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_53_cast_fp16 = reshape(shape = concat_179, x = matmul_17_cast_fp16)[name = string("reshape_53_cast_fp16")];
            tensor<int32, [4]> ctx_35_perm_0 = const()[name = string("ctx_35_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_4663 = const()[name = string("op_4663"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_35_cast_fp16 = transpose(perm = ctx_35_perm_0, x = reshape_53_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 1024, 1, 188]> input_467_cast_fp16 = reshape(shape = var_4663, x = ctx_35_cast_fp16)[name = string("input_467_cast_fp16")];
            string var_4670_pad_type_0 = const()[name = string("op_4670_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4670_strides_0 = const()[name = string("op_4670_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4670_pad_0 = const()[name = string("op_4670_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4670_dilations_0 = const()[name = string("op_4670_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4670_groups_0 = const()[name = string("op_4670_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336442304))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337228800))))[name = string("layers_17_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4670_cast_fp16 = conv(dilations = var_4670_dilations_0, groups = var_4670_groups_0, pad = var_4670_pad_0, pad_type = var_4670_pad_type_0, strides = var_4670_strides_0, weight = layers_17_self_attn_linear_out_weight_to_fp16_palettized, x = input_467_cast_fp16)[name = string("op_4670_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_391_cast_fp16 = add(x = x_379_cast_fp16, y = var_4670_cast_fp16)[name = string("x_391_cast_fp16")];
            tensor<int32, [1]> var_4686_axes_0 = const()[name = string("op_4686_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_17_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_17_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4686_cast_fp16 = layer_norm(axes = var_4686_axes_0, epsilon = layers_17_norm_conv_eps_scaled_to_fp16, x = x_391_cast_fp16)[name = string("op_4686_cast_fp16")];
            tensor<fp16, [1024]> input_469_gamma_0_to_fp16 = const()[name = string("input_469_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337237056)))];
            tensor<fp16, [1024]> input_469_beta_0_to_fp16 = const()[name = string("input_469_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337239168)))];
            fp16 input_469_epsilon_0_to_fp16 = const()[name = string("input_469_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_469_cast_fp16 = batch_norm(beta = input_469_beta_0_to_fp16, epsilon = input_469_epsilon_0_to_fp16, gamma = input_469_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4686_cast_fp16)[name = string("input_469_cast_fp16")];
            string input_471_pad_type_0 = const()[name = string("input_471_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_471_strides_0 = const()[name = string("input_471_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_471_pad_0 = const()[name = string("input_471_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_471_dilations_0 = const()[name = string("input_471_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_471_groups_0 = const()[name = string("input_471_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_17_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337241280))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338814208))))[name = string("layers_17_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_471_cast_fp16 = conv(dilations = input_471_dilations_0, groups = input_471_groups_0, pad = input_471_pad_0, pad_type = input_471_pad_type_0, strides = input_471_strides_0, weight = layers_17_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_469_cast_fp16)[name = string("input_471_cast_fp16")];
            int32 x_393_split_num_splits_0 = const()[name = string("x_393_split_num_splits_0"), val = int32(2)];
            int32 x_393_split_axis_0 = const()[name = string("x_393_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_393_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_393_split_cast_fp16_1 = split(axis = x_393_split_axis_0, num_splits = x_393_split_num_splits_0, x = input_471_cast_fp16)[name = string("x_393_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_393_split_1_sigmoid_cast_fp16 = sigmoid(x = x_393_split_cast_fp16_1)[name = string("x_393_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_393_cast_fp16 = mul(x = x_393_split_cast_fp16_0, y = x_393_split_1_sigmoid_cast_fp16)[name = string("x_393_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_473_cast_fp16 = mul(x = x_393_cast_fp16, y = pad_mask)[name = string("input_473_cast_fp16")];
            string input_475_pad_type_0 = const()[name = string("input_475_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_475_pad_0 = const()[name = string("input_475_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_475_groups_0 = const()[name = string("input_475_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_475_strides_0 = const()[name = string("input_475_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_475_dilations_0 = const()[name = string("input_475_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_137_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338830656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338837632))))[name = string("const_137_to_fp16_palettized")];
            tensor<fp16, [1024]> const_138_to_fp16 = const()[name = string("const_138_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338845888)))];
            tensor<fp16, [1, 1024, 1, 188]> input_477_cast_fp16 = conv(bias = const_138_to_fp16, dilations = input_475_dilations_0, groups = input_475_groups_0, pad = input_475_pad_0, pad_type = input_475_pad_type_0, strides = input_475_strides_0, weight = const_137_to_fp16_palettized, x = input_473_cast_fp16)[name = string("input_477_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_479_cast_fp16 = silu(x = input_477_cast_fp16)[name = string("input_479_cast_fp16")];
            string var_4718_pad_type_0 = const()[name = string("op_4718_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4718_strides_0 = const()[name = string("op_4718_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4718_pad_0 = const()[name = string("op_4718_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4718_dilations_0 = const()[name = string("op_4718_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4718_groups_0 = const()[name = string("op_4718_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_17_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(338848000))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339634496))))[name = string("layers_17_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4718_cast_fp16 = conv(dilations = var_4718_dilations_0, groups = var_4718_groups_0, pad = var_4718_pad_0, pad_type = var_4718_pad_type_0, strides = var_4718_strides_0, weight = layers_17_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_479_cast_fp16)[name = string("op_4718_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_395_cast_fp16 = add(x = x_391_cast_fp16, y = var_4718_cast_fp16)[name = string("x_395_cast_fp16")];
            tensor<int32, [1]> var_4734_axes_0 = const()[name = string("op_4734_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_17_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_17_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4734_cast_fp16 = layer_norm(axes = var_4734_axes_0, epsilon = layers_17_norm_feed_forward2_eps_scaled_to_fp16, x = x_395_cast_fp16)[name = string("op_4734_cast_fp16")];
            tensor<fp16, [1024]> input_481_gamma_0_to_fp16 = const()[name = string("input_481_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339642752)))];
            tensor<fp16, [1024]> input_481_beta_0_to_fp16 = const()[name = string("input_481_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339644864)))];
            fp16 input_481_epsilon_0_to_fp16 = const()[name = string("input_481_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_481_cast_fp16 = batch_norm(beta = input_481_beta_0_to_fp16, epsilon = input_481_epsilon_0_to_fp16, gamma = input_481_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4734_cast_fp16)[name = string("input_481_cast_fp16")];
            string input_483_pad_type_0 = const()[name = string("input_483_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_483_strides_0 = const()[name = string("input_483_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_483_pad_0 = const()[name = string("input_483_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_483_dilations_0 = const()[name = string("input_483_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_483_groups_0 = const()[name = string("input_483_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_17_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339646976))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342792768))))[name = string("layers_17_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_483_cast_fp16 = conv(dilations = input_483_dilations_0, groups = input_483_groups_0, pad = input_483_pad_0, pad_type = input_483_pad_type_0, strides = input_483_strides_0, weight = layers_17_feed_forward2_linear1_weight_to_fp16_palettized, x = input_481_cast_fp16)[name = string("input_483_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_485_cast_fp16 = silu(x = input_483_cast_fp16)[name = string("input_485_cast_fp16")];
            string var_4751_pad_type_0 = const()[name = string("op_4751_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4751_strides_0 = const()[name = string("op_4751_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4751_pad_0 = const()[name = string("op_4751_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4751_dilations_0 = const()[name = string("op_4751_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4751_groups_0 = const()[name = string("op_4751_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_4752_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342825600))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345971392))))[name = string("op_4752_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4752_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_4751_dilations_0, groups = var_4751_groups_0, pad = var_4751_pad_0, pad_type = var_4751_pad_type_0, strides = var_4751_strides_0, weight = op_4752_weight_0_to_fp16_palettized, x = input_485_cast_fp16)[name = string("op_4752_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_397_cast_fp16 = add(x = x_395_cast_fp16, y = var_4752_cast_fp16)[name = string("x_397_cast_fp16")];
            tensor<int32, [1]> var_4768_axes_0 = const()[name = string("op_4768_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_17_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_17_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4768_cast_fp16 = layer_norm(axes = var_4768_axes_0, epsilon = layers_17_norm_out_eps_scaled_to_fp16, x = x_397_cast_fp16)[name = string("op_4768_cast_fp16")];
            tensor<fp16, [1024]> x_399_gamma_0_to_fp16 = const()[name = string("x_399_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345979648)))];
            tensor<fp16, [1024]> x_399_beta_0_to_fp16 = const()[name = string("x_399_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345981760)))];
            fp16 x_399_epsilon_0_to_fp16 = const()[name = string("x_399_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_399_cast_fp16 = batch_norm(beta = x_399_beta_0_to_fp16, epsilon = x_399_epsilon_0_to_fp16, gamma = x_399_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4768_cast_fp16)[name = string("x_399_cast_fp16")];
            int32 var_4787 = const()[name = string("op_4787"), val = int32(1)];
            tensor<int32, [1]> var_4814_axes_0 = const()[name = string("op_4814_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_18_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_18_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4814_cast_fp16 = layer_norm(axes = var_4814_axes_0, epsilon = layers_18_norm_feed_forward1_eps_scaled_to_fp16, x = x_399_cast_fp16)[name = string("op_4814_cast_fp16")];
            tensor<fp16, [1024]> input_487_gamma_0_to_fp16 = const()[name = string("input_487_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345983872)))];
            tensor<fp16, [1024]> input_487_beta_0_to_fp16 = const()[name = string("input_487_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345985984)))];
            fp16 input_487_epsilon_0_to_fp16 = const()[name = string("input_487_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_487_cast_fp16 = batch_norm(beta = input_487_beta_0_to_fp16, epsilon = input_487_epsilon_0_to_fp16, gamma = input_487_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4814_cast_fp16)[name = string("input_487_cast_fp16")];
            string input_489_pad_type_0 = const()[name = string("input_489_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_489_strides_0 = const()[name = string("input_489_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_489_pad_0 = const()[name = string("input_489_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_489_dilations_0 = const()[name = string("input_489_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_489_groups_0 = const()[name = string("input_489_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_18_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345988096))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349133888))))[name = string("layers_18_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_489_cast_fp16 = conv(dilations = input_489_dilations_0, groups = input_489_groups_0, pad = input_489_pad_0, pad_type = input_489_pad_type_0, strides = input_489_strides_0, weight = layers_18_feed_forward1_linear1_weight_to_fp16_palettized, x = input_487_cast_fp16)[name = string("input_489_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_491_cast_fp16 = silu(x = input_489_cast_fp16)[name = string("input_491_cast_fp16")];
            string var_4831_pad_type_0 = const()[name = string("op_4831_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4831_strides_0 = const()[name = string("op_4831_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4831_pad_0 = const()[name = string("op_4831_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4831_dilations_0 = const()[name = string("op_4831_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4831_groups_0 = const()[name = string("op_4831_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_4832_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(349166720))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352312512))))[name = string("op_4832_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4832_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_4831_dilations_0, groups = var_4831_groups_0, pad = var_4831_pad_0, pad_type = var_4831_pad_type_0, strides = var_4831_strides_0, weight = op_4832_weight_0_to_fp16_palettized, x = input_491_cast_fp16)[name = string("op_4832_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_401_cast_fp16 = add(x = x_399_cast_fp16, y = var_4832_cast_fp16)[name = string("x_401_cast_fp16")];
            tensor<int32, [1]> var_4848_axes_0 = const()[name = string("op_4848_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_18_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_18_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4848_cast_fp16 = layer_norm(axes = var_4848_axes_0, epsilon = layers_18_norm_self_att_eps_scaled_to_fp16, x = x_401_cast_fp16)[name = string("op_4848_cast_fp16")];
            tensor<fp16, [1024]> x_403_gamma_0_to_fp16 = const()[name = string("x_403_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352320768)))];
            tensor<fp16, [1024]> x_403_beta_0_to_fp16 = const()[name = string("x_403_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352322880)))];
            fp16 x_403_epsilon_0_to_fp16 = const()[name = string("x_403_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_403_cast_fp16 = batch_norm(beta = x_403_beta_0_to_fp16, epsilon = x_403_epsilon_0_to_fp16, gamma = x_403_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4848_cast_fp16)[name = string("x_403_cast_fp16")];
            string q_37_pad_type_0 = const()[name = string("q_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_37_strides_0 = const()[name = string("q_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_37_pad_0 = const()[name = string("q_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_37_dilations_0 = const()[name = string("q_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_37_groups_0 = const()[name = string("q_37_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352324992))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353111488))))[name = string("layers_18_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_37_cast_fp16 = conv(dilations = q_37_dilations_0, groups = q_37_groups_0, pad = q_37_pad_0, pad_type = q_37_pad_type_0, strides = q_37_strides_0, weight = layers_18_self_attn_linear_q_weight_to_fp16_palettized, x = x_403_cast_fp16)[name = string("q_37_cast_fp16")];
            string k_37_pad_type_0 = const()[name = string("k_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_37_strides_0 = const()[name = string("k_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_37_pad_0 = const()[name = string("k_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_37_dilations_0 = const()[name = string("k_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_37_groups_0 = const()[name = string("k_37_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353119744))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353906240))))[name = string("layers_18_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_37_cast_fp16 = conv(dilations = k_37_dilations_0, groups = k_37_groups_0, pad = k_37_pad_0, pad_type = k_37_pad_type_0, strides = k_37_strides_0, weight = layers_18_self_attn_linear_k_weight_to_fp16_palettized, x = x_403_cast_fp16)[name = string("k_37_cast_fp16")];
            string v_37_pad_type_0 = const()[name = string("v_37_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_37_strides_0 = const()[name = string("v_37_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_37_pad_0 = const()[name = string("v_37_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_37_dilations_0 = const()[name = string("v_37_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_37_groups_0 = const()[name = string("v_37_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353914496))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354700992))))[name = string("layers_18_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_37_cast_fp16 = conv(dilations = v_37_dilations_0, groups = v_37_groups_0, pad = v_37_pad_0, pad_type = v_37_pad_type_0, strides = v_37_strides_0, weight = layers_18_self_attn_linear_v_weight_to_fp16_palettized, x = x_403_cast_fp16)[name = string("v_37_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_37_to_fp16 = const()[name = string("bv_all_37_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354709248)))];
            tensor<fp16, [1, 1024, 1, 188]> var_4880_cast_fp16 = add(x = q_37_cast_fp16, y = bv_all_37_to_fp16)[name = string("op_4880_cast_fp16")];
            tensor<int32, [3]> var_4881 = const()[name = string("op_4881"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_37_cast_fp16 = reshape(shape = var_4881, x = var_4880_cast_fp16)[name = string("qb_37_cast_fp16")];
            bool bd_all_73_transpose_x_0 = const()[name = string("bd_all_73_transpose_x_0"), val = bool(false)];
            bool bd_all_73_transpose_y_0 = const()[name = string("bd_all_73_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_18_self_attn_pos_proj_to_fp16 = const()[name = string("layers_18_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354711360)))];
            tensor<fp16, [8, 375, 188]> bd_all_73_cast_fp16 = matmul(transpose_x = bd_all_73_transpose_x_0, transpose_y = bd_all_73_transpose_y_0, x = layers_18_self_attn_pos_proj_to_fp16, y = qb_37_cast_fp16)[name = string("bd_all_73_cast_fp16")];
            tensor<int32, [3]> x_405_perm_0 = const()[name = string("x_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_407_pad_0 = const()[name = string("x_407_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_407_mode_0 = const()[name = string("x_407_mode_0"), val = string("constant")];
            fp16 const_82_to_fp16 = const()[name = string("const_82_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_405_cast_fp16 = transpose(perm = x_405_perm_0, x = bd_all_73_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [8, 188, 376]> x_407_cast_fp16 = pad(constant_val = const_82_to_fp16, mode = x_407_mode_0, pad = x_407_pad_0, x = x_405_cast_fp16)[name = string("x_407_cast_fp16")];
            tensor<int32, [3]> var_4888 = const()[name = string("op_4888"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_409_cast_fp16 = reshape(shape = var_4888, x = x_407_cast_fp16)[name = string("x_409_cast_fp16")];
            tensor<int32, [3]> var_4891_begin_0 = const()[name = string("op_4891_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_4891_end_0 = const()[name = string("op_4891_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_4891_end_mask_0 = const()[name = string("op_4891_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_4891_cast_fp16 = slice_by_index(begin = var_4891_begin_0, end = var_4891_end_0, end_mask = var_4891_end_mask_0, x = x_409_cast_fp16)[name = string("op_4891_cast_fp16")];
            tensor<int32, [3]> var_4892 = const()[name = string("op_4892"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_411_cast_fp16 = reshape(shape = var_4892, x = var_4891_cast_fp16)[name = string("x_411_cast_fp16")];
            tensor<int32, [3]> bd_all_75_begin_0 = const()[name = string("bd_all_75_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_75_end_0 = const()[name = string("bd_all_75_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_75_end_mask_0 = const()[name = string("bd_all_75_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_75_cast_fp16 = slice_by_index(begin = bd_all_75_begin_0, end = bd_all_75_end_0, end_mask = bd_all_75_end_mask_0, x = x_411_cast_fp16)[name = string("bd_all_75_cast_fp16")];
            tensor<int32, [4]> var_4897 = const()[name = string("op_4897"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_4898_cast_fp16 = reshape(shape = var_4897, x = q_37_cast_fp16)[name = string("op_4898_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_4900_to_fp16 = const()[name = string("op_4900_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355479424)))];
            tensor<fp16, [8, 128, 1, 188]> var_4901_cast_fp16 = add(x = var_4898_cast_fp16, y = var_4900_to_fp16)[name = string("op_4901_cast_fp16")];
            tensor<int32, [4]> var_4902 = const()[name = string("op_4902"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_37_cast_fp16 = reshape(shape = var_4902, x = k_37_cast_fp16)[name = string("kh_37_cast_fp16")];
            tensor<int32, [4]> var_4904 = const()[name = string("op_4904"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_37_cast_fp16 = reshape(shape = var_4904, x = v_37_cast_fp16)[name = string("vh_37_cast_fp16")];
            tensor<int32, [4]> var_4906 = const()[name = string("op_4906"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_37_equation_0 = const()[name = string("ac_37_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_4907_cast_fp16 = transpose(perm = var_4906, x = kh_37_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [8, 188, 1, 188]> ac_37_cast_fp16 = einsum(equation = ac_37_equation_0, values = (var_4907_cast_fp16, var_4901_cast_fp16))[name = string("ac_37_cast_fp16")];
            tensor<int32, [3]> var_4910_perm_0 = const()[name = string("op_4910_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_4911_axes_0 = const()[name = string("op_4911_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_4910_cast_fp16 = transpose(perm = var_4910_perm_0, x = bd_all_75_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [8, 188, 1, 188]> var_4911_cast_fp16 = expand_dims(axes = var_4911_axes_0, x = var_4910_cast_fp16)[name = string("op_4911_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_4912_cast_fp16 = add(x = ac_37_cast_fp16, y = var_4911_cast_fp16)[name = string("op_4912_cast_fp16")];
            fp16 var_4913_to_fp16 = const()[name = string("op_4913_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_73_cast_fp16 = mul(x = var_4912_cast_fp16, y = var_4913_to_fp16)[name = string("scores_73_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_75_cast_fp16 = add(x = scores_73_cast_fp16, y = key_bias)[name = string("scores_75_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_4916_cast_fp16 = softmax(axis = var_4787, x = scores_75_cast_fp16)[name = string("op_4916_cast_fp16")];
            tensor<int32, [4]> transpose_66_perm_0 = const()[name = string("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_36_perm_0 = const()[name = string("transpose_36_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_184 = const()[name = string("concat_184"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_36_cast_fp16 = transpose(perm = transpose_36_perm_0, x = var_4916_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [8, 188, 188]> reshape_54_cast_fp16 = reshape(shape = concat_184, x = transpose_36_cast_fp16)[name = string("reshape_54_cast_fp16")];
            tensor<int32, [3]> concat_185 = const()[name = string("concat_185"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_66_cast_fp16 = transpose(perm = transpose_66_perm_0, x = vh_37_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [8, 188, 128]> reshape_55_cast_fp16 = reshape(shape = concat_185, x = transpose_66_cast_fp16)[name = string("reshape_55_cast_fp16")];
            bool matmul_18_transpose_x_0 = const()[name = string("matmul_18_transpose_x_0"), val = bool(false)];
            bool matmul_18_transpose_y_0 = const()[name = string("matmul_18_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_18_cast_fp16 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = reshape_54_cast_fp16, y = reshape_55_cast_fp16)[name = string("matmul_18_cast_fp16")];
            tensor<int32, [4]> concat_189 = const()[name = string("concat_189"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_56_cast_fp16 = reshape(shape = concat_189, x = matmul_18_cast_fp16)[name = string("reshape_56_cast_fp16")];
            tensor<int32, [4]> ctx_37_perm_0 = const()[name = string("ctx_37_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_4921 = const()[name = string("op_4921"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_37_cast_fp16 = transpose(perm = ctx_37_perm_0, x = reshape_56_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 1024, 1, 188]> input_493_cast_fp16 = reshape(shape = var_4921, x = ctx_37_cast_fp16)[name = string("input_493_cast_fp16")];
            string var_4928_pad_type_0 = const()[name = string("op_4928_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4928_strides_0 = const()[name = string("op_4928_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4928_pad_0 = const()[name = string("op_4928_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4928_dilations_0 = const()[name = string("op_4928_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4928_groups_0 = const()[name = string("op_4928_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355481536))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356268032))))[name = string("layers_18_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4928_cast_fp16 = conv(dilations = var_4928_dilations_0, groups = var_4928_groups_0, pad = var_4928_pad_0, pad_type = var_4928_pad_type_0, strides = var_4928_strides_0, weight = layers_18_self_attn_linear_out_weight_to_fp16_palettized, x = input_493_cast_fp16)[name = string("op_4928_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_413_cast_fp16 = add(x = x_401_cast_fp16, y = var_4928_cast_fp16)[name = string("x_413_cast_fp16")];
            tensor<int32, [1]> var_4944_axes_0 = const()[name = string("op_4944_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_18_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_18_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4944_cast_fp16 = layer_norm(axes = var_4944_axes_0, epsilon = layers_18_norm_conv_eps_scaled_to_fp16, x = x_413_cast_fp16)[name = string("op_4944_cast_fp16")];
            tensor<fp16, [1024]> input_495_gamma_0_to_fp16 = const()[name = string("input_495_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356276288)))];
            tensor<fp16, [1024]> input_495_beta_0_to_fp16 = const()[name = string("input_495_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356278400)))];
            fp16 input_495_epsilon_0_to_fp16 = const()[name = string("input_495_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_495_cast_fp16 = batch_norm(beta = input_495_beta_0_to_fp16, epsilon = input_495_epsilon_0_to_fp16, gamma = input_495_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4944_cast_fp16)[name = string("input_495_cast_fp16")];
            string input_497_pad_type_0 = const()[name = string("input_497_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_497_strides_0 = const()[name = string("input_497_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_497_pad_0 = const()[name = string("input_497_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_497_dilations_0 = const()[name = string("input_497_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_497_groups_0 = const()[name = string("input_497_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_18_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356280512))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357853440))))[name = string("layers_18_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_497_cast_fp16 = conv(dilations = input_497_dilations_0, groups = input_497_groups_0, pad = input_497_pad_0, pad_type = input_497_pad_type_0, strides = input_497_strides_0, weight = layers_18_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_495_cast_fp16)[name = string("input_497_cast_fp16")];
            int32 x_415_split_num_splits_0 = const()[name = string("x_415_split_num_splits_0"), val = int32(2)];
            int32 x_415_split_axis_0 = const()[name = string("x_415_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_415_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_415_split_cast_fp16_1 = split(axis = x_415_split_axis_0, num_splits = x_415_split_num_splits_0, x = input_497_cast_fp16)[name = string("x_415_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_415_split_1_sigmoid_cast_fp16 = sigmoid(x = x_415_split_cast_fp16_1)[name = string("x_415_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_415_cast_fp16 = mul(x = x_415_split_cast_fp16_0, y = x_415_split_1_sigmoid_cast_fp16)[name = string("x_415_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_499_cast_fp16 = mul(x = x_415_cast_fp16, y = pad_mask)[name = string("input_499_cast_fp16")];
            string input_501_pad_type_0 = const()[name = string("input_501_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_501_pad_0 = const()[name = string("input_501_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_501_groups_0 = const()[name = string("input_501_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_501_strides_0 = const()[name = string("input_501_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_501_dilations_0 = const()[name = string("input_501_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_139_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357869888))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357876864))))[name = string("const_139_to_fp16_palettized")];
            tensor<fp16, [1024]> const_140_to_fp16 = const()[name = string("const_140_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357885120)))];
            tensor<fp16, [1, 1024, 1, 188]> input_503_cast_fp16 = conv(bias = const_140_to_fp16, dilations = input_501_dilations_0, groups = input_501_groups_0, pad = input_501_pad_0, pad_type = input_501_pad_type_0, strides = input_501_strides_0, weight = const_139_to_fp16_palettized, x = input_499_cast_fp16)[name = string("input_503_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_505_cast_fp16 = silu(x = input_503_cast_fp16)[name = string("input_505_cast_fp16")];
            string var_4976_pad_type_0 = const()[name = string("op_4976_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_4976_strides_0 = const()[name = string("op_4976_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_4976_pad_0 = const()[name = string("op_4976_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_4976_dilations_0 = const()[name = string("op_4976_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_4976_groups_0 = const()[name = string("op_4976_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_18_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357887232))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(358673728))))[name = string("layers_18_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_4976_cast_fp16 = conv(dilations = var_4976_dilations_0, groups = var_4976_groups_0, pad = var_4976_pad_0, pad_type = var_4976_pad_type_0, strides = var_4976_strides_0, weight = layers_18_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_505_cast_fp16)[name = string("op_4976_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_417_cast_fp16 = add(x = x_413_cast_fp16, y = var_4976_cast_fp16)[name = string("x_417_cast_fp16")];
            tensor<int32, [1]> var_4992_axes_0 = const()[name = string("op_4992_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_18_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_18_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_4992_cast_fp16 = layer_norm(axes = var_4992_axes_0, epsilon = layers_18_norm_feed_forward2_eps_scaled_to_fp16, x = x_417_cast_fp16)[name = string("op_4992_cast_fp16")];
            tensor<fp16, [1024]> input_507_gamma_0_to_fp16 = const()[name = string("input_507_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(358681984)))];
            tensor<fp16, [1024]> input_507_beta_0_to_fp16 = const()[name = string("input_507_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(358684096)))];
            fp16 input_507_epsilon_0_to_fp16 = const()[name = string("input_507_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_507_cast_fp16 = batch_norm(beta = input_507_beta_0_to_fp16, epsilon = input_507_epsilon_0_to_fp16, gamma = input_507_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_4992_cast_fp16)[name = string("input_507_cast_fp16")];
            string input_509_pad_type_0 = const()[name = string("input_509_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_509_strides_0 = const()[name = string("input_509_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_509_pad_0 = const()[name = string("input_509_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_509_dilations_0 = const()[name = string("input_509_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_509_groups_0 = const()[name = string("input_509_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_18_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(358686208))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361832000))))[name = string("layers_18_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_509_cast_fp16 = conv(dilations = input_509_dilations_0, groups = input_509_groups_0, pad = input_509_pad_0, pad_type = input_509_pad_type_0, strides = input_509_strides_0, weight = layers_18_feed_forward2_linear1_weight_to_fp16_palettized, x = input_507_cast_fp16)[name = string("input_509_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_511_cast_fp16 = silu(x = input_509_cast_fp16)[name = string("input_511_cast_fp16")];
            string var_5009_pad_type_0 = const()[name = string("op_5009_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5009_strides_0 = const()[name = string("op_5009_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5009_pad_0 = const()[name = string("op_5009_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5009_dilations_0 = const()[name = string("op_5009_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5009_groups_0 = const()[name = string("op_5009_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_5010_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(361864832))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365010624))))[name = string("op_5010_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5010_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_5009_dilations_0, groups = var_5009_groups_0, pad = var_5009_pad_0, pad_type = var_5009_pad_type_0, strides = var_5009_strides_0, weight = op_5010_weight_0_to_fp16_palettized, x = input_511_cast_fp16)[name = string("op_5010_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_419_cast_fp16 = add(x = x_417_cast_fp16, y = var_5010_cast_fp16)[name = string("x_419_cast_fp16")];
            tensor<int32, [1]> var_5026_axes_0 = const()[name = string("op_5026_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_18_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_18_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5026_cast_fp16 = layer_norm(axes = var_5026_axes_0, epsilon = layers_18_norm_out_eps_scaled_to_fp16, x = x_419_cast_fp16)[name = string("op_5026_cast_fp16")];
            tensor<fp16, [1024]> x_421_gamma_0_to_fp16 = const()[name = string("x_421_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365018880)))];
            tensor<fp16, [1024]> x_421_beta_0_to_fp16 = const()[name = string("x_421_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365020992)))];
            fp16 x_421_epsilon_0_to_fp16 = const()[name = string("x_421_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_421_cast_fp16 = batch_norm(beta = x_421_beta_0_to_fp16, epsilon = x_421_epsilon_0_to_fp16, gamma = x_421_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5026_cast_fp16)[name = string("x_421_cast_fp16")];
            int32 var_5045 = const()[name = string("op_5045"), val = int32(1)];
            tensor<int32, [1]> var_5072_axes_0 = const()[name = string("op_5072_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_19_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_19_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5072_cast_fp16 = layer_norm(axes = var_5072_axes_0, epsilon = layers_19_norm_feed_forward1_eps_scaled_to_fp16, x = x_421_cast_fp16)[name = string("op_5072_cast_fp16")];
            tensor<fp16, [1024]> input_513_gamma_0_to_fp16 = const()[name = string("input_513_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365023104)))];
            tensor<fp16, [1024]> input_513_beta_0_to_fp16 = const()[name = string("input_513_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365025216)))];
            fp16 input_513_epsilon_0_to_fp16 = const()[name = string("input_513_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_513_cast_fp16 = batch_norm(beta = input_513_beta_0_to_fp16, epsilon = input_513_epsilon_0_to_fp16, gamma = input_513_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5072_cast_fp16)[name = string("input_513_cast_fp16")];
            string input_515_pad_type_0 = const()[name = string("input_515_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_515_strides_0 = const()[name = string("input_515_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_515_pad_0 = const()[name = string("input_515_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_515_dilations_0 = const()[name = string("input_515_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_515_groups_0 = const()[name = string("input_515_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_19_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(365027328))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368173120))))[name = string("layers_19_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_515_cast_fp16 = conv(dilations = input_515_dilations_0, groups = input_515_groups_0, pad = input_515_pad_0, pad_type = input_515_pad_type_0, strides = input_515_strides_0, weight = layers_19_feed_forward1_linear1_weight_to_fp16_palettized, x = input_513_cast_fp16)[name = string("input_515_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_517_cast_fp16 = silu(x = input_515_cast_fp16)[name = string("input_517_cast_fp16")];
            string var_5089_pad_type_0 = const()[name = string("op_5089_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5089_strides_0 = const()[name = string("op_5089_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5089_pad_0 = const()[name = string("op_5089_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5089_dilations_0 = const()[name = string("op_5089_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5089_groups_0 = const()[name = string("op_5089_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_5090_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368205952))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371351744))))[name = string("op_5090_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5090_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_5089_dilations_0, groups = var_5089_groups_0, pad = var_5089_pad_0, pad_type = var_5089_pad_type_0, strides = var_5089_strides_0, weight = op_5090_weight_0_to_fp16_palettized, x = input_517_cast_fp16)[name = string("op_5090_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_423_cast_fp16 = add(x = x_421_cast_fp16, y = var_5090_cast_fp16)[name = string("x_423_cast_fp16")];
            tensor<int32, [1]> var_5106_axes_0 = const()[name = string("op_5106_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_19_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_19_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5106_cast_fp16 = layer_norm(axes = var_5106_axes_0, epsilon = layers_19_norm_self_att_eps_scaled_to_fp16, x = x_423_cast_fp16)[name = string("op_5106_cast_fp16")];
            tensor<fp16, [1024]> x_425_gamma_0_to_fp16 = const()[name = string("x_425_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371360000)))];
            tensor<fp16, [1024]> x_425_beta_0_to_fp16 = const()[name = string("x_425_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371362112)))];
            fp16 x_425_epsilon_0_to_fp16 = const()[name = string("x_425_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_425_cast_fp16 = batch_norm(beta = x_425_beta_0_to_fp16, epsilon = x_425_epsilon_0_to_fp16, gamma = x_425_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5106_cast_fp16)[name = string("x_425_cast_fp16")];
            string q_39_pad_type_0 = const()[name = string("q_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_39_strides_0 = const()[name = string("q_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_39_pad_0 = const()[name = string("q_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_39_dilations_0 = const()[name = string("q_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_39_groups_0 = const()[name = string("q_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371364224))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372150720))))[name = string("layers_19_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_39_cast_fp16 = conv(dilations = q_39_dilations_0, groups = q_39_groups_0, pad = q_39_pad_0, pad_type = q_39_pad_type_0, strides = q_39_strides_0, weight = layers_19_self_attn_linear_q_weight_to_fp16_palettized, x = x_425_cast_fp16)[name = string("q_39_cast_fp16")];
            string k_39_pad_type_0 = const()[name = string("k_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_39_strides_0 = const()[name = string("k_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_39_pad_0 = const()[name = string("k_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_39_dilations_0 = const()[name = string("k_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_39_groups_0 = const()[name = string("k_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372158976))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372945472))))[name = string("layers_19_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_39_cast_fp16 = conv(dilations = k_39_dilations_0, groups = k_39_groups_0, pad = k_39_pad_0, pad_type = k_39_pad_type_0, strides = k_39_strides_0, weight = layers_19_self_attn_linear_k_weight_to_fp16_palettized, x = x_425_cast_fp16)[name = string("k_39_cast_fp16")];
            string v_39_pad_type_0 = const()[name = string("v_39_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_39_strides_0 = const()[name = string("v_39_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_39_pad_0 = const()[name = string("v_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_39_dilations_0 = const()[name = string("v_39_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_39_groups_0 = const()[name = string("v_39_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372953728))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373740224))))[name = string("layers_19_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_39_cast_fp16 = conv(dilations = v_39_dilations_0, groups = v_39_groups_0, pad = v_39_pad_0, pad_type = v_39_pad_type_0, strides = v_39_strides_0, weight = layers_19_self_attn_linear_v_weight_to_fp16_palettized, x = x_425_cast_fp16)[name = string("v_39_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_39_to_fp16 = const()[name = string("bv_all_39_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373748480)))];
            tensor<fp16, [1, 1024, 1, 188]> var_5138_cast_fp16 = add(x = q_39_cast_fp16, y = bv_all_39_to_fp16)[name = string("op_5138_cast_fp16")];
            tensor<int32, [3]> var_5139 = const()[name = string("op_5139"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_39_cast_fp16 = reshape(shape = var_5139, x = var_5138_cast_fp16)[name = string("qb_39_cast_fp16")];
            bool bd_all_77_transpose_x_0 = const()[name = string("bd_all_77_transpose_x_0"), val = bool(false)];
            bool bd_all_77_transpose_y_0 = const()[name = string("bd_all_77_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_19_self_attn_pos_proj_to_fp16 = const()[name = string("layers_19_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373750592)))];
            tensor<fp16, [8, 375, 188]> bd_all_77_cast_fp16 = matmul(transpose_x = bd_all_77_transpose_x_0, transpose_y = bd_all_77_transpose_y_0, x = layers_19_self_attn_pos_proj_to_fp16, y = qb_39_cast_fp16)[name = string("bd_all_77_cast_fp16")];
            tensor<int32, [3]> x_427_perm_0 = const()[name = string("x_427_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_429_pad_0 = const()[name = string("x_429_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_429_mode_0 = const()[name = string("x_429_mode_0"), val = string("constant")];
            fp16 const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_427_cast_fp16 = transpose(perm = x_427_perm_0, x = bd_all_77_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [8, 188, 376]> x_429_cast_fp16 = pad(constant_val = const_86_to_fp16, mode = x_429_mode_0, pad = x_429_pad_0, x = x_427_cast_fp16)[name = string("x_429_cast_fp16")];
            tensor<int32, [3]> var_5146 = const()[name = string("op_5146"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_431_cast_fp16 = reshape(shape = var_5146, x = x_429_cast_fp16)[name = string("x_431_cast_fp16")];
            tensor<int32, [3]> var_5149_begin_0 = const()[name = string("op_5149_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_5149_end_0 = const()[name = string("op_5149_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_5149_end_mask_0 = const()[name = string("op_5149_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_5149_cast_fp16 = slice_by_index(begin = var_5149_begin_0, end = var_5149_end_0, end_mask = var_5149_end_mask_0, x = x_431_cast_fp16)[name = string("op_5149_cast_fp16")];
            tensor<int32, [3]> var_5150 = const()[name = string("op_5150"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_433_cast_fp16 = reshape(shape = var_5150, x = var_5149_cast_fp16)[name = string("x_433_cast_fp16")];
            tensor<int32, [3]> bd_all_79_begin_0 = const()[name = string("bd_all_79_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_79_end_0 = const()[name = string("bd_all_79_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_79_end_mask_0 = const()[name = string("bd_all_79_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_79_cast_fp16 = slice_by_index(begin = bd_all_79_begin_0, end = bd_all_79_end_0, end_mask = bd_all_79_end_mask_0, x = x_433_cast_fp16)[name = string("bd_all_79_cast_fp16")];
            tensor<int32, [4]> var_5155 = const()[name = string("op_5155"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_5156_cast_fp16 = reshape(shape = var_5155, x = q_39_cast_fp16)[name = string("op_5156_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_5158_to_fp16 = const()[name = string("op_5158_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(374518656)))];
            tensor<fp16, [8, 128, 1, 188]> var_5159_cast_fp16 = add(x = var_5156_cast_fp16, y = var_5158_to_fp16)[name = string("op_5159_cast_fp16")];
            tensor<int32, [4]> var_5160 = const()[name = string("op_5160"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_39_cast_fp16 = reshape(shape = var_5160, x = k_39_cast_fp16)[name = string("kh_39_cast_fp16")];
            tensor<int32, [4]> var_5162 = const()[name = string("op_5162"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_39_cast_fp16 = reshape(shape = var_5162, x = v_39_cast_fp16)[name = string("vh_39_cast_fp16")];
            tensor<int32, [4]> var_5164 = const()[name = string("op_5164"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_39_equation_0 = const()[name = string("ac_39_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_5165_cast_fp16 = transpose(perm = var_5164, x = kh_39_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [8, 188, 1, 188]> ac_39_cast_fp16 = einsum(equation = ac_39_equation_0, values = (var_5165_cast_fp16, var_5159_cast_fp16))[name = string("ac_39_cast_fp16")];
            tensor<int32, [3]> var_5168_perm_0 = const()[name = string("op_5168_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_5169_axes_0 = const()[name = string("op_5169_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_5168_cast_fp16 = transpose(perm = var_5168_perm_0, x = bd_all_79_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [8, 188, 1, 188]> var_5169_cast_fp16 = expand_dims(axes = var_5169_axes_0, x = var_5168_cast_fp16)[name = string("op_5169_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_5170_cast_fp16 = add(x = ac_39_cast_fp16, y = var_5169_cast_fp16)[name = string("op_5170_cast_fp16")];
            fp16 var_5171_to_fp16 = const()[name = string("op_5171_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_77_cast_fp16 = mul(x = var_5170_cast_fp16, y = var_5171_to_fp16)[name = string("scores_77_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_79_cast_fp16 = add(x = scores_77_cast_fp16, y = key_bias)[name = string("scores_79_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_5174_cast_fp16 = softmax(axis = var_5045, x = scores_79_cast_fp16)[name = string("op_5174_cast_fp16")];
            tensor<int32, [4]> transpose_67_perm_0 = const()[name = string("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_38_perm_0 = const()[name = string("transpose_38_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_194 = const()[name = string("concat_194"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_38_cast_fp16 = transpose(perm = transpose_38_perm_0, x = var_5174_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [8, 188, 188]> reshape_57_cast_fp16 = reshape(shape = concat_194, x = transpose_38_cast_fp16)[name = string("reshape_57_cast_fp16")];
            tensor<int32, [3]> concat_195 = const()[name = string("concat_195"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_67_cast_fp16 = transpose(perm = transpose_67_perm_0, x = vh_39_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [8, 188, 128]> reshape_58_cast_fp16 = reshape(shape = concat_195, x = transpose_67_cast_fp16)[name = string("reshape_58_cast_fp16")];
            bool matmul_19_transpose_x_0 = const()[name = string("matmul_19_transpose_x_0"), val = bool(false)];
            bool matmul_19_transpose_y_0 = const()[name = string("matmul_19_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_19_cast_fp16 = matmul(transpose_x = matmul_19_transpose_x_0, transpose_y = matmul_19_transpose_y_0, x = reshape_57_cast_fp16, y = reshape_58_cast_fp16)[name = string("matmul_19_cast_fp16")];
            tensor<int32, [4]> concat_199 = const()[name = string("concat_199"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_59_cast_fp16 = reshape(shape = concat_199, x = matmul_19_cast_fp16)[name = string("reshape_59_cast_fp16")];
            tensor<int32, [4]> ctx_39_perm_0 = const()[name = string("ctx_39_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_5179 = const()[name = string("op_5179"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_39_cast_fp16 = transpose(perm = ctx_39_perm_0, x = reshape_59_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 1024, 1, 188]> input_519_cast_fp16 = reshape(shape = var_5179, x = ctx_39_cast_fp16)[name = string("input_519_cast_fp16")];
            string var_5186_pad_type_0 = const()[name = string("op_5186_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5186_strides_0 = const()[name = string("op_5186_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5186_pad_0 = const()[name = string("op_5186_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5186_dilations_0 = const()[name = string("op_5186_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5186_groups_0 = const()[name = string("op_5186_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(374520768))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375307264))))[name = string("layers_19_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5186_cast_fp16 = conv(dilations = var_5186_dilations_0, groups = var_5186_groups_0, pad = var_5186_pad_0, pad_type = var_5186_pad_type_0, strides = var_5186_strides_0, weight = layers_19_self_attn_linear_out_weight_to_fp16_palettized, x = input_519_cast_fp16)[name = string("op_5186_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_435_cast_fp16 = add(x = x_423_cast_fp16, y = var_5186_cast_fp16)[name = string("x_435_cast_fp16")];
            tensor<int32, [1]> var_5202_axes_0 = const()[name = string("op_5202_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_19_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_19_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5202_cast_fp16 = layer_norm(axes = var_5202_axes_0, epsilon = layers_19_norm_conv_eps_scaled_to_fp16, x = x_435_cast_fp16)[name = string("op_5202_cast_fp16")];
            tensor<fp16, [1024]> input_521_gamma_0_to_fp16 = const()[name = string("input_521_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375315520)))];
            tensor<fp16, [1024]> input_521_beta_0_to_fp16 = const()[name = string("input_521_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375317632)))];
            fp16 input_521_epsilon_0_to_fp16 = const()[name = string("input_521_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_521_cast_fp16 = batch_norm(beta = input_521_beta_0_to_fp16, epsilon = input_521_epsilon_0_to_fp16, gamma = input_521_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5202_cast_fp16)[name = string("input_521_cast_fp16")];
            string input_523_pad_type_0 = const()[name = string("input_523_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_523_strides_0 = const()[name = string("input_523_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_523_pad_0 = const()[name = string("input_523_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_523_dilations_0 = const()[name = string("input_523_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_523_groups_0 = const()[name = string("input_523_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_19_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375319744))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376892672))))[name = string("layers_19_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_523_cast_fp16 = conv(dilations = input_523_dilations_0, groups = input_523_groups_0, pad = input_523_pad_0, pad_type = input_523_pad_type_0, strides = input_523_strides_0, weight = layers_19_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_521_cast_fp16)[name = string("input_523_cast_fp16")];
            int32 x_437_split_num_splits_0 = const()[name = string("x_437_split_num_splits_0"), val = int32(2)];
            int32 x_437_split_axis_0 = const()[name = string("x_437_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_437_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_437_split_cast_fp16_1 = split(axis = x_437_split_axis_0, num_splits = x_437_split_num_splits_0, x = input_523_cast_fp16)[name = string("x_437_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_437_split_1_sigmoid_cast_fp16 = sigmoid(x = x_437_split_cast_fp16_1)[name = string("x_437_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_437_cast_fp16 = mul(x = x_437_split_cast_fp16_0, y = x_437_split_1_sigmoid_cast_fp16)[name = string("x_437_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_525_cast_fp16 = mul(x = x_437_cast_fp16, y = pad_mask)[name = string("input_525_cast_fp16")];
            string input_527_pad_type_0 = const()[name = string("input_527_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_527_pad_0 = const()[name = string("input_527_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_527_groups_0 = const()[name = string("input_527_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_527_strides_0 = const()[name = string("input_527_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_527_dilations_0 = const()[name = string("input_527_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_141_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376909120))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376916096))))[name = string("const_141_to_fp16_palettized")];
            tensor<fp16, [1024]> const_142_to_fp16 = const()[name = string("const_142_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376924352)))];
            tensor<fp16, [1, 1024, 1, 188]> input_529_cast_fp16 = conv(bias = const_142_to_fp16, dilations = input_527_dilations_0, groups = input_527_groups_0, pad = input_527_pad_0, pad_type = input_527_pad_type_0, strides = input_527_strides_0, weight = const_141_to_fp16_palettized, x = input_525_cast_fp16)[name = string("input_529_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_531_cast_fp16 = silu(x = input_529_cast_fp16)[name = string("input_531_cast_fp16")];
            string var_5234_pad_type_0 = const()[name = string("op_5234_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5234_strides_0 = const()[name = string("op_5234_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5234_pad_0 = const()[name = string("op_5234_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5234_dilations_0 = const()[name = string("op_5234_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5234_groups_0 = const()[name = string("op_5234_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_19_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376926464))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377712960))))[name = string("layers_19_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5234_cast_fp16 = conv(dilations = var_5234_dilations_0, groups = var_5234_groups_0, pad = var_5234_pad_0, pad_type = var_5234_pad_type_0, strides = var_5234_strides_0, weight = layers_19_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_531_cast_fp16)[name = string("op_5234_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_439_cast_fp16 = add(x = x_435_cast_fp16, y = var_5234_cast_fp16)[name = string("x_439_cast_fp16")];
            tensor<int32, [1]> var_5250_axes_0 = const()[name = string("op_5250_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_19_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_19_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5250_cast_fp16 = layer_norm(axes = var_5250_axes_0, epsilon = layers_19_norm_feed_forward2_eps_scaled_to_fp16, x = x_439_cast_fp16)[name = string("op_5250_cast_fp16")];
            tensor<fp16, [1024]> input_533_gamma_0_to_fp16 = const()[name = string("input_533_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377721216)))];
            tensor<fp16, [1024]> input_533_beta_0_to_fp16 = const()[name = string("input_533_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377723328)))];
            fp16 input_533_epsilon_0_to_fp16 = const()[name = string("input_533_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_533_cast_fp16 = batch_norm(beta = input_533_beta_0_to_fp16, epsilon = input_533_epsilon_0_to_fp16, gamma = input_533_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5250_cast_fp16)[name = string("input_533_cast_fp16")];
            string input_535_pad_type_0 = const()[name = string("input_535_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_535_strides_0 = const()[name = string("input_535_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_535_pad_0 = const()[name = string("input_535_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_535_dilations_0 = const()[name = string("input_535_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_535_groups_0 = const()[name = string("input_535_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_19_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377725440))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380871232))))[name = string("layers_19_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_535_cast_fp16 = conv(dilations = input_535_dilations_0, groups = input_535_groups_0, pad = input_535_pad_0, pad_type = input_535_pad_type_0, strides = input_535_strides_0, weight = layers_19_feed_forward2_linear1_weight_to_fp16_palettized, x = input_533_cast_fp16)[name = string("input_535_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_537_cast_fp16 = silu(x = input_535_cast_fp16)[name = string("input_537_cast_fp16")];
            string var_5267_pad_type_0 = const()[name = string("op_5267_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5267_strides_0 = const()[name = string("op_5267_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5267_pad_0 = const()[name = string("op_5267_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5267_dilations_0 = const()[name = string("op_5267_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5267_groups_0 = const()[name = string("op_5267_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_5268_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380904064))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384049856))))[name = string("op_5268_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5268_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_5267_dilations_0, groups = var_5267_groups_0, pad = var_5267_pad_0, pad_type = var_5267_pad_type_0, strides = var_5267_strides_0, weight = op_5268_weight_0_to_fp16_palettized, x = input_537_cast_fp16)[name = string("op_5268_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_441_cast_fp16 = add(x = x_439_cast_fp16, y = var_5268_cast_fp16)[name = string("x_441_cast_fp16")];
            tensor<int32, [1]> var_5284_axes_0 = const()[name = string("op_5284_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_19_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_19_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5284_cast_fp16 = layer_norm(axes = var_5284_axes_0, epsilon = layers_19_norm_out_eps_scaled_to_fp16, x = x_441_cast_fp16)[name = string("op_5284_cast_fp16")];
            tensor<fp16, [1024]> x_443_gamma_0_to_fp16 = const()[name = string("x_443_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384058112)))];
            tensor<fp16, [1024]> x_443_beta_0_to_fp16 = const()[name = string("x_443_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384060224)))];
            fp16 x_443_epsilon_0_to_fp16 = const()[name = string("x_443_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_443_cast_fp16 = batch_norm(beta = x_443_beta_0_to_fp16, epsilon = x_443_epsilon_0_to_fp16, gamma = x_443_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5284_cast_fp16)[name = string("x_443_cast_fp16")];
            int32 var_5303 = const()[name = string("op_5303"), val = int32(1)];
            tensor<int32, [1]> var_5330_axes_0 = const()[name = string("op_5330_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_20_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_20_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5330_cast_fp16 = layer_norm(axes = var_5330_axes_0, epsilon = layers_20_norm_feed_forward1_eps_scaled_to_fp16, x = x_443_cast_fp16)[name = string("op_5330_cast_fp16")];
            tensor<fp16, [1024]> input_539_gamma_0_to_fp16 = const()[name = string("input_539_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384062336)))];
            tensor<fp16, [1024]> input_539_beta_0_to_fp16 = const()[name = string("input_539_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384064448)))];
            fp16 input_539_epsilon_0_to_fp16 = const()[name = string("input_539_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_539_cast_fp16 = batch_norm(beta = input_539_beta_0_to_fp16, epsilon = input_539_epsilon_0_to_fp16, gamma = input_539_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5330_cast_fp16)[name = string("input_539_cast_fp16")];
            string input_541_pad_type_0 = const()[name = string("input_541_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_541_strides_0 = const()[name = string("input_541_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_541_pad_0 = const()[name = string("input_541_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_541_dilations_0 = const()[name = string("input_541_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_541_groups_0 = const()[name = string("input_541_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_20_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384066560))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387212352))))[name = string("layers_20_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_541_cast_fp16 = conv(dilations = input_541_dilations_0, groups = input_541_groups_0, pad = input_541_pad_0, pad_type = input_541_pad_type_0, strides = input_541_strides_0, weight = layers_20_feed_forward1_linear1_weight_to_fp16_palettized, x = input_539_cast_fp16)[name = string("input_541_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_543_cast_fp16 = silu(x = input_541_cast_fp16)[name = string("input_543_cast_fp16")];
            string var_5347_pad_type_0 = const()[name = string("op_5347_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5347_strides_0 = const()[name = string("op_5347_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5347_pad_0 = const()[name = string("op_5347_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5347_dilations_0 = const()[name = string("op_5347_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5347_groups_0 = const()[name = string("op_5347_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_5348_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387245184))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390390976))))[name = string("op_5348_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5348_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_5347_dilations_0, groups = var_5347_groups_0, pad = var_5347_pad_0, pad_type = var_5347_pad_type_0, strides = var_5347_strides_0, weight = op_5348_weight_0_to_fp16_palettized, x = input_543_cast_fp16)[name = string("op_5348_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_445_cast_fp16 = add(x = x_443_cast_fp16, y = var_5348_cast_fp16)[name = string("x_445_cast_fp16")];
            tensor<int32, [1]> var_5364_axes_0 = const()[name = string("op_5364_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_20_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_20_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5364_cast_fp16 = layer_norm(axes = var_5364_axes_0, epsilon = layers_20_norm_self_att_eps_scaled_to_fp16, x = x_445_cast_fp16)[name = string("op_5364_cast_fp16")];
            tensor<fp16, [1024]> x_447_gamma_0_to_fp16 = const()[name = string("x_447_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390399232)))];
            tensor<fp16, [1024]> x_447_beta_0_to_fp16 = const()[name = string("x_447_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390401344)))];
            fp16 x_447_epsilon_0_to_fp16 = const()[name = string("x_447_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_447_cast_fp16 = batch_norm(beta = x_447_beta_0_to_fp16, epsilon = x_447_epsilon_0_to_fp16, gamma = x_447_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5364_cast_fp16)[name = string("x_447_cast_fp16")];
            string q_41_pad_type_0 = const()[name = string("q_41_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_41_strides_0 = const()[name = string("q_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_41_pad_0 = const()[name = string("q_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_41_dilations_0 = const()[name = string("q_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_41_groups_0 = const()[name = string("q_41_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390403456))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391189952))))[name = string("layers_20_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_41_cast_fp16 = conv(dilations = q_41_dilations_0, groups = q_41_groups_0, pad = q_41_pad_0, pad_type = q_41_pad_type_0, strides = q_41_strides_0, weight = layers_20_self_attn_linear_q_weight_to_fp16_palettized, x = x_447_cast_fp16)[name = string("q_41_cast_fp16")];
            string k_41_pad_type_0 = const()[name = string("k_41_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_41_strides_0 = const()[name = string("k_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_41_pad_0 = const()[name = string("k_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_41_dilations_0 = const()[name = string("k_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_41_groups_0 = const()[name = string("k_41_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391198208))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391984704))))[name = string("layers_20_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_41_cast_fp16 = conv(dilations = k_41_dilations_0, groups = k_41_groups_0, pad = k_41_pad_0, pad_type = k_41_pad_type_0, strides = k_41_strides_0, weight = layers_20_self_attn_linear_k_weight_to_fp16_palettized, x = x_447_cast_fp16)[name = string("k_41_cast_fp16")];
            string v_41_pad_type_0 = const()[name = string("v_41_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_41_strides_0 = const()[name = string("v_41_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_41_pad_0 = const()[name = string("v_41_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_41_dilations_0 = const()[name = string("v_41_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_41_groups_0 = const()[name = string("v_41_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391992960))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392779456))))[name = string("layers_20_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_41_cast_fp16 = conv(dilations = v_41_dilations_0, groups = v_41_groups_0, pad = v_41_pad_0, pad_type = v_41_pad_type_0, strides = v_41_strides_0, weight = layers_20_self_attn_linear_v_weight_to_fp16_palettized, x = x_447_cast_fp16)[name = string("v_41_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_41_to_fp16 = const()[name = string("bv_all_41_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392787712)))];
            tensor<fp16, [1, 1024, 1, 188]> var_5396_cast_fp16 = add(x = q_41_cast_fp16, y = bv_all_41_to_fp16)[name = string("op_5396_cast_fp16")];
            tensor<int32, [3]> var_5397 = const()[name = string("op_5397"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_41_cast_fp16 = reshape(shape = var_5397, x = var_5396_cast_fp16)[name = string("qb_41_cast_fp16")];
            bool bd_all_81_transpose_x_0 = const()[name = string("bd_all_81_transpose_x_0"), val = bool(false)];
            bool bd_all_81_transpose_y_0 = const()[name = string("bd_all_81_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_20_self_attn_pos_proj_to_fp16 = const()[name = string("layers_20_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392789824)))];
            tensor<fp16, [8, 375, 188]> bd_all_81_cast_fp16 = matmul(transpose_x = bd_all_81_transpose_x_0, transpose_y = bd_all_81_transpose_y_0, x = layers_20_self_attn_pos_proj_to_fp16, y = qb_41_cast_fp16)[name = string("bd_all_81_cast_fp16")];
            tensor<int32, [3]> x_449_perm_0 = const()[name = string("x_449_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_451_pad_0 = const()[name = string("x_451_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_451_mode_0 = const()[name = string("x_451_mode_0"), val = string("constant")];
            fp16 const_90_to_fp16 = const()[name = string("const_90_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_449_cast_fp16 = transpose(perm = x_449_perm_0, x = bd_all_81_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [8, 188, 376]> x_451_cast_fp16 = pad(constant_val = const_90_to_fp16, mode = x_451_mode_0, pad = x_451_pad_0, x = x_449_cast_fp16)[name = string("x_451_cast_fp16")];
            tensor<int32, [3]> var_5404 = const()[name = string("op_5404"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_453_cast_fp16 = reshape(shape = var_5404, x = x_451_cast_fp16)[name = string("x_453_cast_fp16")];
            tensor<int32, [3]> var_5407_begin_0 = const()[name = string("op_5407_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_5407_end_0 = const()[name = string("op_5407_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_5407_end_mask_0 = const()[name = string("op_5407_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_5407_cast_fp16 = slice_by_index(begin = var_5407_begin_0, end = var_5407_end_0, end_mask = var_5407_end_mask_0, x = x_453_cast_fp16)[name = string("op_5407_cast_fp16")];
            tensor<int32, [3]> var_5408 = const()[name = string("op_5408"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_455_cast_fp16 = reshape(shape = var_5408, x = var_5407_cast_fp16)[name = string("x_455_cast_fp16")];
            tensor<int32, [3]> bd_all_83_begin_0 = const()[name = string("bd_all_83_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_83_end_0 = const()[name = string("bd_all_83_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_83_end_mask_0 = const()[name = string("bd_all_83_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_83_cast_fp16 = slice_by_index(begin = bd_all_83_begin_0, end = bd_all_83_end_0, end_mask = bd_all_83_end_mask_0, x = x_455_cast_fp16)[name = string("bd_all_83_cast_fp16")];
            tensor<int32, [4]> var_5413 = const()[name = string("op_5413"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_5414_cast_fp16 = reshape(shape = var_5413, x = q_41_cast_fp16)[name = string("op_5414_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_5416_to_fp16 = const()[name = string("op_5416_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(393557888)))];
            tensor<fp16, [8, 128, 1, 188]> var_5417_cast_fp16 = add(x = var_5414_cast_fp16, y = var_5416_to_fp16)[name = string("op_5417_cast_fp16")];
            tensor<int32, [4]> var_5418 = const()[name = string("op_5418"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_41_cast_fp16 = reshape(shape = var_5418, x = k_41_cast_fp16)[name = string("kh_41_cast_fp16")];
            tensor<int32, [4]> var_5420 = const()[name = string("op_5420"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_41_cast_fp16 = reshape(shape = var_5420, x = v_41_cast_fp16)[name = string("vh_41_cast_fp16")];
            tensor<int32, [4]> var_5422 = const()[name = string("op_5422"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_41_equation_0 = const()[name = string("ac_41_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_5423_cast_fp16 = transpose(perm = var_5422, x = kh_41_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [8, 188, 1, 188]> ac_41_cast_fp16 = einsum(equation = ac_41_equation_0, values = (var_5423_cast_fp16, var_5417_cast_fp16))[name = string("ac_41_cast_fp16")];
            tensor<int32, [3]> var_5426_perm_0 = const()[name = string("op_5426_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_5427_axes_0 = const()[name = string("op_5427_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_5426_cast_fp16 = transpose(perm = var_5426_perm_0, x = bd_all_83_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [8, 188, 1, 188]> var_5427_cast_fp16 = expand_dims(axes = var_5427_axes_0, x = var_5426_cast_fp16)[name = string("op_5427_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_5428_cast_fp16 = add(x = ac_41_cast_fp16, y = var_5427_cast_fp16)[name = string("op_5428_cast_fp16")];
            fp16 var_5429_to_fp16 = const()[name = string("op_5429_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_81_cast_fp16 = mul(x = var_5428_cast_fp16, y = var_5429_to_fp16)[name = string("scores_81_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_83_cast_fp16 = add(x = scores_81_cast_fp16, y = key_bias)[name = string("scores_83_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_5432_cast_fp16 = softmax(axis = var_5303, x = scores_83_cast_fp16)[name = string("op_5432_cast_fp16")];
            tensor<int32, [4]> transpose_68_perm_0 = const()[name = string("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_40_perm_0 = const()[name = string("transpose_40_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_204 = const()[name = string("concat_204"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_40_cast_fp16 = transpose(perm = transpose_40_perm_0, x = var_5432_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [8, 188, 188]> reshape_60_cast_fp16 = reshape(shape = concat_204, x = transpose_40_cast_fp16)[name = string("reshape_60_cast_fp16")];
            tensor<int32, [3]> concat_205 = const()[name = string("concat_205"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_68_cast_fp16 = transpose(perm = transpose_68_perm_0, x = vh_41_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [8, 188, 128]> reshape_61_cast_fp16 = reshape(shape = concat_205, x = transpose_68_cast_fp16)[name = string("reshape_61_cast_fp16")];
            bool matmul_20_transpose_x_0 = const()[name = string("matmul_20_transpose_x_0"), val = bool(false)];
            bool matmul_20_transpose_y_0 = const()[name = string("matmul_20_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_20_cast_fp16 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = reshape_60_cast_fp16, y = reshape_61_cast_fp16)[name = string("matmul_20_cast_fp16")];
            tensor<int32, [4]> concat_209 = const()[name = string("concat_209"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_62_cast_fp16 = reshape(shape = concat_209, x = matmul_20_cast_fp16)[name = string("reshape_62_cast_fp16")];
            tensor<int32, [4]> ctx_41_perm_0 = const()[name = string("ctx_41_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_5437 = const()[name = string("op_5437"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_41_cast_fp16 = transpose(perm = ctx_41_perm_0, x = reshape_62_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 1024, 1, 188]> input_545_cast_fp16 = reshape(shape = var_5437, x = ctx_41_cast_fp16)[name = string("input_545_cast_fp16")];
            string var_5444_pad_type_0 = const()[name = string("op_5444_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5444_strides_0 = const()[name = string("op_5444_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5444_pad_0 = const()[name = string("op_5444_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5444_dilations_0 = const()[name = string("op_5444_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5444_groups_0 = const()[name = string("op_5444_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(393560000))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394346496))))[name = string("layers_20_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5444_cast_fp16 = conv(dilations = var_5444_dilations_0, groups = var_5444_groups_0, pad = var_5444_pad_0, pad_type = var_5444_pad_type_0, strides = var_5444_strides_0, weight = layers_20_self_attn_linear_out_weight_to_fp16_palettized, x = input_545_cast_fp16)[name = string("op_5444_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_457_cast_fp16 = add(x = x_445_cast_fp16, y = var_5444_cast_fp16)[name = string("x_457_cast_fp16")];
            tensor<int32, [1]> var_5460_axes_0 = const()[name = string("op_5460_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_20_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_20_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5460_cast_fp16 = layer_norm(axes = var_5460_axes_0, epsilon = layers_20_norm_conv_eps_scaled_to_fp16, x = x_457_cast_fp16)[name = string("op_5460_cast_fp16")];
            tensor<fp16, [1024]> input_547_gamma_0_to_fp16 = const()[name = string("input_547_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394354752)))];
            tensor<fp16, [1024]> input_547_beta_0_to_fp16 = const()[name = string("input_547_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394356864)))];
            fp16 input_547_epsilon_0_to_fp16 = const()[name = string("input_547_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_547_cast_fp16 = batch_norm(beta = input_547_beta_0_to_fp16, epsilon = input_547_epsilon_0_to_fp16, gamma = input_547_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5460_cast_fp16)[name = string("input_547_cast_fp16")];
            string input_549_pad_type_0 = const()[name = string("input_549_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_549_strides_0 = const()[name = string("input_549_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_549_pad_0 = const()[name = string("input_549_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_549_dilations_0 = const()[name = string("input_549_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_549_groups_0 = const()[name = string("input_549_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_20_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394358976))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395931904))))[name = string("layers_20_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_549_cast_fp16 = conv(dilations = input_549_dilations_0, groups = input_549_groups_0, pad = input_549_pad_0, pad_type = input_549_pad_type_0, strides = input_549_strides_0, weight = layers_20_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_547_cast_fp16)[name = string("input_549_cast_fp16")];
            int32 x_459_split_num_splits_0 = const()[name = string("x_459_split_num_splits_0"), val = int32(2)];
            int32 x_459_split_axis_0 = const()[name = string("x_459_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_459_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_459_split_cast_fp16_1 = split(axis = x_459_split_axis_0, num_splits = x_459_split_num_splits_0, x = input_549_cast_fp16)[name = string("x_459_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_459_split_1_sigmoid_cast_fp16 = sigmoid(x = x_459_split_cast_fp16_1)[name = string("x_459_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_459_cast_fp16 = mul(x = x_459_split_cast_fp16_0, y = x_459_split_1_sigmoid_cast_fp16)[name = string("x_459_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_551_cast_fp16 = mul(x = x_459_cast_fp16, y = pad_mask)[name = string("input_551_cast_fp16")];
            string input_553_pad_type_0 = const()[name = string("input_553_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_553_pad_0 = const()[name = string("input_553_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_553_groups_0 = const()[name = string("input_553_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_553_strides_0 = const()[name = string("input_553_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_553_dilations_0 = const()[name = string("input_553_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_143_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395948352))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395955328))))[name = string("const_143_to_fp16_palettized")];
            tensor<fp16, [1024]> const_144_to_fp16 = const()[name = string("const_144_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395963584)))];
            tensor<fp16, [1, 1024, 1, 188]> input_555_cast_fp16 = conv(bias = const_144_to_fp16, dilations = input_553_dilations_0, groups = input_553_groups_0, pad = input_553_pad_0, pad_type = input_553_pad_type_0, strides = input_553_strides_0, weight = const_143_to_fp16_palettized, x = input_551_cast_fp16)[name = string("input_555_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_557_cast_fp16 = silu(x = input_555_cast_fp16)[name = string("input_557_cast_fp16")];
            string var_5492_pad_type_0 = const()[name = string("op_5492_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5492_strides_0 = const()[name = string("op_5492_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5492_pad_0 = const()[name = string("op_5492_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5492_dilations_0 = const()[name = string("op_5492_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5492_groups_0 = const()[name = string("op_5492_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_20_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395965696))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396752192))))[name = string("layers_20_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5492_cast_fp16 = conv(dilations = var_5492_dilations_0, groups = var_5492_groups_0, pad = var_5492_pad_0, pad_type = var_5492_pad_type_0, strides = var_5492_strides_0, weight = layers_20_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_557_cast_fp16)[name = string("op_5492_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_461_cast_fp16 = add(x = x_457_cast_fp16, y = var_5492_cast_fp16)[name = string("x_461_cast_fp16")];
            tensor<int32, [1]> var_5508_axes_0 = const()[name = string("op_5508_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_20_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_20_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5508_cast_fp16 = layer_norm(axes = var_5508_axes_0, epsilon = layers_20_norm_feed_forward2_eps_scaled_to_fp16, x = x_461_cast_fp16)[name = string("op_5508_cast_fp16")];
            tensor<fp16, [1024]> input_559_gamma_0_to_fp16 = const()[name = string("input_559_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396760448)))];
            tensor<fp16, [1024]> input_559_beta_0_to_fp16 = const()[name = string("input_559_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396762560)))];
            fp16 input_559_epsilon_0_to_fp16 = const()[name = string("input_559_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_559_cast_fp16 = batch_norm(beta = input_559_beta_0_to_fp16, epsilon = input_559_epsilon_0_to_fp16, gamma = input_559_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5508_cast_fp16)[name = string("input_559_cast_fp16")];
            string input_561_pad_type_0 = const()[name = string("input_561_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_561_strides_0 = const()[name = string("input_561_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_561_pad_0 = const()[name = string("input_561_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_561_dilations_0 = const()[name = string("input_561_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_561_groups_0 = const()[name = string("input_561_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_20_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396764672))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399910464))))[name = string("layers_20_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_561_cast_fp16 = conv(dilations = input_561_dilations_0, groups = input_561_groups_0, pad = input_561_pad_0, pad_type = input_561_pad_type_0, strides = input_561_strides_0, weight = layers_20_feed_forward2_linear1_weight_to_fp16_palettized, x = input_559_cast_fp16)[name = string("input_561_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_563_cast_fp16 = silu(x = input_561_cast_fp16)[name = string("input_563_cast_fp16")];
            string var_5525_pad_type_0 = const()[name = string("op_5525_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5525_strides_0 = const()[name = string("op_5525_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5525_pad_0 = const()[name = string("op_5525_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5525_dilations_0 = const()[name = string("op_5525_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5525_groups_0 = const()[name = string("op_5525_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_5526_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399943296))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403089088))))[name = string("op_5526_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5526_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_5525_dilations_0, groups = var_5525_groups_0, pad = var_5525_pad_0, pad_type = var_5525_pad_type_0, strides = var_5525_strides_0, weight = op_5526_weight_0_to_fp16_palettized, x = input_563_cast_fp16)[name = string("op_5526_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_463_cast_fp16 = add(x = x_461_cast_fp16, y = var_5526_cast_fp16)[name = string("x_463_cast_fp16")];
            tensor<int32, [1]> var_5542_axes_0 = const()[name = string("op_5542_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_20_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_20_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5542_cast_fp16 = layer_norm(axes = var_5542_axes_0, epsilon = layers_20_norm_out_eps_scaled_to_fp16, x = x_463_cast_fp16)[name = string("op_5542_cast_fp16")];
            tensor<fp16, [1024]> x_465_gamma_0_to_fp16 = const()[name = string("x_465_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403097344)))];
            tensor<fp16, [1024]> x_465_beta_0_to_fp16 = const()[name = string("x_465_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403099456)))];
            fp16 x_465_epsilon_0_to_fp16 = const()[name = string("x_465_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_465_cast_fp16 = batch_norm(beta = x_465_beta_0_to_fp16, epsilon = x_465_epsilon_0_to_fp16, gamma = x_465_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5542_cast_fp16)[name = string("x_465_cast_fp16")];
            int32 var_5561 = const()[name = string("op_5561"), val = int32(1)];
            tensor<int32, [1]> var_5588_axes_0 = const()[name = string("op_5588_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_21_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_21_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5588_cast_fp16 = layer_norm(axes = var_5588_axes_0, epsilon = layers_21_norm_feed_forward1_eps_scaled_to_fp16, x = x_465_cast_fp16)[name = string("op_5588_cast_fp16")];
            tensor<fp16, [1024]> input_565_gamma_0_to_fp16 = const()[name = string("input_565_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403101568)))];
            tensor<fp16, [1024]> input_565_beta_0_to_fp16 = const()[name = string("input_565_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403103680)))];
            fp16 input_565_epsilon_0_to_fp16 = const()[name = string("input_565_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_565_cast_fp16 = batch_norm(beta = input_565_beta_0_to_fp16, epsilon = input_565_epsilon_0_to_fp16, gamma = input_565_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5588_cast_fp16)[name = string("input_565_cast_fp16")];
            string input_567_pad_type_0 = const()[name = string("input_567_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_567_strides_0 = const()[name = string("input_567_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_567_pad_0 = const()[name = string("input_567_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_567_dilations_0 = const()[name = string("input_567_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_567_groups_0 = const()[name = string("input_567_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_21_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403105792))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406251584))))[name = string("layers_21_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_567_cast_fp16 = conv(dilations = input_567_dilations_0, groups = input_567_groups_0, pad = input_567_pad_0, pad_type = input_567_pad_type_0, strides = input_567_strides_0, weight = layers_21_feed_forward1_linear1_weight_to_fp16_palettized, x = input_565_cast_fp16)[name = string("input_567_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_569_cast_fp16 = silu(x = input_567_cast_fp16)[name = string("input_569_cast_fp16")];
            string var_5605_pad_type_0 = const()[name = string("op_5605_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5605_strides_0 = const()[name = string("op_5605_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5605_pad_0 = const()[name = string("op_5605_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5605_dilations_0 = const()[name = string("op_5605_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5605_groups_0 = const()[name = string("op_5605_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_5606_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406284416))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409430208))))[name = string("op_5606_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5606_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_5605_dilations_0, groups = var_5605_groups_0, pad = var_5605_pad_0, pad_type = var_5605_pad_type_0, strides = var_5605_strides_0, weight = op_5606_weight_0_to_fp16_palettized, x = input_569_cast_fp16)[name = string("op_5606_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_467_cast_fp16 = add(x = x_465_cast_fp16, y = var_5606_cast_fp16)[name = string("x_467_cast_fp16")];
            tensor<int32, [1]> var_5622_axes_0 = const()[name = string("op_5622_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_21_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_21_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5622_cast_fp16 = layer_norm(axes = var_5622_axes_0, epsilon = layers_21_norm_self_att_eps_scaled_to_fp16, x = x_467_cast_fp16)[name = string("op_5622_cast_fp16")];
            tensor<fp16, [1024]> x_469_gamma_0_to_fp16 = const()[name = string("x_469_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409438464)))];
            tensor<fp16, [1024]> x_469_beta_0_to_fp16 = const()[name = string("x_469_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409440576)))];
            fp16 x_469_epsilon_0_to_fp16 = const()[name = string("x_469_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_469_cast_fp16 = batch_norm(beta = x_469_beta_0_to_fp16, epsilon = x_469_epsilon_0_to_fp16, gamma = x_469_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5622_cast_fp16)[name = string("x_469_cast_fp16")];
            string q_43_pad_type_0 = const()[name = string("q_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_43_strides_0 = const()[name = string("q_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_43_pad_0 = const()[name = string("q_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_43_dilations_0 = const()[name = string("q_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_43_groups_0 = const()[name = string("q_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409442688))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410229184))))[name = string("layers_21_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_43_cast_fp16 = conv(dilations = q_43_dilations_0, groups = q_43_groups_0, pad = q_43_pad_0, pad_type = q_43_pad_type_0, strides = q_43_strides_0, weight = layers_21_self_attn_linear_q_weight_to_fp16_palettized, x = x_469_cast_fp16)[name = string("q_43_cast_fp16")];
            string k_43_pad_type_0 = const()[name = string("k_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_43_strides_0 = const()[name = string("k_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_43_pad_0 = const()[name = string("k_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_43_dilations_0 = const()[name = string("k_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_43_groups_0 = const()[name = string("k_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410237440))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411023936))))[name = string("layers_21_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_43_cast_fp16 = conv(dilations = k_43_dilations_0, groups = k_43_groups_0, pad = k_43_pad_0, pad_type = k_43_pad_type_0, strides = k_43_strides_0, weight = layers_21_self_attn_linear_k_weight_to_fp16_palettized, x = x_469_cast_fp16)[name = string("k_43_cast_fp16")];
            string v_43_pad_type_0 = const()[name = string("v_43_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_43_strides_0 = const()[name = string("v_43_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_43_pad_0 = const()[name = string("v_43_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_43_dilations_0 = const()[name = string("v_43_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_43_groups_0 = const()[name = string("v_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411032192))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411818688))))[name = string("layers_21_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_43_cast_fp16 = conv(dilations = v_43_dilations_0, groups = v_43_groups_0, pad = v_43_pad_0, pad_type = v_43_pad_type_0, strides = v_43_strides_0, weight = layers_21_self_attn_linear_v_weight_to_fp16_palettized, x = x_469_cast_fp16)[name = string("v_43_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_43_to_fp16 = const()[name = string("bv_all_43_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411826944)))];
            tensor<fp16, [1, 1024, 1, 188]> var_5654_cast_fp16 = add(x = q_43_cast_fp16, y = bv_all_43_to_fp16)[name = string("op_5654_cast_fp16")];
            tensor<int32, [3]> var_5655 = const()[name = string("op_5655"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_43_cast_fp16 = reshape(shape = var_5655, x = var_5654_cast_fp16)[name = string("qb_43_cast_fp16")];
            bool bd_all_85_transpose_x_0 = const()[name = string("bd_all_85_transpose_x_0"), val = bool(false)];
            bool bd_all_85_transpose_y_0 = const()[name = string("bd_all_85_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_21_self_attn_pos_proj_to_fp16 = const()[name = string("layers_21_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411829056)))];
            tensor<fp16, [8, 375, 188]> bd_all_85_cast_fp16 = matmul(transpose_x = bd_all_85_transpose_x_0, transpose_y = bd_all_85_transpose_y_0, x = layers_21_self_attn_pos_proj_to_fp16, y = qb_43_cast_fp16)[name = string("bd_all_85_cast_fp16")];
            tensor<int32, [3]> x_471_perm_0 = const()[name = string("x_471_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_473_pad_0 = const()[name = string("x_473_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_473_mode_0 = const()[name = string("x_473_mode_0"), val = string("constant")];
            fp16 const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_471_cast_fp16 = transpose(perm = x_471_perm_0, x = bd_all_85_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [8, 188, 376]> x_473_cast_fp16 = pad(constant_val = const_94_to_fp16, mode = x_473_mode_0, pad = x_473_pad_0, x = x_471_cast_fp16)[name = string("x_473_cast_fp16")];
            tensor<int32, [3]> var_5662 = const()[name = string("op_5662"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_475_cast_fp16 = reshape(shape = var_5662, x = x_473_cast_fp16)[name = string("x_475_cast_fp16")];
            tensor<int32, [3]> var_5665_begin_0 = const()[name = string("op_5665_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_5665_end_0 = const()[name = string("op_5665_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_5665_end_mask_0 = const()[name = string("op_5665_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_5665_cast_fp16 = slice_by_index(begin = var_5665_begin_0, end = var_5665_end_0, end_mask = var_5665_end_mask_0, x = x_475_cast_fp16)[name = string("op_5665_cast_fp16")];
            tensor<int32, [3]> var_5666 = const()[name = string("op_5666"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_477_cast_fp16 = reshape(shape = var_5666, x = var_5665_cast_fp16)[name = string("x_477_cast_fp16")];
            tensor<int32, [3]> bd_all_87_begin_0 = const()[name = string("bd_all_87_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_87_end_0 = const()[name = string("bd_all_87_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_87_end_mask_0 = const()[name = string("bd_all_87_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_87_cast_fp16 = slice_by_index(begin = bd_all_87_begin_0, end = bd_all_87_end_0, end_mask = bd_all_87_end_mask_0, x = x_477_cast_fp16)[name = string("bd_all_87_cast_fp16")];
            tensor<int32, [4]> var_5671 = const()[name = string("op_5671"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_5672_cast_fp16 = reshape(shape = var_5671, x = q_43_cast_fp16)[name = string("op_5672_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_5674_to_fp16 = const()[name = string("op_5674_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412597120)))];
            tensor<fp16, [8, 128, 1, 188]> var_5675_cast_fp16 = add(x = var_5672_cast_fp16, y = var_5674_to_fp16)[name = string("op_5675_cast_fp16")];
            tensor<int32, [4]> var_5676 = const()[name = string("op_5676"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_43_cast_fp16 = reshape(shape = var_5676, x = k_43_cast_fp16)[name = string("kh_43_cast_fp16")];
            tensor<int32, [4]> var_5678 = const()[name = string("op_5678"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_43_cast_fp16 = reshape(shape = var_5678, x = v_43_cast_fp16)[name = string("vh_43_cast_fp16")];
            tensor<int32, [4]> var_5680 = const()[name = string("op_5680"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_43_equation_0 = const()[name = string("ac_43_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_5681_cast_fp16 = transpose(perm = var_5680, x = kh_43_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [8, 188, 1, 188]> ac_43_cast_fp16 = einsum(equation = ac_43_equation_0, values = (var_5681_cast_fp16, var_5675_cast_fp16))[name = string("ac_43_cast_fp16")];
            tensor<int32, [3]> var_5684_perm_0 = const()[name = string("op_5684_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_5685_axes_0 = const()[name = string("op_5685_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_5684_cast_fp16 = transpose(perm = var_5684_perm_0, x = bd_all_87_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [8, 188, 1, 188]> var_5685_cast_fp16 = expand_dims(axes = var_5685_axes_0, x = var_5684_cast_fp16)[name = string("op_5685_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_5686_cast_fp16 = add(x = ac_43_cast_fp16, y = var_5685_cast_fp16)[name = string("op_5686_cast_fp16")];
            fp16 var_5687_to_fp16 = const()[name = string("op_5687_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_85_cast_fp16 = mul(x = var_5686_cast_fp16, y = var_5687_to_fp16)[name = string("scores_85_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_87_cast_fp16 = add(x = scores_85_cast_fp16, y = key_bias)[name = string("scores_87_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_5690_cast_fp16 = softmax(axis = var_5561, x = scores_87_cast_fp16)[name = string("op_5690_cast_fp16")];
            tensor<int32, [4]> transpose_69_perm_0 = const()[name = string("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_42_perm_0 = const()[name = string("transpose_42_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_214 = const()[name = string("concat_214"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_42_cast_fp16 = transpose(perm = transpose_42_perm_0, x = var_5690_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [8, 188, 188]> reshape_63_cast_fp16 = reshape(shape = concat_214, x = transpose_42_cast_fp16)[name = string("reshape_63_cast_fp16")];
            tensor<int32, [3]> concat_215 = const()[name = string("concat_215"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_69_cast_fp16 = transpose(perm = transpose_69_perm_0, x = vh_43_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [8, 188, 128]> reshape_64_cast_fp16 = reshape(shape = concat_215, x = transpose_69_cast_fp16)[name = string("reshape_64_cast_fp16")];
            bool matmul_21_transpose_x_0 = const()[name = string("matmul_21_transpose_x_0"), val = bool(false)];
            bool matmul_21_transpose_y_0 = const()[name = string("matmul_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_21_cast_fp16 = matmul(transpose_x = matmul_21_transpose_x_0, transpose_y = matmul_21_transpose_y_0, x = reshape_63_cast_fp16, y = reshape_64_cast_fp16)[name = string("matmul_21_cast_fp16")];
            tensor<int32, [4]> concat_219 = const()[name = string("concat_219"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_65_cast_fp16 = reshape(shape = concat_219, x = matmul_21_cast_fp16)[name = string("reshape_65_cast_fp16")];
            tensor<int32, [4]> ctx_43_perm_0 = const()[name = string("ctx_43_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_5695 = const()[name = string("op_5695"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_43_cast_fp16 = transpose(perm = ctx_43_perm_0, x = reshape_65_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 1024, 1, 188]> input_571_cast_fp16 = reshape(shape = var_5695, x = ctx_43_cast_fp16)[name = string("input_571_cast_fp16")];
            string var_5702_pad_type_0 = const()[name = string("op_5702_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5702_strides_0 = const()[name = string("op_5702_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5702_pad_0 = const()[name = string("op_5702_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5702_dilations_0 = const()[name = string("op_5702_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5702_groups_0 = const()[name = string("op_5702_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412599232))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413385728))))[name = string("layers_21_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5702_cast_fp16 = conv(dilations = var_5702_dilations_0, groups = var_5702_groups_0, pad = var_5702_pad_0, pad_type = var_5702_pad_type_0, strides = var_5702_strides_0, weight = layers_21_self_attn_linear_out_weight_to_fp16_palettized, x = input_571_cast_fp16)[name = string("op_5702_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_479_cast_fp16 = add(x = x_467_cast_fp16, y = var_5702_cast_fp16)[name = string("x_479_cast_fp16")];
            tensor<int32, [1]> var_5718_axes_0 = const()[name = string("op_5718_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_21_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_21_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5718_cast_fp16 = layer_norm(axes = var_5718_axes_0, epsilon = layers_21_norm_conv_eps_scaled_to_fp16, x = x_479_cast_fp16)[name = string("op_5718_cast_fp16")];
            tensor<fp16, [1024]> input_573_gamma_0_to_fp16 = const()[name = string("input_573_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413393984)))];
            tensor<fp16, [1024]> input_573_beta_0_to_fp16 = const()[name = string("input_573_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413396096)))];
            fp16 input_573_epsilon_0_to_fp16 = const()[name = string("input_573_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_573_cast_fp16 = batch_norm(beta = input_573_beta_0_to_fp16, epsilon = input_573_epsilon_0_to_fp16, gamma = input_573_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5718_cast_fp16)[name = string("input_573_cast_fp16")];
            string input_575_pad_type_0 = const()[name = string("input_575_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_575_strides_0 = const()[name = string("input_575_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_575_pad_0 = const()[name = string("input_575_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_575_dilations_0 = const()[name = string("input_575_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_575_groups_0 = const()[name = string("input_575_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_21_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413398208))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414971136))))[name = string("layers_21_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_575_cast_fp16 = conv(dilations = input_575_dilations_0, groups = input_575_groups_0, pad = input_575_pad_0, pad_type = input_575_pad_type_0, strides = input_575_strides_0, weight = layers_21_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_573_cast_fp16)[name = string("input_575_cast_fp16")];
            int32 x_481_split_num_splits_0 = const()[name = string("x_481_split_num_splits_0"), val = int32(2)];
            int32 x_481_split_axis_0 = const()[name = string("x_481_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_481_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_481_split_cast_fp16_1 = split(axis = x_481_split_axis_0, num_splits = x_481_split_num_splits_0, x = input_575_cast_fp16)[name = string("x_481_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_481_split_1_sigmoid_cast_fp16 = sigmoid(x = x_481_split_cast_fp16_1)[name = string("x_481_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_481_cast_fp16 = mul(x = x_481_split_cast_fp16_0, y = x_481_split_1_sigmoid_cast_fp16)[name = string("x_481_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_577_cast_fp16 = mul(x = x_481_cast_fp16, y = pad_mask)[name = string("input_577_cast_fp16")];
            string input_579_pad_type_0 = const()[name = string("input_579_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_579_pad_0 = const()[name = string("input_579_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_579_groups_0 = const()[name = string("input_579_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_579_strides_0 = const()[name = string("input_579_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_579_dilations_0 = const()[name = string("input_579_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_145_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414987584))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(414994560))))[name = string("const_145_to_fp16_palettized")];
            tensor<fp16, [1024]> const_146_to_fp16 = const()[name = string("const_146_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415002816)))];
            tensor<fp16, [1, 1024, 1, 188]> input_581_cast_fp16 = conv(bias = const_146_to_fp16, dilations = input_579_dilations_0, groups = input_579_groups_0, pad = input_579_pad_0, pad_type = input_579_pad_type_0, strides = input_579_strides_0, weight = const_145_to_fp16_palettized, x = input_577_cast_fp16)[name = string("input_581_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_583_cast_fp16 = silu(x = input_581_cast_fp16)[name = string("input_583_cast_fp16")];
            string var_5750_pad_type_0 = const()[name = string("op_5750_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5750_strides_0 = const()[name = string("op_5750_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5750_pad_0 = const()[name = string("op_5750_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5750_dilations_0 = const()[name = string("op_5750_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5750_groups_0 = const()[name = string("op_5750_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_21_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415004928))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415791424))))[name = string("layers_21_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5750_cast_fp16 = conv(dilations = var_5750_dilations_0, groups = var_5750_groups_0, pad = var_5750_pad_0, pad_type = var_5750_pad_type_0, strides = var_5750_strides_0, weight = layers_21_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_583_cast_fp16)[name = string("op_5750_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_483_cast_fp16 = add(x = x_479_cast_fp16, y = var_5750_cast_fp16)[name = string("x_483_cast_fp16")];
            tensor<int32, [1]> var_5766_axes_0 = const()[name = string("op_5766_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_21_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_21_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5766_cast_fp16 = layer_norm(axes = var_5766_axes_0, epsilon = layers_21_norm_feed_forward2_eps_scaled_to_fp16, x = x_483_cast_fp16)[name = string("op_5766_cast_fp16")];
            tensor<fp16, [1024]> input_585_gamma_0_to_fp16 = const()[name = string("input_585_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415799680)))];
            tensor<fp16, [1024]> input_585_beta_0_to_fp16 = const()[name = string("input_585_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415801792)))];
            fp16 input_585_epsilon_0_to_fp16 = const()[name = string("input_585_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_585_cast_fp16 = batch_norm(beta = input_585_beta_0_to_fp16, epsilon = input_585_epsilon_0_to_fp16, gamma = input_585_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5766_cast_fp16)[name = string("input_585_cast_fp16")];
            string input_587_pad_type_0 = const()[name = string("input_587_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_587_strides_0 = const()[name = string("input_587_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_587_pad_0 = const()[name = string("input_587_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_587_dilations_0 = const()[name = string("input_587_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_587_groups_0 = const()[name = string("input_587_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_21_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415803904))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418949696))))[name = string("layers_21_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_587_cast_fp16 = conv(dilations = input_587_dilations_0, groups = input_587_groups_0, pad = input_587_pad_0, pad_type = input_587_pad_type_0, strides = input_587_strides_0, weight = layers_21_feed_forward2_linear1_weight_to_fp16_palettized, x = input_585_cast_fp16)[name = string("input_587_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_589_cast_fp16 = silu(x = input_587_cast_fp16)[name = string("input_589_cast_fp16")];
            string var_5783_pad_type_0 = const()[name = string("op_5783_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5783_strides_0 = const()[name = string("op_5783_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5783_pad_0 = const()[name = string("op_5783_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5783_dilations_0 = const()[name = string("op_5783_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5783_groups_0 = const()[name = string("op_5783_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_5784_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(418982528))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422128320))))[name = string("op_5784_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5784_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_5783_dilations_0, groups = var_5783_groups_0, pad = var_5783_pad_0, pad_type = var_5783_pad_type_0, strides = var_5783_strides_0, weight = op_5784_weight_0_to_fp16_palettized, x = input_589_cast_fp16)[name = string("op_5784_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_485_cast_fp16 = add(x = x_483_cast_fp16, y = var_5784_cast_fp16)[name = string("x_485_cast_fp16")];
            tensor<int32, [1]> var_5800_axes_0 = const()[name = string("op_5800_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_21_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_21_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5800_cast_fp16 = layer_norm(axes = var_5800_axes_0, epsilon = layers_21_norm_out_eps_scaled_to_fp16, x = x_485_cast_fp16)[name = string("op_5800_cast_fp16")];
            tensor<fp16, [1024]> x_487_gamma_0_to_fp16 = const()[name = string("x_487_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422136576)))];
            tensor<fp16, [1024]> x_487_beta_0_to_fp16 = const()[name = string("x_487_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422138688)))];
            fp16 x_487_epsilon_0_to_fp16 = const()[name = string("x_487_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_487_cast_fp16 = batch_norm(beta = x_487_beta_0_to_fp16, epsilon = x_487_epsilon_0_to_fp16, gamma = x_487_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5800_cast_fp16)[name = string("x_487_cast_fp16")];
            int32 var_5819 = const()[name = string("op_5819"), val = int32(1)];
            tensor<int32, [1]> var_5846_axes_0 = const()[name = string("op_5846_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_22_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_22_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5846_cast_fp16 = layer_norm(axes = var_5846_axes_0, epsilon = layers_22_norm_feed_forward1_eps_scaled_to_fp16, x = x_487_cast_fp16)[name = string("op_5846_cast_fp16")];
            tensor<fp16, [1024]> input_591_gamma_0_to_fp16 = const()[name = string("input_591_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422140800)))];
            tensor<fp16, [1024]> input_591_beta_0_to_fp16 = const()[name = string("input_591_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422142912)))];
            fp16 input_591_epsilon_0_to_fp16 = const()[name = string("input_591_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_591_cast_fp16 = batch_norm(beta = input_591_beta_0_to_fp16, epsilon = input_591_epsilon_0_to_fp16, gamma = input_591_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5846_cast_fp16)[name = string("input_591_cast_fp16")];
            string input_593_pad_type_0 = const()[name = string("input_593_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_593_strides_0 = const()[name = string("input_593_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_593_pad_0 = const()[name = string("input_593_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_593_dilations_0 = const()[name = string("input_593_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_593_groups_0 = const()[name = string("input_593_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_22_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422145024))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(425290816))))[name = string("layers_22_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_593_cast_fp16 = conv(dilations = input_593_dilations_0, groups = input_593_groups_0, pad = input_593_pad_0, pad_type = input_593_pad_type_0, strides = input_593_strides_0, weight = layers_22_feed_forward1_linear1_weight_to_fp16_palettized, x = input_591_cast_fp16)[name = string("input_593_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_595_cast_fp16 = silu(x = input_593_cast_fp16)[name = string("input_595_cast_fp16")];
            string var_5863_pad_type_0 = const()[name = string("op_5863_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5863_strides_0 = const()[name = string("op_5863_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5863_pad_0 = const()[name = string("op_5863_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5863_dilations_0 = const()[name = string("op_5863_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5863_groups_0 = const()[name = string("op_5863_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_5864_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(425323648))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428469440))))[name = string("op_5864_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5864_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_5863_dilations_0, groups = var_5863_groups_0, pad = var_5863_pad_0, pad_type = var_5863_pad_type_0, strides = var_5863_strides_0, weight = op_5864_weight_0_to_fp16_palettized, x = input_595_cast_fp16)[name = string("op_5864_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_489_cast_fp16 = add(x = x_487_cast_fp16, y = var_5864_cast_fp16)[name = string("x_489_cast_fp16")];
            tensor<int32, [1]> var_5880_axes_0 = const()[name = string("op_5880_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_22_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_22_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5880_cast_fp16 = layer_norm(axes = var_5880_axes_0, epsilon = layers_22_norm_self_att_eps_scaled_to_fp16, x = x_489_cast_fp16)[name = string("op_5880_cast_fp16")];
            tensor<fp16, [1024]> x_491_gamma_0_to_fp16 = const()[name = string("x_491_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428477696)))];
            tensor<fp16, [1024]> x_491_beta_0_to_fp16 = const()[name = string("x_491_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428479808)))];
            fp16 x_491_epsilon_0_to_fp16 = const()[name = string("x_491_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_491_cast_fp16 = batch_norm(beta = x_491_beta_0_to_fp16, epsilon = x_491_epsilon_0_to_fp16, gamma = x_491_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5880_cast_fp16)[name = string("x_491_cast_fp16")];
            string q_45_pad_type_0 = const()[name = string("q_45_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_45_strides_0 = const()[name = string("q_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_45_pad_0 = const()[name = string("q_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_45_dilations_0 = const()[name = string("q_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_45_groups_0 = const()[name = string("q_45_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428481920))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429268416))))[name = string("layers_22_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_45_cast_fp16 = conv(dilations = q_45_dilations_0, groups = q_45_groups_0, pad = q_45_pad_0, pad_type = q_45_pad_type_0, strides = q_45_strides_0, weight = layers_22_self_attn_linear_q_weight_to_fp16_palettized, x = x_491_cast_fp16)[name = string("q_45_cast_fp16")];
            string k_45_pad_type_0 = const()[name = string("k_45_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_45_strides_0 = const()[name = string("k_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_45_pad_0 = const()[name = string("k_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_45_dilations_0 = const()[name = string("k_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_45_groups_0 = const()[name = string("k_45_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429276672))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430063168))))[name = string("layers_22_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_45_cast_fp16 = conv(dilations = k_45_dilations_0, groups = k_45_groups_0, pad = k_45_pad_0, pad_type = k_45_pad_type_0, strides = k_45_strides_0, weight = layers_22_self_attn_linear_k_weight_to_fp16_palettized, x = x_491_cast_fp16)[name = string("k_45_cast_fp16")];
            string v_45_pad_type_0 = const()[name = string("v_45_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_45_strides_0 = const()[name = string("v_45_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_45_pad_0 = const()[name = string("v_45_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_45_dilations_0 = const()[name = string("v_45_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_45_groups_0 = const()[name = string("v_45_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430071424))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430857920))))[name = string("layers_22_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_45_cast_fp16 = conv(dilations = v_45_dilations_0, groups = v_45_groups_0, pad = v_45_pad_0, pad_type = v_45_pad_type_0, strides = v_45_strides_0, weight = layers_22_self_attn_linear_v_weight_to_fp16_palettized, x = x_491_cast_fp16)[name = string("v_45_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_45_to_fp16 = const()[name = string("bv_all_45_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430866176)))];
            tensor<fp16, [1, 1024, 1, 188]> var_5912_cast_fp16 = add(x = q_45_cast_fp16, y = bv_all_45_to_fp16)[name = string("op_5912_cast_fp16")];
            tensor<int32, [3]> var_5913 = const()[name = string("op_5913"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_45_cast_fp16 = reshape(shape = var_5913, x = var_5912_cast_fp16)[name = string("qb_45_cast_fp16")];
            bool bd_all_89_transpose_x_0 = const()[name = string("bd_all_89_transpose_x_0"), val = bool(false)];
            bool bd_all_89_transpose_y_0 = const()[name = string("bd_all_89_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_22_self_attn_pos_proj_to_fp16 = const()[name = string("layers_22_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430868288)))];
            tensor<fp16, [8, 375, 188]> bd_all_89_cast_fp16 = matmul(transpose_x = bd_all_89_transpose_x_0, transpose_y = bd_all_89_transpose_y_0, x = layers_22_self_attn_pos_proj_to_fp16, y = qb_45_cast_fp16)[name = string("bd_all_89_cast_fp16")];
            tensor<int32, [3]> x_493_perm_0 = const()[name = string("x_493_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_495_pad_0 = const()[name = string("x_495_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_495_mode_0 = const()[name = string("x_495_mode_0"), val = string("constant")];
            fp16 const_98_to_fp16 = const()[name = string("const_98_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_493_cast_fp16 = transpose(perm = x_493_perm_0, x = bd_all_89_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [8, 188, 376]> x_495_cast_fp16 = pad(constant_val = const_98_to_fp16, mode = x_495_mode_0, pad = x_495_pad_0, x = x_493_cast_fp16)[name = string("x_495_cast_fp16")];
            tensor<int32, [3]> var_5920 = const()[name = string("op_5920"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_497_cast_fp16 = reshape(shape = var_5920, x = x_495_cast_fp16)[name = string("x_497_cast_fp16")];
            tensor<int32, [3]> var_5923_begin_0 = const()[name = string("op_5923_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_5923_end_0 = const()[name = string("op_5923_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_5923_end_mask_0 = const()[name = string("op_5923_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_5923_cast_fp16 = slice_by_index(begin = var_5923_begin_0, end = var_5923_end_0, end_mask = var_5923_end_mask_0, x = x_497_cast_fp16)[name = string("op_5923_cast_fp16")];
            tensor<int32, [3]> var_5924 = const()[name = string("op_5924"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_499_cast_fp16 = reshape(shape = var_5924, x = var_5923_cast_fp16)[name = string("x_499_cast_fp16")];
            tensor<int32, [3]> bd_all_91_begin_0 = const()[name = string("bd_all_91_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_91_end_0 = const()[name = string("bd_all_91_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_91_end_mask_0 = const()[name = string("bd_all_91_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_91_cast_fp16 = slice_by_index(begin = bd_all_91_begin_0, end = bd_all_91_end_0, end_mask = bd_all_91_end_mask_0, x = x_499_cast_fp16)[name = string("bd_all_91_cast_fp16")];
            tensor<int32, [4]> var_5929 = const()[name = string("op_5929"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_5930_cast_fp16 = reshape(shape = var_5929, x = q_45_cast_fp16)[name = string("op_5930_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_5932_to_fp16 = const()[name = string("op_5932_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431636352)))];
            tensor<fp16, [8, 128, 1, 188]> var_5933_cast_fp16 = add(x = var_5930_cast_fp16, y = var_5932_to_fp16)[name = string("op_5933_cast_fp16")];
            tensor<int32, [4]> var_5934 = const()[name = string("op_5934"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_45_cast_fp16 = reshape(shape = var_5934, x = k_45_cast_fp16)[name = string("kh_45_cast_fp16")];
            tensor<int32, [4]> var_5936 = const()[name = string("op_5936"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_45_cast_fp16 = reshape(shape = var_5936, x = v_45_cast_fp16)[name = string("vh_45_cast_fp16")];
            tensor<int32, [4]> var_5938 = const()[name = string("op_5938"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_45_equation_0 = const()[name = string("ac_45_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_5939_cast_fp16 = transpose(perm = var_5938, x = kh_45_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [8, 188, 1, 188]> ac_45_cast_fp16 = einsum(equation = ac_45_equation_0, values = (var_5939_cast_fp16, var_5933_cast_fp16))[name = string("ac_45_cast_fp16")];
            tensor<int32, [3]> var_5942_perm_0 = const()[name = string("op_5942_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_5943_axes_0 = const()[name = string("op_5943_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_5942_cast_fp16 = transpose(perm = var_5942_perm_0, x = bd_all_91_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [8, 188, 1, 188]> var_5943_cast_fp16 = expand_dims(axes = var_5943_axes_0, x = var_5942_cast_fp16)[name = string("op_5943_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_5944_cast_fp16 = add(x = ac_45_cast_fp16, y = var_5943_cast_fp16)[name = string("op_5944_cast_fp16")];
            fp16 var_5945_to_fp16 = const()[name = string("op_5945_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_89_cast_fp16 = mul(x = var_5944_cast_fp16, y = var_5945_to_fp16)[name = string("scores_89_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_91_cast_fp16 = add(x = scores_89_cast_fp16, y = key_bias)[name = string("scores_91_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_5948_cast_fp16 = softmax(axis = var_5819, x = scores_91_cast_fp16)[name = string("op_5948_cast_fp16")];
            tensor<int32, [4]> transpose_70_perm_0 = const()[name = string("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_44_perm_0 = const()[name = string("transpose_44_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_224 = const()[name = string("concat_224"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_44_cast_fp16 = transpose(perm = transpose_44_perm_0, x = var_5948_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [8, 188, 188]> reshape_66_cast_fp16 = reshape(shape = concat_224, x = transpose_44_cast_fp16)[name = string("reshape_66_cast_fp16")];
            tensor<int32, [3]> concat_225 = const()[name = string("concat_225"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_70_cast_fp16 = transpose(perm = transpose_70_perm_0, x = vh_45_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [8, 188, 128]> reshape_67_cast_fp16 = reshape(shape = concat_225, x = transpose_70_cast_fp16)[name = string("reshape_67_cast_fp16")];
            bool matmul_22_transpose_x_0 = const()[name = string("matmul_22_transpose_x_0"), val = bool(false)];
            bool matmul_22_transpose_y_0 = const()[name = string("matmul_22_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_22_cast_fp16 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = reshape_66_cast_fp16, y = reshape_67_cast_fp16)[name = string("matmul_22_cast_fp16")];
            tensor<int32, [4]> concat_229 = const()[name = string("concat_229"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_68_cast_fp16 = reshape(shape = concat_229, x = matmul_22_cast_fp16)[name = string("reshape_68_cast_fp16")];
            tensor<int32, [4]> ctx_45_perm_0 = const()[name = string("ctx_45_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_5953 = const()[name = string("op_5953"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_45_cast_fp16 = transpose(perm = ctx_45_perm_0, x = reshape_68_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 1024, 1, 188]> input_597_cast_fp16 = reshape(shape = var_5953, x = ctx_45_cast_fp16)[name = string("input_597_cast_fp16")];
            string var_5960_pad_type_0 = const()[name = string("op_5960_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_5960_strides_0 = const()[name = string("op_5960_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_5960_pad_0 = const()[name = string("op_5960_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_5960_dilations_0 = const()[name = string("op_5960_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_5960_groups_0 = const()[name = string("op_5960_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431638464))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(432424960))))[name = string("layers_22_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_5960_cast_fp16 = conv(dilations = var_5960_dilations_0, groups = var_5960_groups_0, pad = var_5960_pad_0, pad_type = var_5960_pad_type_0, strides = var_5960_strides_0, weight = layers_22_self_attn_linear_out_weight_to_fp16_palettized, x = input_597_cast_fp16)[name = string("op_5960_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_501_cast_fp16 = add(x = x_489_cast_fp16, y = var_5960_cast_fp16)[name = string("x_501_cast_fp16")];
            tensor<int32, [1]> var_5976_axes_0 = const()[name = string("op_5976_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_22_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_22_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_5976_cast_fp16 = layer_norm(axes = var_5976_axes_0, epsilon = layers_22_norm_conv_eps_scaled_to_fp16, x = x_501_cast_fp16)[name = string("op_5976_cast_fp16")];
            tensor<fp16, [1024]> input_599_gamma_0_to_fp16 = const()[name = string("input_599_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(432433216)))];
            tensor<fp16, [1024]> input_599_beta_0_to_fp16 = const()[name = string("input_599_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(432435328)))];
            fp16 input_599_epsilon_0_to_fp16 = const()[name = string("input_599_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_599_cast_fp16 = batch_norm(beta = input_599_beta_0_to_fp16, epsilon = input_599_epsilon_0_to_fp16, gamma = input_599_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_5976_cast_fp16)[name = string("input_599_cast_fp16")];
            string input_601_pad_type_0 = const()[name = string("input_601_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_601_strides_0 = const()[name = string("input_601_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_601_pad_0 = const()[name = string("input_601_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_601_dilations_0 = const()[name = string("input_601_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_601_groups_0 = const()[name = string("input_601_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_22_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(432437440))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434010368))))[name = string("layers_22_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_601_cast_fp16 = conv(dilations = input_601_dilations_0, groups = input_601_groups_0, pad = input_601_pad_0, pad_type = input_601_pad_type_0, strides = input_601_strides_0, weight = layers_22_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_599_cast_fp16)[name = string("input_601_cast_fp16")];
            int32 x_503_split_num_splits_0 = const()[name = string("x_503_split_num_splits_0"), val = int32(2)];
            int32 x_503_split_axis_0 = const()[name = string("x_503_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_503_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_503_split_cast_fp16_1 = split(axis = x_503_split_axis_0, num_splits = x_503_split_num_splits_0, x = input_601_cast_fp16)[name = string("x_503_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_503_split_1_sigmoid_cast_fp16 = sigmoid(x = x_503_split_cast_fp16_1)[name = string("x_503_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_503_cast_fp16 = mul(x = x_503_split_cast_fp16_0, y = x_503_split_1_sigmoid_cast_fp16)[name = string("x_503_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_603_cast_fp16 = mul(x = x_503_cast_fp16, y = pad_mask)[name = string("input_603_cast_fp16")];
            string input_605_pad_type_0 = const()[name = string("input_605_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_605_pad_0 = const()[name = string("input_605_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_605_groups_0 = const()[name = string("input_605_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_605_strides_0 = const()[name = string("input_605_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_605_dilations_0 = const()[name = string("input_605_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_147_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434026816))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434033792))))[name = string("const_147_to_fp16_palettized")];
            tensor<fp16, [1024]> const_148_to_fp16 = const()[name = string("const_148_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434042048)))];
            tensor<fp16, [1, 1024, 1, 188]> input_607_cast_fp16 = conv(bias = const_148_to_fp16, dilations = input_605_dilations_0, groups = input_605_groups_0, pad = input_605_pad_0, pad_type = input_605_pad_type_0, strides = input_605_strides_0, weight = const_147_to_fp16_palettized, x = input_603_cast_fp16)[name = string("input_607_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_609_cast_fp16 = silu(x = input_607_cast_fp16)[name = string("input_609_cast_fp16")];
            string var_6008_pad_type_0 = const()[name = string("op_6008_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_6008_strides_0 = const()[name = string("op_6008_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_6008_pad_0 = const()[name = string("op_6008_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_6008_dilations_0 = const()[name = string("op_6008_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_6008_groups_0 = const()[name = string("op_6008_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_22_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434044160))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434830656))))[name = string("layers_22_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_6008_cast_fp16 = conv(dilations = var_6008_dilations_0, groups = var_6008_groups_0, pad = var_6008_pad_0, pad_type = var_6008_pad_type_0, strides = var_6008_strides_0, weight = layers_22_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_609_cast_fp16)[name = string("op_6008_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_505_cast_fp16 = add(x = x_501_cast_fp16, y = var_6008_cast_fp16)[name = string("x_505_cast_fp16")];
            tensor<int32, [1]> var_6024_axes_0 = const()[name = string("op_6024_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_22_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_22_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_6024_cast_fp16 = layer_norm(axes = var_6024_axes_0, epsilon = layers_22_norm_feed_forward2_eps_scaled_to_fp16, x = x_505_cast_fp16)[name = string("op_6024_cast_fp16")];
            tensor<fp16, [1024]> input_611_gamma_0_to_fp16 = const()[name = string("input_611_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434838912)))];
            tensor<fp16, [1024]> input_611_beta_0_to_fp16 = const()[name = string("input_611_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434841024)))];
            fp16 input_611_epsilon_0_to_fp16 = const()[name = string("input_611_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_611_cast_fp16 = batch_norm(beta = input_611_beta_0_to_fp16, epsilon = input_611_epsilon_0_to_fp16, gamma = input_611_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_6024_cast_fp16)[name = string("input_611_cast_fp16")];
            string input_613_pad_type_0 = const()[name = string("input_613_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_613_strides_0 = const()[name = string("input_613_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_613_pad_0 = const()[name = string("input_613_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_613_dilations_0 = const()[name = string("input_613_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_613_groups_0 = const()[name = string("input_613_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_22_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(434843136))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437988928))))[name = string("layers_22_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_613_cast_fp16 = conv(dilations = input_613_dilations_0, groups = input_613_groups_0, pad = input_613_pad_0, pad_type = input_613_pad_type_0, strides = input_613_strides_0, weight = layers_22_feed_forward2_linear1_weight_to_fp16_palettized, x = input_611_cast_fp16)[name = string("input_613_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_615_cast_fp16 = silu(x = input_613_cast_fp16)[name = string("input_615_cast_fp16")];
            string var_6041_pad_type_0 = const()[name = string("op_6041_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_6041_strides_0 = const()[name = string("op_6041_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_6041_pad_0 = const()[name = string("op_6041_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_6041_dilations_0 = const()[name = string("op_6041_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_6041_groups_0 = const()[name = string("op_6041_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_6042_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438021760))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441167552))))[name = string("op_6042_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_6042_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_6041_dilations_0, groups = var_6041_groups_0, pad = var_6041_pad_0, pad_type = var_6041_pad_type_0, strides = var_6041_strides_0, weight = op_6042_weight_0_to_fp16_palettized, x = input_615_cast_fp16)[name = string("op_6042_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_507_cast_fp16 = add(x = x_505_cast_fp16, y = var_6042_cast_fp16)[name = string("x_507_cast_fp16")];
            tensor<int32, [1]> var_6058_axes_0 = const()[name = string("op_6058_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_22_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_22_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_6058_cast_fp16 = layer_norm(axes = var_6058_axes_0, epsilon = layers_22_norm_out_eps_scaled_to_fp16, x = x_507_cast_fp16)[name = string("op_6058_cast_fp16")];
            tensor<fp16, [1024]> x_509_gamma_0_to_fp16 = const()[name = string("x_509_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441175808)))];
            tensor<fp16, [1024]> x_509_beta_0_to_fp16 = const()[name = string("x_509_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441177920)))];
            fp16 x_509_epsilon_0_to_fp16 = const()[name = string("x_509_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_509_cast_fp16 = batch_norm(beta = x_509_beta_0_to_fp16, epsilon = x_509_epsilon_0_to_fp16, gamma = x_509_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_6058_cast_fp16)[name = string("x_509_cast_fp16")];
            int32 var_6077 = const()[name = string("op_6077"), val = int32(1)];
            tensor<int32, [1]> var_6104_axes_0 = const()[name = string("op_6104_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_23_norm_feed_forward1_eps_scaled_to_fp16 = const()[name = string("layers_23_norm_feed_forward1_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_6104_cast_fp16 = layer_norm(axes = var_6104_axes_0, epsilon = layers_23_norm_feed_forward1_eps_scaled_to_fp16, x = x_509_cast_fp16)[name = string("op_6104_cast_fp16")];
            tensor<fp16, [1024]> input_617_gamma_0_to_fp16 = const()[name = string("input_617_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441180032)))];
            tensor<fp16, [1024]> input_617_beta_0_to_fp16 = const()[name = string("input_617_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441182144)))];
            fp16 input_617_epsilon_0_to_fp16 = const()[name = string("input_617_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_617_cast_fp16 = batch_norm(beta = input_617_beta_0_to_fp16, epsilon = input_617_epsilon_0_to_fp16, gamma = input_617_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_6104_cast_fp16)[name = string("input_617_cast_fp16")];
            string input_619_pad_type_0 = const()[name = string("input_619_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_619_strides_0 = const()[name = string("input_619_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_619_pad_0 = const()[name = string("input_619_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_619_dilations_0 = const()[name = string("input_619_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_619_groups_0 = const()[name = string("input_619_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_23_feed_forward1_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(441184256))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444330048))))[name = string("layers_23_feed_forward1_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_619_cast_fp16 = conv(dilations = input_619_dilations_0, groups = input_619_groups_0, pad = input_619_pad_0, pad_type = input_619_pad_type_0, strides = input_619_strides_0, weight = layers_23_feed_forward1_linear1_weight_to_fp16_palettized, x = input_617_cast_fp16)[name = string("input_619_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_621_cast_fp16 = silu(x = input_619_cast_fp16)[name = string("input_621_cast_fp16")];
            string var_6121_pad_type_0 = const()[name = string("op_6121_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_6121_strides_0 = const()[name = string("op_6121_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_6121_pad_0 = const()[name = string("op_6121_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_6121_dilations_0 = const()[name = string("op_6121_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_6121_groups_0 = const()[name = string("op_6121_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_6122_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444362880))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447508672))))[name = string("op_6122_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_6122_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_6121_dilations_0, groups = var_6121_groups_0, pad = var_6121_pad_0, pad_type = var_6121_pad_type_0, strides = var_6121_strides_0, weight = op_6122_weight_0_to_fp16_palettized, x = input_621_cast_fp16)[name = string("op_6122_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_511_cast_fp16 = add(x = x_509_cast_fp16, y = var_6122_cast_fp16)[name = string("x_511_cast_fp16")];
            tensor<int32, [1]> var_6138_axes_0 = const()[name = string("op_6138_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_23_norm_self_att_eps_scaled_to_fp16 = const()[name = string("layers_23_norm_self_att_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_6138_cast_fp16 = layer_norm(axes = var_6138_axes_0, epsilon = layers_23_norm_self_att_eps_scaled_to_fp16, x = x_511_cast_fp16)[name = string("op_6138_cast_fp16")];
            tensor<fp16, [1024]> x_513_gamma_0_to_fp16 = const()[name = string("x_513_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447516928)))];
            tensor<fp16, [1024]> x_513_beta_0_to_fp16 = const()[name = string("x_513_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447519040)))];
            fp16 x_513_epsilon_0_to_fp16 = const()[name = string("x_513_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> x_513_cast_fp16 = batch_norm(beta = x_513_beta_0_to_fp16, epsilon = x_513_epsilon_0_to_fp16, gamma = x_513_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_6138_cast_fp16)[name = string("x_513_cast_fp16")];
            string q_pad_type_0 = const()[name = string("q_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> q_strides_0 = const()[name = string("q_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> q_pad_0 = const()[name = string("q_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> q_dilations_0 = const()[name = string("q_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 q_groups_0 = const()[name = string("q_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_linear_q_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(447521152))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(448307648))))[name = string("layers_23_self_attn_linear_q_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> q_cast_fp16 = conv(dilations = q_dilations_0, groups = q_groups_0, pad = q_pad_0, pad_type = q_pad_type_0, strides = q_strides_0, weight = layers_23_self_attn_linear_q_weight_to_fp16_palettized, x = x_513_cast_fp16)[name = string("q_cast_fp16")];
            string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_strides_0 = const()[name = string("k_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_pad_0 = const()[name = string("k_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_linear_k_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(448315904))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449102400))))[name = string("layers_23_self_attn_linear_k_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = layers_23_self_attn_linear_k_weight_to_fp16_palettized, x = x_513_cast_fp16)[name = string("k_cast_fp16")];
            string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_strides_0 = const()[name = string("v_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_pad_0 = const()[name = string("v_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_linear_v_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449110656))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449897152))))[name = string("layers_23_self_attn_linear_v_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> v_cast_fp16 = conv(dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = layers_23_self_attn_linear_v_weight_to_fp16_palettized, x = x_513_cast_fp16)[name = string("v_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 1]> bv_all_to_fp16 = const()[name = string("bv_all_to_fp16"), val = tensor<fp16, [1, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449905408)))];
            tensor<fp16, [1, 1024, 1, 188]> var_6170_cast_fp16 = add(x = q_cast_fp16, y = bv_all_to_fp16)[name = string("op_6170_cast_fp16")];
            tensor<int32, [3]> var_6171 = const()[name = string("op_6171"), val = tensor<int32, [3]>([8, 128, 188])];
            tensor<fp16, [8, 128, 188]> qb_cast_fp16 = reshape(shape = var_6171, x = var_6170_cast_fp16)[name = string("qb_cast_fp16")];
            bool bd_all_93_transpose_x_0 = const()[name = string("bd_all_93_transpose_x_0"), val = bool(false)];
            bool bd_all_93_transpose_y_0 = const()[name = string("bd_all_93_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 375, 128]> layers_23_self_attn_pos_proj_to_fp16 = const()[name = string("layers_23_self_attn_pos_proj_to_fp16"), val = tensor<fp16, [8, 375, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(449907520)))];
            tensor<fp16, [8, 375, 188]> bd_all_93_cast_fp16 = matmul(transpose_x = bd_all_93_transpose_x_0, transpose_y = bd_all_93_transpose_y_0, x = layers_23_self_attn_pos_proj_to_fp16, y = qb_cast_fp16)[name = string("bd_all_93_cast_fp16")];
            tensor<int32, [3]> x_515_perm_0 = const()[name = string("x_515_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [6]> x_517_pad_0 = const()[name = string("x_517_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 1, 0])];
            string x_517_mode_0 = const()[name = string("x_517_mode_0"), val = string("constant")];
            fp16 const_102_to_fp16 = const()[name = string("const_102_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [8, 188, 375]> x_515_cast_fp16 = transpose(perm = x_515_perm_0, x = bd_all_93_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [8, 188, 376]> x_517_cast_fp16 = pad(constant_val = const_102_to_fp16, mode = x_517_mode_0, pad = x_517_pad_0, x = x_515_cast_fp16)[name = string("x_517_cast_fp16")];
            tensor<int32, [3]> var_6178 = const()[name = string("op_6178"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<fp16, [8, 376, 188]> x_519_cast_fp16 = reshape(shape = var_6178, x = x_517_cast_fp16)[name = string("x_519_cast_fp16")];
            tensor<int32, [3]> var_6181_begin_0 = const()[name = string("op_6181_begin_0"), val = tensor<int32, [3]>([0, 1, 0])];
            tensor<int32, [3]> var_6181_end_0 = const()[name = string("op_6181_end_0"), val = tensor<int32, [3]>([8, 376, 188])];
            tensor<bool, [3]> var_6181_end_mask_0 = const()[name = string("op_6181_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [8, 375, 188]> var_6181_cast_fp16 = slice_by_index(begin = var_6181_begin_0, end = var_6181_end_0, end_mask = var_6181_end_mask_0, x = x_519_cast_fp16)[name = string("op_6181_cast_fp16")];
            tensor<int32, [3]> var_6182 = const()[name = string("op_6182"), val = tensor<int32, [3]>([8, 188, 375])];
            tensor<fp16, [8, 188, 375]> x_521_cast_fp16 = reshape(shape = var_6182, x = var_6181_cast_fp16)[name = string("x_521_cast_fp16")];
            tensor<int32, [3]> bd_all_begin_0 = const()[name = string("bd_all_begin_0"), val = tensor<int32, [3]>([0, 0, 0])];
            tensor<int32, [3]> bd_all_end_0 = const()[name = string("bd_all_end_0"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<bool, [3]> bd_all_end_mask_0 = const()[name = string("bd_all_end_mask_0"), val = tensor<bool, [3]>([true, true, false])];
            tensor<fp16, [8, 188, 188]> bd_all_cast_fp16 = slice_by_index(begin = bd_all_begin_0, end = bd_all_end_0, end_mask = bd_all_end_mask_0, x = x_521_cast_fp16)[name = string("bd_all_cast_fp16")];
            tensor<int32, [4]> var_6187 = const()[name = string("op_6187"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> var_6188_cast_fp16 = reshape(shape = var_6187, x = q_cast_fp16)[name = string("op_6188_cast_fp16")];
            tensor<fp16, [8, 128, 1, 1]> var_6190_to_fp16 = const()[name = string("op_6190_to_fp16"), val = tensor<fp16, [8, 128, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450675584)))];
            tensor<fp16, [8, 128, 1, 188]> var_6191_cast_fp16 = add(x = var_6188_cast_fp16, y = var_6190_to_fp16)[name = string("op_6191_cast_fp16")];
            tensor<int32, [4]> var_6192 = const()[name = string("op_6192"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> kh_cast_fp16 = reshape(shape = var_6192, x = k_cast_fp16)[name = string("kh_cast_fp16")];
            tensor<int32, [4]> var_6194 = const()[name = string("op_6194"), val = tensor<int32, [4]>([8, 128, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> vh_cast_fp16 = reshape(shape = var_6194, x = v_cast_fp16)[name = string("vh_cast_fp16")];
            tensor<int32, [4]> var_6196 = const()[name = string("op_6196"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            string ac_equation_0 = const()[name = string("ac_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [8, 188, 1, 128]> var_6197_cast_fp16 = transpose(perm = var_6196, x = kh_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [8, 188, 1, 188]> ac_cast_fp16 = einsum(equation = ac_equation_0, values = (var_6197_cast_fp16, var_6191_cast_fp16))[name = string("ac_cast_fp16")];
            tensor<int32, [3]> var_6200_perm_0 = const()[name = string("op_6200_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> var_6201_axes_0 = const()[name = string("op_6201_axes_0"), val = tensor<int32, [1]>([2])];
            tensor<fp16, [8, 188, 188]> var_6200_cast_fp16 = transpose(perm = var_6200_perm_0, x = bd_all_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [8, 188, 1, 188]> var_6201_cast_fp16 = expand_dims(axes = var_6201_axes_0, x = var_6200_cast_fp16)[name = string("op_6201_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_6202_cast_fp16 = add(x = ac_cast_fp16, y = var_6201_cast_fp16)[name = string("op_6202_cast_fp16")];
            fp16 var_6203_to_fp16 = const()[name = string("op_6203_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [8, 188, 1, 188]> scores_93_cast_fp16 = mul(x = var_6202_cast_fp16, y = var_6203_to_fp16)[name = string("scores_93_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> scores_cast_fp16 = add(x = scores_93_cast_fp16, y = key_bias)[name = string("scores_cast_fp16")];
            tensor<fp16, [8, 188, 1, 188]> var_6206_cast_fp16 = softmax(axis = var_6077, x = scores_cast_fp16)[name = string("op_6206_cast_fp16")];
            tensor<int32, [4]> transpose_71_perm_0 = const()[name = string("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [4]> transpose_46_perm_0 = const()[name = string("transpose_46_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])];
            tensor<int32, [3]> concat_234 = const()[name = string("concat_234"), val = tensor<int32, [3]>([8, 188, 188])];
            tensor<fp16, [8, 1, 188, 188]> transpose_46_cast_fp16 = transpose(perm = transpose_46_perm_0, x = var_6206_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [8, 188, 188]> reshape_69_cast_fp16 = reshape(shape = concat_234, x = transpose_46_cast_fp16)[name = string("reshape_69_cast_fp16")];
            tensor<int32, [3]> concat_235 = const()[name = string("concat_235"), val = tensor<int32, [3]>([8, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> transpose_71_cast_fp16 = transpose(perm = transpose_71_perm_0, x = vh_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [8, 188, 128]> reshape_70_cast_fp16 = reshape(shape = concat_235, x = transpose_71_cast_fp16)[name = string("reshape_70_cast_fp16")];
            bool matmul_23_transpose_x_0 = const()[name = string("matmul_23_transpose_x_0"), val = bool(false)];
            bool matmul_23_transpose_y_0 = const()[name = string("matmul_23_transpose_y_0"), val = bool(false)];
            tensor<fp16, [8, 188, 128]> matmul_23_cast_fp16 = matmul(transpose_x = matmul_23_transpose_x_0, transpose_y = matmul_23_transpose_y_0, x = reshape_69_cast_fp16, y = reshape_70_cast_fp16)[name = string("matmul_23_cast_fp16")];
            tensor<int32, [4]> concat_239 = const()[name = string("concat_239"), val = tensor<int32, [4]>([8, 1, 188, 128])];
            tensor<fp16, [8, 1, 188, 128]> reshape_71_cast_fp16 = reshape(shape = concat_239, x = matmul_23_cast_fp16)[name = string("reshape_71_cast_fp16")];
            tensor<int32, [4]> ctx_perm_0 = const()[name = string("ctx_perm_0"), val = tensor<int32, [4]>([0, 3, 1, 2])];
            tensor<int32, [4]> var_6211 = const()[name = string("op_6211"), val = tensor<int32, [4]>([1, 1024, 1, 188])];
            tensor<fp16, [8, 128, 1, 188]> ctx_cast_fp16 = transpose(perm = ctx_perm_0, x = reshape_71_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [1, 1024, 1, 188]> input_623_cast_fp16 = reshape(shape = var_6211, x = ctx_cast_fp16)[name = string("input_623_cast_fp16")];
            string var_6218_pad_type_0 = const()[name = string("op_6218_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_6218_strides_0 = const()[name = string("op_6218_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_6218_pad_0 = const()[name = string("op_6218_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_6218_dilations_0 = const()[name = string("op_6218_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_6218_groups_0 = const()[name = string("op_6218_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_self_attn_linear_out_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(450677696))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(451464192))))[name = string("layers_23_self_attn_linear_out_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_6218_cast_fp16 = conv(dilations = var_6218_dilations_0, groups = var_6218_groups_0, pad = var_6218_pad_0, pad_type = var_6218_pad_type_0, strides = var_6218_strides_0, weight = layers_23_self_attn_linear_out_weight_to_fp16_palettized, x = input_623_cast_fp16)[name = string("op_6218_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_523_cast_fp16 = add(x = x_511_cast_fp16, y = var_6218_cast_fp16)[name = string("x_523_cast_fp16")];
            tensor<int32, [1]> var_6234_axes_0 = const()[name = string("op_6234_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_23_norm_conv_eps_scaled_to_fp16 = const()[name = string("layers_23_norm_conv_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_6234_cast_fp16 = layer_norm(axes = var_6234_axes_0, epsilon = layers_23_norm_conv_eps_scaled_to_fp16, x = x_523_cast_fp16)[name = string("op_6234_cast_fp16")];
            tensor<fp16, [1024]> input_625_gamma_0_to_fp16 = const()[name = string("input_625_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(451472448)))];
            tensor<fp16, [1024]> input_625_beta_0_to_fp16 = const()[name = string("input_625_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(451474560)))];
            fp16 input_625_epsilon_0_to_fp16 = const()[name = string("input_625_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_625_cast_fp16 = batch_norm(beta = input_625_beta_0_to_fp16, epsilon = input_625_epsilon_0_to_fp16, gamma = input_625_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_6234_cast_fp16)[name = string("input_625_cast_fp16")];
            string input_627_pad_type_0 = const()[name = string("input_627_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_627_strides_0 = const()[name = string("input_627_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_627_pad_0 = const()[name = string("input_627_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_627_dilations_0 = const()[name = string("input_627_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_627_groups_0 = const()[name = string("input_627_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1, 1]> layers_23_conv_pointwise_conv1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [2048, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(451476672))), lut = tensor<fp16, [128, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453049600))))[name = string("layers_23_conv_pointwise_conv1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 2048, 1, 188]> input_627_cast_fp16 = conv(dilations = input_627_dilations_0, groups = input_627_groups_0, pad = input_627_pad_0, pad_type = input_627_pad_type_0, strides = input_627_strides_0, weight = layers_23_conv_pointwise_conv1_weight_to_fp16_palettized, x = input_625_cast_fp16)[name = string("input_627_cast_fp16")];
            int32 x_525_split_num_splits_0 = const()[name = string("x_525_split_num_splits_0"), val = int32(2)];
            int32 x_525_split_axis_0 = const()[name = string("x_525_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 1, 188]> x_525_split_cast_fp16_0, tensor<fp16, [1, 1024, 1, 188]> x_525_split_cast_fp16_1 = split(axis = x_525_split_axis_0, num_splits = x_525_split_num_splits_0, x = input_627_cast_fp16)[name = string("x_525_split_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_525_split_1_sigmoid_cast_fp16 = sigmoid(x = x_525_split_cast_fp16_1)[name = string("x_525_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_525_cast_fp16 = mul(x = x_525_split_cast_fp16_0, y = x_525_split_1_sigmoid_cast_fp16)[name = string("x_525_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_629_cast_fp16 = mul(x = x_525_cast_fp16, y = pad_mask)[name = string("input_629_cast_fp16")];
            string input_631_pad_type_0 = const()[name = string("input_631_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> input_631_pad_0 = const()[name = string("input_631_pad_0"), val = tensor<int32, [4]>([0, 0, 4, 4])];
            int32 input_631_groups_0 = const()[name = string("input_631_groups_0"), val = int32(1024)];
            tensor<int32, [2]> input_631_strides_0 = const()[name = string("input_631_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> input_631_dilations_0 = const()[name = string("input_631_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [1024, 1, 1, 9]> const_149_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453066048))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453073024))))[name = string("const_149_to_fp16_palettized")];
            tensor<fp16, [1024]> const_150_to_fp16 = const()[name = string("const_150_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453081280)))];
            tensor<fp16, [1, 1024, 1, 188]> input_633_cast_fp16 = conv(bias = const_150_to_fp16, dilations = input_631_dilations_0, groups = input_631_groups_0, pad = input_631_pad_0, pad_type = input_631_pad_type_0, strides = input_631_strides_0, weight = const_149_to_fp16_palettized, x = input_629_cast_fp16)[name = string("input_633_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> input_635_cast_fp16 = silu(x = input_633_cast_fp16)[name = string("input_635_cast_fp16")];
            string var_6266_pad_type_0 = const()[name = string("op_6266_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_6266_strides_0 = const()[name = string("op_6266_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_6266_pad_0 = const()[name = string("op_6266_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_6266_dilations_0 = const()[name = string("op_6266_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_6266_groups_0 = const()[name = string("op_6266_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1, 1]> layers_23_conv_pointwise_conv2_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453083392))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453869888))))[name = string("layers_23_conv_pointwise_conv2_weight_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_6266_cast_fp16 = conv(dilations = var_6266_dilations_0, groups = var_6266_groups_0, pad = var_6266_pad_0, pad_type = var_6266_pad_type_0, strides = var_6266_strides_0, weight = layers_23_conv_pointwise_conv2_weight_to_fp16_palettized, x = input_635_cast_fp16)[name = string("op_6266_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_527_cast_fp16 = add(x = x_523_cast_fp16, y = var_6266_cast_fp16)[name = string("x_527_cast_fp16")];
            tensor<int32, [1]> var_6282_axes_0 = const()[name = string("op_6282_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_23_norm_feed_forward2_eps_scaled_to_fp16 = const()[name = string("layers_23_norm_feed_forward2_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_6282_cast_fp16 = layer_norm(axes = var_6282_axes_0, epsilon = layers_23_norm_feed_forward2_eps_scaled_to_fp16, x = x_527_cast_fp16)[name = string("op_6282_cast_fp16")];
            tensor<fp16, [1024]> input_637_gamma_0_to_fp16 = const()[name = string("input_637_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453878144)))];
            tensor<fp16, [1024]> input_637_beta_0_to_fp16 = const()[name = string("input_637_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453880256)))];
            fp16 input_637_epsilon_0_to_fp16 = const()[name = string("input_637_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_637_cast_fp16 = batch_norm(beta = input_637_beta_0_to_fp16, epsilon = input_637_epsilon_0_to_fp16, gamma = input_637_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_6282_cast_fp16)[name = string("input_637_cast_fp16")];
            string input_639_pad_type_0 = const()[name = string("input_639_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_639_strides_0 = const()[name = string("input_639_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_639_pad_0 = const()[name = string("input_639_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_639_dilations_0 = const()[name = string("input_639_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_639_groups_0 = const()[name = string("input_639_groups_0"), val = int32(1)];
            tensor<fp16, [4096, 1024, 1, 1]> layers_23_feed_forward2_linear1_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [4096, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(453882368))), lut = tensor<fp16, [256, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457028160))))[name = string("layers_23_feed_forward2_linear1_weight_to_fp16_palettized")];
            tensor<fp16, [1, 4096, 1, 188]> input_639_cast_fp16 = conv(dilations = input_639_dilations_0, groups = input_639_groups_0, pad = input_639_pad_0, pad_type = input_639_pad_type_0, strides = input_639_strides_0, weight = layers_23_feed_forward2_linear1_weight_to_fp16_palettized, x = input_637_cast_fp16)[name = string("input_639_cast_fp16")];
            tensor<fp16, [1, 4096, 1, 188]> input_641_cast_fp16 = silu(x = input_639_cast_fp16)[name = string("input_641_cast_fp16")];
            string var_6299_pad_type_0 = const()[name = string("op_6299_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_6299_strides_0 = const()[name = string("op_6299_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_6299_pad_0 = const()[name = string("op_6299_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_6299_dilations_0 = const()[name = string("op_6299_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_6299_groups_0 = const()[name = string("op_6299_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 4096, 1, 1]> op_6300_weight_0_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [1024, 4096, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(457060992))), lut = tensor<fp16, [64, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460206784))))[name = string("op_6300_weight_0_to_fp16_palettized")];
            tensor<fp16, [1, 1024, 1, 188]> var_6300_cast_fp16 = conv(bias = input_19_mean_0_to_fp16, dilations = var_6299_dilations_0, groups = var_6299_groups_0, pad = var_6299_pad_0, pad_type = var_6299_pad_type_0, strides = var_6299_strides_0, weight = op_6300_weight_0_to_fp16_palettized, x = input_641_cast_fp16)[name = string("op_6300_cast_fp16")];
            tensor<fp16, [1, 1024, 1, 188]> x_cast_fp16 = add(x = x_527_cast_fp16, y = var_6300_cast_fp16)[name = string("x_cast_fp16")];
            tensor<int32, [1]> var_6316_axes_0 = const()[name = string("op_6316_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 layers_23_norm_out_eps_scaled_to_fp16 = const()[name = string("layers_23_norm_out_eps_scaled_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> var_6316_cast_fp16 = layer_norm(axes = var_6316_axes_0, epsilon = layers_23_norm_out_eps_scaled_to_fp16, x = x_cast_fp16)[name = string("op_6316_cast_fp16")];
            tensor<fp16, [1024]> input_gamma_0_to_fp16 = const()[name = string("input_gamma_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460215040)))];
            tensor<fp16, [1024]> input_beta_0_to_fp16 = const()[name = string("input_beta_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460217152)))];
            fp16 input_epsilon_0_to_fp16 = const()[name = string("input_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 1024, 1, 188]> input_cast_fp16 = batch_norm(beta = input_beta_0_to_fp16, epsilon = input_epsilon_0_to_fp16, gamma = input_gamma_0_to_fp16, mean = input_19_mean_0_to_fp16, variance = input_19_variance_0_to_fp16, x = var_6316_cast_fp16)[name = string("input_cast_fp16")];
            string var_6329_pad_type_0 = const()[name = string("op_6329_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_6329_strides_0 = const()[name = string("op_6329_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_6329_pad_0 = const()[name = string("op_6329_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_6329_dilations_0 = const()[name = string("op_6329_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_6329_groups_0 = const()[name = string("op_6329_groups_0"), val = int32(1)];
            tensor<fp16, [640, 1024, 1, 1]> joint_enc_weight_to_fp16_palettized = constexpr_lut_to_dense(indices = tensor<uint6, [640, 1024, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460219264))), lut = tensor<fp16, [40, 1, 1, 1, 64, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460710848))))[name = string("joint_enc_weight_to_fp16_palettized")];
            tensor<fp16, [640]> joint_enc_bias_to_fp16 = const()[name = string("joint_enc_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(460716032)))];
            tensor<fp16, [1, 640, 1, 188]> enc_proj = conv(bias = joint_enc_bias_to_fp16, dilations = var_6329_dilations_0, groups = var_6329_groups_0, pad = var_6329_pad_0, pad_type = var_6329_pad_type_0, strides = var_6329_strides_0, weight = joint_enc_weight_to_fp16_palettized, x = input_cast_fp16)[name = string("op_6329_cast_fp16")];
        } -> (enc_proj);
}