program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
{
    func main<ios18>(tensor<fp32, [1, 128, 1501]> features, tensor<int32, [1]> features_length) {
            int32 var_23 = const()[name = string("op_23"), val = int32(-1)];
            tensor<int32, [3]> x_1_perm_0 = const()[name = string("x_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string features_to_fp16_dtype_0 = const()[name = string("features_to_fp16_dtype_0"), val = string("fp16")];
            tensor<int32, [1]> tensor_1_axes_0 = const()[name = string("tensor_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 128, 1501]> features_to_fp16 = cast(dtype = features_to_fp16_dtype_0, x = features)[name = string("cast_10")];
            tensor<fp16, [1, 1501, 128]> x_1_cast_fp16 = transpose(perm = x_1_perm_0, x = features_to_fp16)[name = string("transpose_419")];
            tensor<fp16, [1, 1, 1501, 128]> tensor_1_cast_fp16 = expand_dims(axes = tensor_1_axes_0, x = x_1_cast_fp16)[name = string("tensor_1_cast_fp16")];
            tensor<int32, [1, 1501]> expand_dims_0 = const()[name = string("expand_dims_0"), val = tensor<int32, [1, 1501]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<int32, [1]> var_117_axes_0 = const()[name = string("op_117_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1, 1]> var_117 = expand_dims(axes = var_117_axes_0, x = features_length)[name = string("op_117")];
            tensor<bool, [1, 1501]> time_mask_1 = less(x = expand_dims_0, y = var_117)[name = string("time_mask_1")];
            tensor<int32, [1]> var_119_axes_0 = const()[name = string("op_119_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, [1, 1501, 1]> var_119 = expand_dims(axes = var_119_axes_0, x = time_mask_1)[name = string("op_119")];
            tensor<int32, [3]> var_121_reps_0 = const()[name = string("op_121_reps_0"), val = tensor<int32, [3]>([1, 1, 128])];
            tensor<bool, [1, 1501, 128]> var_121 = tile(reps = var_121_reps_0, x = var_119)[name = string("op_121")];
            tensor<int32, [1]> var_127_axes_0 = const()[name = string("op_127_axes_0"), val = tensor<int32, [1]>([1])];
            string mask_1_to_fp16_dtype_0 = const()[name = string("mask_1_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 1501, 128]> var_121_to_fp16 = cast(dtype = mask_1_to_fp16_dtype_0, x = var_121)[name = string("cast_9")];
            tensor<fp16, [1, 1, 1501, 128]> var_127_cast_fp16 = expand_dims(axes = var_127_axes_0, x = var_121_to_fp16)[name = string("op_127_cast_fp16")];
            tensor<fp16, [1, 1, 1501, 128]> input_1_cast_fp16 = mul(x = tensor_1_cast_fp16, y = var_127_cast_fp16)[name = string("input_1_cast_fp16")];
            string tensor_3_pad_type_0 = const()[name = string("tensor_3_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> tensor_3_pad_0 = const()[name = string("tensor_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> tensor_3_strides_0 = const()[name = string("tensor_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> tensor_3_dilations_0 = const()[name = string("tensor_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 tensor_3_groups_0 = const()[name = string("tensor_3_groups_0"), val = int32(1)];
            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_0_weight_to_fp16 = const()[name = string("module_pre_encode_conv_0_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6144)))];
            tensor<fp16, [256]> module_pre_encode_conv_0_bias_to_fp16 = const()[name = string("module_pre_encode_conv_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10816)))];
            tensor<fp16, [1, 256, 751, 64]> tensor_3_cast_fp16 = conv(bias = module_pre_encode_conv_0_bias_to_fp16, dilations = tensor_3_dilations_0, groups = tensor_3_groups_0, pad = tensor_3_pad_0, pad_type = tensor_3_pad_type_0, strides = tensor_3_strides_0, weight = module_pre_encode_conv_0_weight_to_fp16, x = input_1_cast_fp16)[name = string("tensor_3_cast_fp16")];
            string current_lengths_1_to_fp16_dtype_0 = const()[name = string("current_lengths_1_to_fp16_dtype_0"), val = string("fp16")];
            fp16 var_138_promoted_to_fp16 = const()[name = string("op_138_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> features_length_to_fp16 = cast(dtype = current_lengths_1_to_fp16_dtype_0, x = features_length)[name = string("cast_8")];
            tensor<fp16, [1]> var_139_cast_fp16 = add(x = features_length_to_fp16, y = var_138_promoted_to_fp16)[name = string("op_139_cast_fp16")];
            fp16 var_140_promoted_to_fp16 = const()[name = string("op_140_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_141_cast_fp16 = add(x = var_139_cast_fp16, y = var_140_promoted_to_fp16)[name = string("op_141_cast_fp16")];
            fp16 var_142_promoted_to_fp16 = const()[name = string("op_142_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> var_143_cast_fp16 = sub(x = var_141_cast_fp16, y = var_142_promoted_to_fp16)[name = string("op_143_cast_fp16")];
            fp16 var_21_promoted_to_fp16 = const()[name = string("op_21_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> floor_div_0_cast_fp16 = floor_div(x = var_143_cast_fp16, y = var_21_promoted_to_fp16)[name = string("floor_div_0_cast_fp16")];
            fp16 var_145_promoted_to_fp16 = const()[name = string("op_145_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> current_lengths_3_cast_fp16 = add(x = floor_div_0_cast_fp16, y = var_145_promoted_to_fp16)[name = string("current_lengths_3_cast_fp16")];
            string lengths_19_dtype_0 = const()[name = string("lengths_19_dtype_0"), val = string("int32")];
            tensor<int32, [1, 751]> expand_dims_1 = const()[name = string("expand_dims_1"), val = tensor<int32, [1, 751]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11392)))];
            tensor<int32, [1]> var_154_axes_0 = const()[name = string("op_154_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> current_lengths_3_cast_fp16_to_int32 = cast(dtype = lengths_19_dtype_0, x = current_lengths_3_cast_fp16)[name = string("cast_7")];
            tensor<int32, [1, 1]> var_154 = expand_dims(axes = var_154_axes_0, x = current_lengths_3_cast_fp16_to_int32)[name = string("op_154")];
            tensor<bool, [1, 751]> time_mask_3 = less(x = expand_dims_1, y = var_154)[name = string("time_mask_3")];
            tensor<int32, [1]> var_156_axes_0 = const()[name = string("op_156_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, [1, 751, 1]> var_156 = expand_dims(axes = var_156_axes_0, x = time_mask_3)[name = string("op_156")];
            tensor<int32, [3]> var_158_reps_0 = const()[name = string("op_158_reps_0"), val = tensor<int32, [3]>([1, 1, 64])];
            tensor<bool, [1, 751, 64]> var_158 = tile(reps = var_158_reps_0, x = var_156)[name = string("op_158")];
            tensor<int32, [1]> var_164_axes_0 = const()[name = string("op_164_axes_0"), val = tensor<int32, [1]>([1])];
            string mask_3_to_fp16_dtype_0 = const()[name = string("mask_3_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 751, 64]> var_158_to_fp16 = cast(dtype = mask_3_to_fp16_dtype_0, x = var_158)[name = string("cast_6")];
            tensor<fp16, [1, 1, 751, 64]> var_164_cast_fp16 = expand_dims(axes = var_164_axes_0, x = var_158_to_fp16)[name = string("op_164_cast_fp16")];
            tensor<int32, [4]> expanded_mask_3_reps_0 = const()[name = string("expanded_mask_3_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
            tensor<fp16, [1, 256, 751, 64]> expanded_mask_3_cast_fp16 = tile(reps = expanded_mask_3_reps_0, x = var_164_cast_fp16)[name = string("expanded_mask_3_cast_fp16")];
            tensor<fp16, [1, 256, 751, 64]> input_3_cast_fp16 = mul(x = tensor_3_cast_fp16, y = expanded_mask_3_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<fp16, [1, 256, 751, 64]> tensor_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = string("tensor_5_cast_fp16")];
            tensor<fp16, [1, 256, 751, 64]> input_5_cast_fp16 = mul(x = tensor_5_cast_fp16, y = expanded_mask_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string tensor_7_pad_type_0 = const()[name = string("tensor_7_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> tensor_7_pad_0 = const()[name = string("tensor_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> tensor_7_strides_0 = const()[name = string("tensor_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 tensor_7_groups_0 = const()[name = string("tensor_7_groups_0"), val = int32(256)];
            tensor<int32, [2]> tensor_7_dilations_0 = const()[name = string("tensor_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_2_weight_to_fp16 = const()[name = string("module_pre_encode_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14464)))];
            tensor<fp16, [256]> module_pre_encode_conv_2_bias_to_fp16 = const()[name = string("module_pre_encode_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19136)))];
            tensor<fp16, [1, 256, 376, 32]> tensor_7_cast_fp16 = conv(bias = module_pre_encode_conv_2_bias_to_fp16, dilations = tensor_7_dilations_0, groups = tensor_7_groups_0, pad = tensor_7_pad_0, pad_type = tensor_7_pad_type_0, strides = tensor_7_strides_0, weight = module_pre_encode_conv_2_weight_to_fp16, x = input_5_cast_fp16)[name = string("tensor_7_cast_fp16")];
            fp16 var_184_promoted_to_fp16 = const()[name = string("op_184_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_185_cast_fp16 = add(x = current_lengths_3_cast_fp16, y = var_184_promoted_to_fp16)[name = string("op_185_cast_fp16")];
            fp16 var_186_promoted_to_fp16 = const()[name = string("op_186_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_187_cast_fp16 = add(x = var_185_cast_fp16, y = var_186_promoted_to_fp16)[name = string("op_187_cast_fp16")];
            fp16 var_188_promoted_to_fp16 = const()[name = string("op_188_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> var_189_cast_fp16 = sub(x = var_187_cast_fp16, y = var_188_promoted_to_fp16)[name = string("op_189_cast_fp16")];
            fp16 var_21_promoted_1_to_fp16 = const()[name = string("op_21_promoted_1_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> floor_div_1_cast_fp16 = floor_div(x = var_189_cast_fp16, y = var_21_promoted_1_to_fp16)[name = string("floor_div_1_cast_fp16")];
            fp16 var_191_promoted_to_fp16 = const()[name = string("op_191_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> current_lengths_5_cast_fp16 = add(x = floor_div_1_cast_fp16, y = var_191_promoted_to_fp16)[name = string("current_lengths_5_cast_fp16")];
            string lengths_21_dtype_0 = const()[name = string("lengths_21_dtype_0"), val = string("int32")];
            tensor<int32, [1, 376]> expand_dims_2 = const()[name = string("expand_dims_2"), val = tensor<int32, [1, 376]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19712)))];
            tensor<int32, [1]> var_200_axes_0 = const()[name = string("op_200_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> current_lengths_5_cast_fp16_to_int32 = cast(dtype = lengths_21_dtype_0, x = current_lengths_5_cast_fp16)[name = string("cast_5")];
            tensor<int32, [1, 1]> var_200 = expand_dims(axes = var_200_axes_0, x = current_lengths_5_cast_fp16_to_int32)[name = string("op_200")];
            tensor<bool, [1, 376]> time_mask_5 = less(x = expand_dims_2, y = var_200)[name = string("time_mask_5")];
            tensor<int32, [1]> var_202_axes_0 = const()[name = string("op_202_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, [1, 376, 1]> var_202 = expand_dims(axes = var_202_axes_0, x = time_mask_5)[name = string("op_202")];
            tensor<int32, [3]> var_204_reps_0 = const()[name = string("op_204_reps_0"), val = tensor<int32, [3]>([1, 1, 32])];
            tensor<bool, [1, 376, 32]> var_204 = tile(reps = var_204_reps_0, x = var_202)[name = string("op_204")];
            tensor<int32, [1]> var_210_axes_0 = const()[name = string("op_210_axes_0"), val = tensor<int32, [1]>([1])];
            string mask_5_to_fp16_dtype_0 = const()[name = string("mask_5_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 376, 32]> var_204_to_fp16 = cast(dtype = mask_5_to_fp16_dtype_0, x = var_204)[name = string("cast_4")];
            tensor<fp16, [1, 1, 376, 32]> var_210_cast_fp16 = expand_dims(axes = var_210_axes_0, x = var_204_to_fp16)[name = string("op_210_cast_fp16")];
            tensor<int32, [4]> expanded_mask_7_reps_0 = const()[name = string("expanded_mask_7_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
            tensor<fp16, [1, 256, 376, 32]> expanded_mask_7_cast_fp16 = tile(reps = expanded_mask_7_reps_0, x = var_210_cast_fp16)[name = string("expanded_mask_7_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> input_7_cast_fp16 = mul(x = tensor_7_cast_fp16, y = expanded_mask_7_cast_fp16)[name = string("input_7_cast_fp16")];
            string tensor_9_pad_type_0 = const()[name = string("tensor_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> tensor_9_strides_0 = const()[name = string("tensor_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> tensor_9_pad_0 = const()[name = string("tensor_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> tensor_9_dilations_0 = const()[name = string("tensor_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 tensor_9_groups_0 = const()[name = string("tensor_9_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> module_pre_encode_conv_3_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21312))), scale = tensor<fp16, [256, 8, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54144))))[name = string("module_pre_encode_conv_3_weight_to_fp16_quantized")];
            tensor<fp16, [256]> module_pre_encode_conv_3_bias_to_fp16 = const()[name = string("module_pre_encode_conv_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58304)))];
            tensor<fp16, [1, 256, 376, 32]> tensor_9_cast_fp16 = conv(bias = module_pre_encode_conv_3_bias_to_fp16, dilations = tensor_9_dilations_0, groups = tensor_9_groups_0, pad = tensor_9_pad_0, pad_type = tensor_9_pad_type_0, strides = tensor_9_strides_0, weight = module_pre_encode_conv_3_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = string("tensor_9_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> input_9_cast_fp16 = mul(x = tensor_9_cast_fp16, y = expanded_mask_7_cast_fp16)[name = string("input_9_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> tensor_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = string("tensor_11_cast_fp16")];
            tensor<fp16, [1, 256, 376, 32]> input_11_cast_fp16 = mul(x = tensor_11_cast_fp16, y = expanded_mask_7_cast_fp16)[name = string("input_11_cast_fp16")];
            string tensor_13_pad_type_0 = const()[name = string("tensor_13_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> tensor_13_pad_0 = const()[name = string("tensor_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> tensor_13_strides_0 = const()[name = string("tensor_13_strides_0"), val = tensor<int32, [2]>([2, 2])];
            int32 tensor_13_groups_0 = const()[name = string("tensor_13_groups_0"), val = int32(256)];
            tensor<int32, [2]> tensor_13_dilations_0 = const()[name = string("tensor_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> module_pre_encode_conv_5_weight_to_fp16 = const()[name = string("module_pre_encode_conv_5_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58880)))];
            tensor<fp16, [256]> module_pre_encode_conv_5_bias_to_fp16 = const()[name = string("module_pre_encode_conv_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63552)))];
            tensor<fp16, [1, 256, 188, 16]> tensor_13_cast_fp16 = conv(bias = module_pre_encode_conv_5_bias_to_fp16, dilations = tensor_13_dilations_0, groups = tensor_13_groups_0, pad = tensor_13_pad_0, pad_type = tensor_13_pad_type_0, strides = tensor_13_strides_0, weight = module_pre_encode_conv_5_weight_to_fp16, x = input_11_cast_fp16)[name = string("tensor_13_cast_fp16")];
            fp16 var_245_promoted_to_fp16 = const()[name = string("op_245_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_246_cast_fp16 = add(x = current_lengths_5_cast_fp16, y = var_245_promoted_to_fp16)[name = string("op_246_cast_fp16")];
            fp16 var_247_promoted_to_fp16 = const()[name = string("op_247_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> var_248_cast_fp16 = add(x = var_246_cast_fp16, y = var_247_promoted_to_fp16)[name = string("op_248_cast_fp16")];
            fp16 var_249_promoted_to_fp16 = const()[name = string("op_249_promoted_to_fp16"), val = fp16(0x1.8p+1)];
            tensor<fp16, [1]> var_250_cast_fp16 = sub(x = var_248_cast_fp16, y = var_249_promoted_to_fp16)[name = string("op_250_cast_fp16")];
            fp16 var_21_promoted_2_to_fp16 = const()[name = string("op_21_promoted_2_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1]> floor_div_2_cast_fp16 = floor_div(x = var_250_cast_fp16, y = var_21_promoted_2_to_fp16)[name = string("floor_div_2_cast_fp16")];
            fp16 var_252_promoted_to_fp16 = const()[name = string("op_252_promoted_to_fp16"), val = fp16(0x1p+0)];
            tensor<fp16, [1]> current_lengths_cast_fp16 = add(x = floor_div_2_cast_fp16, y = var_252_promoted_to_fp16)[name = string("current_lengths_cast_fp16")];
            string lengths_dtype_0 = const()[name = string("lengths_dtype_0"), val = string("int32")];
            tensor<int32, [1, 188]> expand_dims_3 = const()[name = string("expand_dims_3"), val = tensor<int32, [1, 188]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64128)))];
            tensor<int32, [1]> var_261_axes_0 = const()[name = string("op_261_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> current_lengths_cast_fp16_to_int32 = cast(dtype = lengths_dtype_0, x = current_lengths_cast_fp16)[name = string("cast_3")];
            tensor<int32, [1, 1]> var_261 = expand_dims(axes = var_261_axes_0, x = current_lengths_cast_fp16_to_int32)[name = string("op_261")];
            tensor<bool, [1, 188]> time_mask = less(x = expand_dims_3, y = var_261)[name = string("time_mask")];
            tensor<int32, [1]> var_263_axes_0 = const()[name = string("op_263_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, [1, 188, 1]> var_263 = expand_dims(axes = var_263_axes_0, x = time_mask)[name = string("op_263")];
            tensor<int32, [3]> var_265_reps_0 = const()[name = string("op_265_reps_0"), val = tensor<int32, [3]>([1, 1, 16])];
            tensor<bool, [1, 188, 16]> var_265 = tile(reps = var_265_reps_0, x = var_263)[name = string("op_265")];
            tensor<int32, [1]> var_271_axes_0 = const()[name = string("op_271_axes_0"), val = tensor<int32, [1]>([1])];
            string mask_7_to_fp16_dtype_0 = const()[name = string("mask_7_to_fp16_dtype_0"), val = string("fp16")];
            tensor<fp16, [1, 188, 16]> var_265_to_fp16 = cast(dtype = mask_7_to_fp16_dtype_0, x = var_265)[name = string("cast_2")];
            tensor<fp16, [1, 1, 188, 16]> var_271_cast_fp16 = expand_dims(axes = var_271_axes_0, x = var_265_to_fp16)[name = string("op_271_cast_fp16")];
            tensor<int32, [4]> expanded_mask_13_reps_0 = const()[name = string("expanded_mask_13_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
            tensor<fp16, [1, 256, 188, 16]> expanded_mask_13_cast_fp16 = tile(reps = expanded_mask_13_reps_0, x = var_271_cast_fp16)[name = string("expanded_mask_13_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> input_13_cast_fp16 = mul(x = tensor_13_cast_fp16, y = expanded_mask_13_cast_fp16)[name = string("input_13_cast_fp16")];
            string tensor_15_pad_type_0 = const()[name = string("tensor_15_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> tensor_15_strides_0 = const()[name = string("tensor_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> tensor_15_pad_0 = const()[name = string("tensor_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> tensor_15_dilations_0 = const()[name = string("tensor_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 tensor_15_groups_0 = const()[name = string("tensor_15_groups_0"), val = int32(1)];
            tensor<fp16, [256, 256, 1, 1]> module_pre_encode_conv_6_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [256, 256, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64960))), scale = tensor<fp16, [256, 8, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(97792))))[name = string("module_pre_encode_conv_6_weight_to_fp16_quantized")];
            tensor<fp16, [256]> module_pre_encode_conv_6_bias_to_fp16 = const()[name = string("module_pre_encode_conv_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101952)))];
            tensor<fp16, [1, 256, 188, 16]> tensor_15_cast_fp16 = conv(bias = module_pre_encode_conv_6_bias_to_fp16, dilations = tensor_15_dilations_0, groups = tensor_15_groups_0, pad = tensor_15_pad_0, pad_type = tensor_15_pad_type_0, strides = tensor_15_strides_0, weight = module_pre_encode_conv_6_weight_to_fp16_quantized, x = input_13_cast_fp16)[name = string("tensor_15_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> input_15_cast_fp16 = mul(x = tensor_15_cast_fp16, y = expanded_mask_13_cast_fp16)[name = string("input_15_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> tensor_cast_fp16 = relu(x = input_15_cast_fp16)[name = string("tensor_cast_fp16")];
            tensor<fp16, [1, 256, 188, 16]> x_3_cast_fp16 = mul(x = tensor_cast_fp16, y = expanded_mask_13_cast_fp16)[name = string("x_3_cast_fp16")];
            tensor<int32, [4]> var_305_perm_0 = const()[name = string("op_305_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_306 = const()[name = string("op_306"), val = tensor<int32, [3]>([1, 188, -1])];
            tensor<fp16, [1, 188, 256, 16]> var_305_cast_fp16 = transpose(perm = var_305_perm_0, x = x_3_cast_fp16)[name = string("transpose_418")];
            tensor<fp16, [1, 188, 4096]> input_17_cast_fp16 = reshape(shape = var_306, x = var_305_cast_fp16)[name = string("input_17_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_pre_encode_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102528))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2199744))))[name = string("module_pre_encode_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_pre_encode_out_bias_to_fp16 = const()[name = string("module_pre_encode_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2461952)))];
            tensor<fp16, [1, 188, 1024]> linear_0_cast_fp16 = linear(bias = module_pre_encode_out_bias_to_fp16, weight = module_pre_encode_out_weight_to_fp16_quantized, x = input_17_cast_fp16)[name = string("linear_0_cast_fp16")];
            string padding_length_dtype_0 = const()[name = string("padding_length_dtype_0"), val = string("int32")];
            tensor<int32, [1]> var_344_axes_0 = const()[name = string("op_344_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<int32, [1]> encoder_length = cast(dtype = padding_length_dtype_0, x = current_lengths_cast_fp16)[name = string("cast_1")];
            tensor<int32, [1, 1]> var_344 = expand_dims(axes = var_344_axes_0, x = encoder_length)[name = string("op_344")];
            tensor<bool, [1, 188]> pad_mask_1 = less(x = expand_dims_3, y = var_344)[name = string("pad_mask_1")];
            tensor<int32, [1]> var_346_axes_0 = const()[name = string("op_346_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 188]> var_346 = expand_dims(axes = var_346_axes_0, x = pad_mask_1)[name = string("op_346")];
            tensor<int32, [3]> var_347 = const()[name = string("op_347"), val = tensor<int32, [3]>([1, 188, 1])];
            tensor<bool, [1, 188, 188]> pad_mask_for_att_mask_1 = tile(reps = var_347, x = var_346)[name = string("pad_mask_for_att_mask_1")];
            tensor<int32, [3]> var_349_perm_0 = const()[name = string("op_349_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<bool, [1, 188, 188]> var_349 = transpose(perm = var_349_perm_0, x = pad_mask_for_att_mask_1)[name = string("transpose_417")];
            tensor<bool, [1, 188, 188]> pad_mask_for_att_mask = logical_and(x = pad_mask_for_att_mask_1, y = var_349)[name = string("pad_mask_for_att_mask")];
            tensor<bool, [1, 188, 188]> const_63 = const()[name = string("const_63"), val = tensor<bool, [1, 188, 188]>([[[true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true]]])];
            tensor<bool, [1, 188, 188]> att_mask = logical_and(x = pad_mask_for_att_mask, y = const_63)[name = string("att_mask")];
            tensor<bool, [1, 188, 188]> mask_9 = logical_not(x = att_mask)[name = string("mask_9")];
            tensor<bool, [1, 188]> pad_mask = logical_not(x = pad_mask_1)[name = string("pad_mask")];
            tensor<int32, [1]> input_21_axes_0 = const()[name = string("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2464064)))];
            tensor<fp16, [1024]> module_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2466176)))];
            fp16 var_9_to_fp16 = const()[name = string("op_9_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 188, 1024]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = module_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_0_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2468288))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4565504))))[name = string("module_layers_0_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_0_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_0_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4827712)))];
            tensor<fp16, [1, 188, 4096]> linear_1_cast_fp16 = linear(bias = module_layers_0_feed_forward1_linear1_bias_to_fp16, weight = module_layers_0_feed_forward1_linear1_weight_to_fp16_quantized, x = input_21_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = string("input_25_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_0_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4835968))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6933184))))[name = string("module_layers_0_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_0_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_0_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7195392)))];
            tensor<fp16, [1, 188, 1024]> linear_2_cast_fp16 = linear(bias = module_layers_0_feed_forward1_linear2_bias_to_fp16, weight = module_layers_0_feed_forward1_linear2_weight_to_fp16_quantized, x = input_25_cast_fp16)[name = string("linear_2_cast_fp16")];
            fp16 var_382_to_fp16 = const()[name = string("op_382_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_383_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_382_to_fp16)[name = string("op_383_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_31_cast_fp16 = add(x = linear_0_cast_fp16, y = var_383_cast_fp16)[name = string("input_31_cast_fp16")];
            tensor<int32, [1]> query_1_axes_0 = const()[name = string("query_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_0_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7197504)))];
            tensor<fp16, [1024]> module_layers_0_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7199616)))];
            tensor<fp16, [1, 188, 1024]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = module_layers_0_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_self_att_weight_to_fp16, x = input_31_cast_fp16)[name = string("query_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7201728))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7726080))))[name = string("module_layers_0_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_0_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_0_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7791680)))];
            tensor<fp16, [1, 188, 1024]> linear_3_cast_fp16 = linear(bias = module_layers_0_self_attn_linear_q_bias_to_fp16, weight = module_layers_0_self_attn_linear_q_weight_to_fp16_quantized, x = query_1_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<int32, [4]> var_400 = const()[name = string("op_400"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_1_cast_fp16 = reshape(shape = var_400, x = linear_3_cast_fp16)[name = string("q_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7793792))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8318144))))[name = string("module_layers_0_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_0_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_0_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8383744)))];
            tensor<fp16, [1, 188, 1024]> linear_4_cast_fp16 = linear(bias = module_layers_0_self_attn_linear_k_bias_to_fp16, weight = module_layers_0_self_attn_linear_k_weight_to_fp16_quantized, x = query_1_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<int32, [4]> var_405 = const()[name = string("op_405"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_1_cast_fp16 = reshape(shape = var_405, x = linear_4_cast_fp16)[name = string("k_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8385856))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8910208))))[name = string("module_layers_0_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_0_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_0_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8975808)))];
            tensor<fp16, [1, 188, 1024]> linear_5_cast_fp16 = linear(bias = module_layers_0_self_attn_linear_v_bias_to_fp16, weight = module_layers_0_self_attn_linear_v_weight_to_fp16_quantized, x = query_1_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<int32, [4]> var_410 = const()[name = string("op_410"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_1_cast_fp16 = reshape(shape = var_410, x = linear_5_cast_fp16)[name = string("v_1_cast_fp16")];
            tensor<int32, [4]> value_3_perm_0 = const()[name = string("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8977920))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8978496))))[name = string("module_layers_0_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_422_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_422_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_0_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8978624))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8979200))))[name = string("module_layers_0_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_424_cast_fp16 = add(x = q_1_cast_fp16, y = module_layers_0_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_424_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_1_perm_0 = const()[name = string("q_with_bias_v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_7_transpose_x_0 = const()[name = string("x_7_transpose_x_0"), val = bool(false)];
            bool x_7_transpose_y_0 = const()[name = string("x_7_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_426_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8979328))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9171392))))[name = string("op_426_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_1_cast_fp16 = transpose(perm = q_with_bias_v_1_perm_0, x = var_424_cast_fp16)[name = string("transpose_416")];
            tensor<fp16, [1, 8, 188, 375]> x_7_cast_fp16 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = q_with_bias_v_1_cast_fp16, y = op_426_to_fp16_quantized)[name = string("x_7_cast_fp16")];
            tensor<int32, [8]> x_9_pad_0 = const()[name = string("x_9_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_9_mode_0 = const()[name = string("x_9_mode_0"), val = string("constant")];
            fp16 const_70_to_fp16 = const()[name = string("const_70_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_9_cast_fp16 = pad(constant_val = const_70_to_fp16, mode = x_9_mode_0, pad = x_9_pad_0, x = x_7_cast_fp16)[name = string("x_9_cast_fp16")];
            tensor<int32, [4]> var_434 = const()[name = string("op_434"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_11_cast_fp16 = reshape(shape = var_434, x = x_9_cast_fp16)[name = string("x_11_cast_fp16")];
            tensor<int32, [4]> var_438_begin_0 = const()[name = string("op_438_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_438_end_0 = const()[name = string("op_438_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_438_end_mask_0 = const()[name = string("op_438_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_438_cast_fp16 = slice_by_index(begin = var_438_begin_0, end = var_438_end_0, end_mask = var_438_end_mask_0, x = x_11_cast_fp16)[name = string("op_438_cast_fp16")];
            tensor<int32, [4]> var_439 = const()[name = string("op_439"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_1_cast_fp16 = reshape(shape = var_439, x = var_438_cast_fp16)[name = string("matrix_bd_1_cast_fp16")];
            bool matrix_ac_1_transpose_x_0 = const()[name = string("matrix_ac_1_transpose_x_0"), val = bool(false)];
            bool matrix_ac_1_transpose_y_0 = const()[name = string("matrix_ac_1_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_128_perm_0 = const()[name = string("transpose_128_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_129_perm_0 = const()[name = string("transpose_129_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_129 = transpose(perm = transpose_129_perm_0, x = k_1_cast_fp16)[name = string("transpose_414")];
            tensor<fp16, [1, 8, 188, 128]> transpose_128 = transpose(perm = transpose_128_perm_0, x = var_422_cast_fp16)[name = string("transpose_415")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_1_cast_fp16 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_128, y = transpose_129)[name = string("matrix_ac_1_cast_fp16")];
            tensor<int32, [4]> matrix_bd_3_begin_0 = const()[name = string("matrix_bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_3_end_0 = const()[name = string("matrix_bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_3_end_mask_0 = const()[name = string("matrix_bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_3_cast_fp16 = slice_by_index(begin = matrix_bd_3_begin_0, end = matrix_bd_3_end_0, end_mask = matrix_bd_3_end_mask_0, x = matrix_bd_1_cast_fp16)[name = string("matrix_bd_3_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_448_cast_fp16 = add(x = matrix_ac_1_cast_fp16, y = matrix_bd_3_cast_fp16)[name = string("op_448_cast_fp16")];
            fp16 _inversed_scores_1_y_0_to_fp16 = const()[name = string("_inversed_scores_1_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_1_cast_fp16 = mul(x = var_448_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
            tensor<int32, [1]> mask_11_axes_0 = const()[name = string("mask_11_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 188, 188]> mask_11 = expand_dims(axes = mask_11_axes_0, x = mask_9)[name = string("mask_11")];
            fp16 var_12_to_fp16 = const()[name = string("op_12_to_fp16"), val = fp16(-0x1.388p+13)];
            tensor<fp16, [1, 8, 188, 188]> scores_3_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_1_cast_fp16, cond = mask_11)[name = string("scores_3_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_454_cast_fp16 = softmax(axis = var_23, x = scores_3_cast_fp16)[name = string("op_454_cast_fp16")];
            fp16 var_11_to_fp16 = const()[name = string("op_11_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 188]> input_33_cast_fp16 = select(a = var_11_to_fp16, b = var_454_cast_fp16, cond = mask_11)[name = string("input_33_cast_fp16")];
            bool x_13_transpose_x_0 = const()[name = string("x_13_transpose_x_0"), val = bool(false)];
            bool x_13_transpose_y_0 = const()[name = string("x_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = v_1_cast_fp16)[name = string("transpose_413")];
            tensor<fp16, [1, 8, 188, 128]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_33_cast_fp16, y = value_3_cast_fp16)[name = string("x_13_cast_fp16")];
            tensor<int32, [4]> var_458_perm_0 = const()[name = string("op_458_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_459 = const()[name = string("op_459"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_458_cast_fp16 = transpose(perm = var_458_perm_0, x = x_13_cast_fp16)[name = string("transpose_412")];
            tensor<fp16, [1, 188, 1024]> input_35_cast_fp16 = reshape(shape = var_459, x = var_458_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_0_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9174464))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9698816))))[name = string("module_layers_0_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_0_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_0_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9764416)))];
            tensor<fp16, [1, 188, 1024]> linear_7_cast_fp16 = linear(bias = module_layers_0_self_attn_linear_out_bias_to_fp16, weight = module_layers_0_self_attn_linear_out_weight_to_fp16_quantized, x = input_35_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_7_cast_fp16)[name = string("input_39_cast_fp16")];
            tensor<int32, [1]> x_17_axes_0 = const()[name = string("x_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_0_norm_conv_weight_to_fp16 = const()[name = string("module_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9766528)))];
            tensor<fp16, [1024]> module_layers_0_norm_conv_bias_to_fp16 = const()[name = string("module_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9768640)))];
            tensor<fp16, [1, 188, 1024]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = module_layers_0_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = string("x_17_cast_fp16")];
            tensor<int32, [3]> input_41_perm_0 = const()[name = string("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_43_pad_type_0 = const()[name = string("input_43_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_43_strides_0 = const()[name = string("input_43_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_43_pad_0 = const()[name = string("input_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_43_dilations_0 = const()[name = string("input_43_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_43_groups_0 = const()[name = string("input_43_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9770752))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10819392))))[name = string("module_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_0_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_0_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10950528)))];
            tensor<fp16, [1, 1024, 188]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = string("transpose_411")];
            tensor<fp16, [1, 2048, 188]> input_43_cast_fp16 = conv(bias = module_layers_0_conv_pointwise_conv1_bias_to_fp16, dilations = input_43_dilations_0, groups = input_43_groups_0, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = input_43_strides_0, weight = module_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_41_cast_fp16)[name = string("input_43_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, 188]> x_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_43_cast_fp16)[name = string("x_19_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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<int32, [1]> var_483_axes_0 = const()[name = string("op_483_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 188]> var_483 = expand_dims(axes = var_483_axes_0, x = pad_mask)[name = string("op_483")];
            tensor<fp16, [1, 1024, 188]> input_45_cast_fp16 = select(a = var_11_to_fp16, b = x_19_cast_fp16, cond = var_483)[name = string("input_45_cast_fp16")];
            tensor<int32, [6]> input_47_pad_0 = const()[name = string("input_47_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("constant")];
            fp16 const_73_to_fp16 = const()[name = string("const_73_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_47_cast_fp16 = pad(constant_val = const_73_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")];
            string input_49_pad_type_0 = const()[name = string("input_49_pad_type_0"), val = string("valid")];
            int32 input_49_groups_0 = const()[name = string("input_49_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_49_strides_0 = const()[name = string("input_49_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_49_pad_0 = const()[name = string("input_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_49_dilations_0 = const()[name = string("input_49_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_384_to_fp16 = const()[name = string("const_384_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10954688)))];
            tensor<fp16, [1024]> const_385_to_fp16 = const()[name = string("const_385_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10973184)))];
            tensor<fp16, [1, 1024, 188]> input_51_cast_fp16 = conv(bias = const_385_to_fp16, dilations = input_49_dilations_0, groups = input_49_groups_0, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = input_49_strides_0, weight = const_384_to_fp16, x = input_47_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_53_cast_fp16 = silu(x = input_51_cast_fp16)[name = string("input_53_cast_fp16")];
            string x_21_pad_type_0 = const()[name = string("x_21_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_21_strides_0 = const()[name = string("x_21_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_21_pad_0 = const()[name = string("x_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_21_dilations_0 = const()[name = string("x_21_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_21_groups_0 = const()[name = string("x_21_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10975296))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11499648))))[name = string("module_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_0_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_0_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11565248)))];
            tensor<fp16, [1, 1024, 188]> x_21_cast_fp16 = conv(bias = module_layers_0_conv_pointwise_conv2_bias_to_fp16, dilations = x_21_dilations_0, groups = x_21_groups_0, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = x_21_strides_0, weight = module_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_53_cast_fp16)[name = string("x_21_cast_fp16")];
            tensor<int32, [3]> input_55_perm_0 = const()[name = string("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = x_21_cast_fp16)[name = string("transpose_410")];
            tensor<fp16, [1, 188, 1024]> input_57_cast_fp16 = add(x = input_39_cast_fp16, y = input_55_cast_fp16)[name = string("input_57_cast_fp16")];
            tensor<int32, [1]> input_59_axes_0 = const()[name = string("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11567360)))];
            tensor<fp16, [1024]> module_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11569472)))];
            tensor<fp16, [1, 188, 1024]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = module_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_feed_forward2_weight_to_fp16, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_0_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11571584))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13668800))))[name = string("module_layers_0_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_0_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_0_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13931008)))];
            tensor<fp16, [1, 188, 4096]> linear_8_cast_fp16 = linear(bias = module_layers_0_feed_forward2_linear1_bias_to_fp16, weight = module_layers_0_feed_forward2_linear1_weight_to_fp16_quantized, x = input_59_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_63_cast_fp16 = silu(x = linear_8_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_0_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13939264))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16036480))))[name = string("module_layers_0_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_0_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_0_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16298688)))];
            tensor<fp16, [1, 188, 1024]> linear_9_cast_fp16 = linear(bias = module_layers_0_feed_forward2_linear2_bias_to_fp16, weight = module_layers_0_feed_forward2_linear2_weight_to_fp16_quantized, x = input_63_cast_fp16)[name = string("linear_9_cast_fp16")];
            fp16 var_525_to_fp16 = const()[name = string("op_525_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_526_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_525_to_fp16)[name = string("op_526_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_69_cast_fp16 = add(x = input_57_cast_fp16, y = var_526_cast_fp16)[name = string("input_69_cast_fp16")];
            tensor<int32, [1]> input_71_axes_0 = const()[name = string("input_71_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_0_norm_out_weight_to_fp16 = const()[name = string("module_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16300800)))];
            tensor<fp16, [1024]> module_layers_0_norm_out_bias_to_fp16 = const()[name = string("module_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16302912)))];
            tensor<fp16, [1, 188, 1024]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = module_layers_0_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_0_norm_out_weight_to_fp16, x = input_69_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<int32, [1]> input_73_axes_0 = const()[name = string("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16305024)))];
            tensor<fp16, [1024]> module_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16307136)))];
            tensor<fp16, [1, 188, 1024]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = module_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward1_weight_to_fp16, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_1_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16309248))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18406464))))[name = string("module_layers_1_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_1_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_1_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18668672)))];
            tensor<fp16, [1, 188, 4096]> linear_10_cast_fp16 = linear(bias = module_layers_1_feed_forward1_linear1_bias_to_fp16, weight = module_layers_1_feed_forward1_linear1_weight_to_fp16_quantized, x = input_73_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_77_cast_fp16 = silu(x = linear_10_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_1_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18676928))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20774144))))[name = string("module_layers_1_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_1_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_1_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21036352)))];
            tensor<fp16, [1, 188, 1024]> linear_11_cast_fp16 = linear(bias = module_layers_1_feed_forward1_linear2_bias_to_fp16, weight = module_layers_1_feed_forward1_linear2_weight_to_fp16_quantized, x = input_77_cast_fp16)[name = string("linear_11_cast_fp16")];
            fp16 var_556_to_fp16 = const()[name = string("op_556_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_557_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_556_to_fp16)[name = string("op_557_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_83_cast_fp16 = add(x = input_71_cast_fp16, y = var_557_cast_fp16)[name = string("input_83_cast_fp16")];
            tensor<int32, [1]> query_3_axes_0 = const()[name = string("query_3_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_1_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21038464)))];
            tensor<fp16, [1024]> module_layers_1_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21040576)))];
            tensor<fp16, [1, 188, 1024]> query_3_cast_fp16 = layer_norm(axes = query_3_axes_0, beta = module_layers_1_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_self_att_weight_to_fp16, x = input_83_cast_fp16)[name = string("query_3_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21042688))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21567040))))[name = string("module_layers_1_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_1_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_1_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21632640)))];
            tensor<fp16, [1, 188, 1024]> linear_12_cast_fp16 = linear(bias = module_layers_1_self_attn_linear_q_bias_to_fp16, weight = module_layers_1_self_attn_linear_q_weight_to_fp16_quantized, x = query_3_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<int32, [4]> var_574 = const()[name = string("op_574"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_7_cast_fp16 = reshape(shape = var_574, x = linear_12_cast_fp16)[name = string("q_7_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21634752))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22159104))))[name = string("module_layers_1_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_1_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_1_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22224704)))];
            tensor<fp16, [1, 188, 1024]> linear_13_cast_fp16 = linear(bias = module_layers_1_self_attn_linear_k_bias_to_fp16, weight = module_layers_1_self_attn_linear_k_weight_to_fp16_quantized, x = query_3_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<int32, [4]> var_579 = const()[name = string("op_579"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_5_cast_fp16 = reshape(shape = var_579, x = linear_13_cast_fp16)[name = string("k_5_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22226816))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22751168))))[name = string("module_layers_1_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_1_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_1_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22816768)))];
            tensor<fp16, [1, 188, 1024]> linear_14_cast_fp16 = linear(bias = module_layers_1_self_attn_linear_v_bias_to_fp16, weight = module_layers_1_self_attn_linear_v_weight_to_fp16_quantized, x = query_3_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<int32, [4]> var_584 = const()[name = string("op_584"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_3_cast_fp16 = reshape(shape = var_584, x = linear_14_cast_fp16)[name = string("v_3_cast_fp16")];
            tensor<int32, [4]> value_5_perm_0 = const()[name = string("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22818880))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22819456))))[name = string("module_layers_1_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_596_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_596_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_1_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22819584))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22820160))))[name = string("module_layers_1_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_598_cast_fp16 = add(x = q_7_cast_fp16, y = module_layers_1_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_598_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_3_perm_0 = const()[name = string("q_with_bias_v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_29_transpose_x_0 = const()[name = string("x_29_transpose_x_0"), val = bool(false)];
            bool x_29_transpose_y_0 = const()[name = string("x_29_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_600_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22820288))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23012352))))[name = string("op_600_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_3_cast_fp16 = transpose(perm = q_with_bias_v_3_perm_0, x = var_598_cast_fp16)[name = string("transpose_409")];
            tensor<fp16, [1, 8, 188, 375]> x_29_cast_fp16 = matmul(transpose_x = x_29_transpose_x_0, transpose_y = x_29_transpose_y_0, x = q_with_bias_v_3_cast_fp16, y = op_600_to_fp16_quantized)[name = string("x_29_cast_fp16")];
            tensor<int32, [8]> x_31_pad_0 = const()[name = string("x_31_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_31_mode_0 = const()[name = string("x_31_mode_0"), val = string("constant")];
            fp16 const_80_to_fp16 = const()[name = string("const_80_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_31_cast_fp16 = pad(constant_val = const_80_to_fp16, mode = x_31_mode_0, pad = x_31_pad_0, x = x_29_cast_fp16)[name = string("x_31_cast_fp16")];
            tensor<int32, [4]> var_608 = const()[name = string("op_608"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_33_cast_fp16 = reshape(shape = var_608, x = x_31_cast_fp16)[name = string("x_33_cast_fp16")];
            tensor<int32, [4]> var_612_begin_0 = const()[name = string("op_612_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_612_end_0 = const()[name = string("op_612_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_612_end_mask_0 = const()[name = string("op_612_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_612_cast_fp16 = slice_by_index(begin = var_612_begin_0, end = var_612_end_0, end_mask = var_612_end_mask_0, x = x_33_cast_fp16)[name = string("op_612_cast_fp16")];
            tensor<int32, [4]> var_613 = const()[name = string("op_613"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_5_cast_fp16 = reshape(shape = var_613, x = var_612_cast_fp16)[name = string("matrix_bd_5_cast_fp16")];
            bool matrix_ac_3_transpose_x_0 = const()[name = string("matrix_ac_3_transpose_x_0"), val = bool(false)];
            bool matrix_ac_3_transpose_y_0 = const()[name = string("matrix_ac_3_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_130_perm_0 = const()[name = string("transpose_130_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_131_perm_0 = const()[name = string("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_131 = transpose(perm = transpose_131_perm_0, x = k_5_cast_fp16)[name = string("transpose_407")];
            tensor<fp16, [1, 8, 188, 128]> transpose_130 = transpose(perm = transpose_130_perm_0, x = var_596_cast_fp16)[name = string("transpose_408")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_3_cast_fp16 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_130, y = transpose_131)[name = string("matrix_ac_3_cast_fp16")];
            tensor<int32, [4]> matrix_bd_7_begin_0 = const()[name = string("matrix_bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_7_end_0 = const()[name = string("matrix_bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_7_end_mask_0 = const()[name = string("matrix_bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_7_cast_fp16 = slice_by_index(begin = matrix_bd_7_begin_0, end = matrix_bd_7_end_0, end_mask = matrix_bd_7_end_mask_0, x = matrix_bd_5_cast_fp16)[name = string("matrix_bd_7_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_622_cast_fp16 = add(x = matrix_ac_3_cast_fp16, y = matrix_bd_7_cast_fp16)[name = string("op_622_cast_fp16")];
            fp16 _inversed_scores_5_y_0_to_fp16 = const()[name = string("_inversed_scores_5_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_5_cast_fp16 = mul(x = var_622_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = string("_inversed_scores_5_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_7_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_5_cast_fp16, cond = mask_11)[name = string("scores_7_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_628_cast_fp16 = softmax(axis = var_23, x = scores_7_cast_fp16)[name = string("op_628_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_85_cast_fp16 = select(a = var_11_to_fp16, b = var_628_cast_fp16, cond = mask_11)[name = string("input_85_cast_fp16")];
            bool x_35_transpose_x_0 = const()[name = string("x_35_transpose_x_0"), val = bool(false)];
            bool x_35_transpose_y_0 = const()[name = string("x_35_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = v_3_cast_fp16)[name = string("transpose_406")];
            tensor<fp16, [1, 8, 188, 128]> x_35_cast_fp16 = matmul(transpose_x = x_35_transpose_x_0, transpose_y = x_35_transpose_y_0, x = input_85_cast_fp16, y = value_5_cast_fp16)[name = string("x_35_cast_fp16")];
            tensor<int32, [4]> var_632_perm_0 = const()[name = string("op_632_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_633 = const()[name = string("op_633"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_632_cast_fp16 = transpose(perm = var_632_perm_0, x = x_35_cast_fp16)[name = string("transpose_405")];
            tensor<fp16, [1, 188, 1024]> input_87_cast_fp16 = reshape(shape = var_633, x = var_632_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_1_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23015424))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23539776))))[name = string("module_layers_1_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_1_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_1_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23605376)))];
            tensor<fp16, [1, 188, 1024]> linear_16_cast_fp16 = linear(bias = module_layers_1_self_attn_linear_out_bias_to_fp16, weight = module_layers_1_self_attn_linear_out_weight_to_fp16_quantized, x = input_87_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_16_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<int32, [1]> x_39_axes_0 = const()[name = string("x_39_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_1_norm_conv_weight_to_fp16 = const()[name = string("module_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23607488)))];
            tensor<fp16, [1024]> module_layers_1_norm_conv_bias_to_fp16 = const()[name = string("module_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23609600)))];
            tensor<fp16, [1, 188, 1024]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, beta = module_layers_1_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_conv_weight_to_fp16, x = input_91_cast_fp16)[name = string("x_39_cast_fp16")];
            tensor<int32, [3]> input_93_perm_0 = const()[name = string("input_93_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_95_pad_type_0 = const()[name = string("input_95_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_95_strides_0 = const()[name = string("input_95_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_95_pad_0 = const()[name = string("input_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_95_dilations_0 = const()[name = string("input_95_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_95_groups_0 = const()[name = string("input_95_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23611712))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24660352))))[name = string("module_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_1_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_1_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24791488)))];
            tensor<fp16, [1, 1024, 188]> input_93_cast_fp16 = transpose(perm = input_93_perm_0, x = x_39_cast_fp16)[name = string("transpose_404")];
            tensor<fp16, [1, 2048, 188]> input_95_cast_fp16 = conv(bias = module_layers_1_conv_pointwise_conv1_bias_to_fp16, dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = module_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_93_cast_fp16)[name = string("input_95_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, 188]> x_41_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_95_cast_fp16)[name = string("x_41_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_97_cast_fp16 = select(a = var_11_to_fp16, b = x_41_cast_fp16, cond = var_483)[name = string("input_97_cast_fp16")];
            tensor<int32, [6]> input_99_pad_0 = const()[name = string("input_99_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_99_mode_0 = const()[name = string("input_99_mode_0"), val = string("constant")];
            fp16 const_83_to_fp16 = const()[name = string("const_83_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_99_cast_fp16 = pad(constant_val = const_83_to_fp16, mode = input_99_mode_0, pad = input_99_pad_0, x = input_97_cast_fp16)[name = string("input_99_cast_fp16")];
            string input_101_pad_type_0 = const()[name = string("input_101_pad_type_0"), val = string("valid")];
            int32 input_101_groups_0 = const()[name = string("input_101_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_101_strides_0 = const()[name = string("input_101_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_101_pad_0 = const()[name = string("input_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_101_dilations_0 = const()[name = string("input_101_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_386_to_fp16 = const()[name = string("const_386_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24795648)))];
            tensor<fp16, [1024]> const_387_to_fp16 = const()[name = string("const_387_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24814144)))];
            tensor<fp16, [1, 1024, 188]> input_103_cast_fp16 = conv(bias = const_387_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_386_to_fp16, x = input_99_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_105_cast_fp16 = silu(x = input_103_cast_fp16)[name = string("input_105_cast_fp16")];
            string x_43_pad_type_0 = const()[name = string("x_43_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_43_strides_0 = const()[name = string("x_43_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_43_pad_0 = const()[name = string("x_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_43_dilations_0 = const()[name = string("x_43_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_43_groups_0 = const()[name = string("x_43_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24816256))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25340608))))[name = string("module_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_1_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_1_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25406208)))];
            tensor<fp16, [1, 1024, 188]> x_43_cast_fp16 = conv(bias = module_layers_1_conv_pointwise_conv2_bias_to_fp16, dilations = x_43_dilations_0, groups = x_43_groups_0, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = x_43_strides_0, weight = module_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_105_cast_fp16)[name = string("x_43_cast_fp16")];
            tensor<int32, [3]> input_107_perm_0 = const()[name = string("input_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_107_cast_fp16 = transpose(perm = input_107_perm_0, x = x_43_cast_fp16)[name = string("transpose_403")];
            tensor<fp16, [1, 188, 1024]> input_109_cast_fp16 = add(x = input_91_cast_fp16, y = input_107_cast_fp16)[name = string("input_109_cast_fp16")];
            tensor<int32, [1]> input_111_axes_0 = const()[name = string("input_111_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25408320)))];
            tensor<fp16, [1024]> module_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25410432)))];
            tensor<fp16, [1, 188, 1024]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = module_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = string("input_111_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_1_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25412544))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27509760))))[name = string("module_layers_1_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_1_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_1_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27771968)))];
            tensor<fp16, [1, 188, 4096]> linear_17_cast_fp16 = linear(bias = module_layers_1_feed_forward2_linear1_bias_to_fp16, weight = module_layers_1_feed_forward2_linear1_weight_to_fp16_quantized, x = input_111_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_115_cast_fp16 = silu(x = linear_17_cast_fp16)[name = string("input_115_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_1_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27780224))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29877440))))[name = string("module_layers_1_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_1_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_1_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30139648)))];
            tensor<fp16, [1, 188, 1024]> linear_18_cast_fp16 = linear(bias = module_layers_1_feed_forward2_linear2_bias_to_fp16, weight = module_layers_1_feed_forward2_linear2_weight_to_fp16_quantized, x = input_115_cast_fp16)[name = string("linear_18_cast_fp16")];
            fp16 var_699_to_fp16 = const()[name = string("op_699_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_700_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_699_to_fp16)[name = string("op_700_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_121_cast_fp16 = add(x = input_109_cast_fp16, y = var_700_cast_fp16)[name = string("input_121_cast_fp16")];
            tensor<int32, [1]> input_123_axes_0 = const()[name = string("input_123_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_1_norm_out_weight_to_fp16 = const()[name = string("module_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30141760)))];
            tensor<fp16, [1024]> module_layers_1_norm_out_bias_to_fp16 = const()[name = string("module_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30143872)))];
            tensor<fp16, [1, 188, 1024]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = module_layers_1_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_1_norm_out_weight_to_fp16, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
            tensor<int32, [1]> input_125_axes_0 = const()[name = string("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30145984)))];
            tensor<fp16, [1024]> module_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30148096)))];
            tensor<fp16, [1, 188, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = module_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward1_weight_to_fp16, x = input_123_cast_fp16)[name = string("input_125_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_2_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30150208))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32247424))))[name = string("module_layers_2_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_2_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_2_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32509632)))];
            tensor<fp16, [1, 188, 4096]> linear_19_cast_fp16 = linear(bias = module_layers_2_feed_forward1_linear1_bias_to_fp16, weight = module_layers_2_feed_forward1_linear1_weight_to_fp16_quantized, x = input_125_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_129_cast_fp16 = silu(x = linear_19_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_2_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32517888))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34615104))))[name = string("module_layers_2_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_2_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_2_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34877312)))];
            tensor<fp16, [1, 188, 1024]> linear_20_cast_fp16 = linear(bias = module_layers_2_feed_forward1_linear2_bias_to_fp16, weight = module_layers_2_feed_forward1_linear2_weight_to_fp16_quantized, x = input_129_cast_fp16)[name = string("linear_20_cast_fp16")];
            fp16 var_730_to_fp16 = const()[name = string("op_730_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_731_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_730_to_fp16)[name = string("op_731_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_135_cast_fp16 = add(x = input_123_cast_fp16, y = var_731_cast_fp16)[name = string("input_135_cast_fp16")];
            tensor<int32, [1]> query_5_axes_0 = const()[name = string("query_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_2_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34879424)))];
            tensor<fp16, [1024]> module_layers_2_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34881536)))];
            tensor<fp16, [1, 188, 1024]> query_5_cast_fp16 = layer_norm(axes = query_5_axes_0, beta = module_layers_2_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_self_att_weight_to_fp16, x = input_135_cast_fp16)[name = string("query_5_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34883648))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35408000))))[name = string("module_layers_2_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_2_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_2_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35473600)))];
            tensor<fp16, [1, 188, 1024]> linear_21_cast_fp16 = linear(bias = module_layers_2_self_attn_linear_q_bias_to_fp16, weight = module_layers_2_self_attn_linear_q_weight_to_fp16_quantized, x = query_5_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<int32, [4]> var_748 = const()[name = string("op_748"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_13_cast_fp16 = reshape(shape = var_748, x = linear_21_cast_fp16)[name = string("q_13_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35475712))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36000064))))[name = string("module_layers_2_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_2_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_2_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36065664)))];
            tensor<fp16, [1, 188, 1024]> linear_22_cast_fp16 = linear(bias = module_layers_2_self_attn_linear_k_bias_to_fp16, weight = module_layers_2_self_attn_linear_k_weight_to_fp16_quantized, x = query_5_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [4]> var_753 = const()[name = string("op_753"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_9_cast_fp16 = reshape(shape = var_753, x = linear_22_cast_fp16)[name = string("k_9_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36067776))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36592128))))[name = string("module_layers_2_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_2_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_2_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36657728)))];
            tensor<fp16, [1, 188, 1024]> linear_23_cast_fp16 = linear(bias = module_layers_2_self_attn_linear_v_bias_to_fp16, weight = module_layers_2_self_attn_linear_v_weight_to_fp16_quantized, x = query_5_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<int32, [4]> var_758 = const()[name = string("op_758"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_5_cast_fp16 = reshape(shape = var_758, x = linear_23_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> value_7_perm_0 = const()[name = string("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36659840))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36660416))))[name = string("module_layers_2_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_770_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_770_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_2_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36660544))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36661120))))[name = string("module_layers_2_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_772_cast_fp16 = add(x = q_13_cast_fp16, y = module_layers_2_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_772_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_5_perm_0 = const()[name = string("q_with_bias_v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_51_transpose_x_0 = const()[name = string("x_51_transpose_x_0"), val = bool(false)];
            bool x_51_transpose_y_0 = const()[name = string("x_51_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_774_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36661248))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36853312))))[name = string("op_774_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_5_cast_fp16 = transpose(perm = q_with_bias_v_5_perm_0, x = var_772_cast_fp16)[name = string("transpose_402")];
            tensor<fp16, [1, 8, 188, 375]> x_51_cast_fp16 = matmul(transpose_x = x_51_transpose_x_0, transpose_y = x_51_transpose_y_0, x = q_with_bias_v_5_cast_fp16, y = op_774_to_fp16_quantized)[name = string("x_51_cast_fp16")];
            tensor<int32, [8]> x_53_pad_0 = const()[name = string("x_53_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("constant")];
            fp16 const_90_to_fp16 = const()[name = string("const_90_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_53_cast_fp16 = pad(constant_val = const_90_to_fp16, mode = x_53_mode_0, pad = x_53_pad_0, x = x_51_cast_fp16)[name = string("x_53_cast_fp16")];
            tensor<int32, [4]> var_782 = const()[name = string("op_782"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_55_cast_fp16 = reshape(shape = var_782, x = x_53_cast_fp16)[name = string("x_55_cast_fp16")];
            tensor<int32, [4]> var_786_begin_0 = const()[name = string("op_786_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_786_end_0 = const()[name = string("op_786_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_786_end_mask_0 = const()[name = string("op_786_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_786_cast_fp16 = slice_by_index(begin = var_786_begin_0, end = var_786_end_0, end_mask = var_786_end_mask_0, x = x_55_cast_fp16)[name = string("op_786_cast_fp16")];
            tensor<int32, [4]> var_787 = const()[name = string("op_787"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_9_cast_fp16 = reshape(shape = var_787, x = var_786_cast_fp16)[name = string("matrix_bd_9_cast_fp16")];
            bool matrix_ac_5_transpose_x_0 = const()[name = string("matrix_ac_5_transpose_x_0"), val = bool(false)];
            bool matrix_ac_5_transpose_y_0 = const()[name = string("matrix_ac_5_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_132_perm_0 = const()[name = string("transpose_132_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_133_perm_0 = const()[name = string("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_133 = transpose(perm = transpose_133_perm_0, x = k_9_cast_fp16)[name = string("transpose_400")];
            tensor<fp16, [1, 8, 188, 128]> transpose_132 = transpose(perm = transpose_132_perm_0, x = var_770_cast_fp16)[name = string("transpose_401")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_5_cast_fp16 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_132, y = transpose_133)[name = string("matrix_ac_5_cast_fp16")];
            tensor<int32, [4]> matrix_bd_11_begin_0 = const()[name = string("matrix_bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_11_end_0 = const()[name = string("matrix_bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_11_end_mask_0 = const()[name = string("matrix_bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_11_cast_fp16 = slice_by_index(begin = matrix_bd_11_begin_0, end = matrix_bd_11_end_0, end_mask = matrix_bd_11_end_mask_0, x = matrix_bd_9_cast_fp16)[name = string("matrix_bd_11_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_796_cast_fp16 = add(x = matrix_ac_5_cast_fp16, y = matrix_bd_11_cast_fp16)[name = string("op_796_cast_fp16")];
            fp16 _inversed_scores_9_y_0_to_fp16 = const()[name = string("_inversed_scores_9_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_9_cast_fp16 = mul(x = var_796_cast_fp16, y = _inversed_scores_9_y_0_to_fp16)[name = string("_inversed_scores_9_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_11_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_9_cast_fp16, cond = mask_11)[name = string("scores_11_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_802_cast_fp16 = softmax(axis = var_23, x = scores_11_cast_fp16)[name = string("op_802_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_137_cast_fp16 = select(a = var_11_to_fp16, b = var_802_cast_fp16, cond = mask_11)[name = string("input_137_cast_fp16")];
            bool x_57_transpose_x_0 = const()[name = string("x_57_transpose_x_0"), val = bool(false)];
            bool x_57_transpose_y_0 = const()[name = string("x_57_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = v_5_cast_fp16)[name = string("transpose_399")];
            tensor<fp16, [1, 8, 188, 128]> x_57_cast_fp16 = matmul(transpose_x = x_57_transpose_x_0, transpose_y = x_57_transpose_y_0, x = input_137_cast_fp16, y = value_7_cast_fp16)[name = string("x_57_cast_fp16")];
            tensor<int32, [4]> var_806_perm_0 = const()[name = string("op_806_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_807 = const()[name = string("op_807"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_806_cast_fp16 = transpose(perm = var_806_perm_0, x = x_57_cast_fp16)[name = string("transpose_398")];
            tensor<fp16, [1, 188, 1024]> input_139_cast_fp16 = reshape(shape = var_807, x = var_806_cast_fp16)[name = string("input_139_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_2_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36856384))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37380736))))[name = string("module_layers_2_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_2_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_2_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37446336)))];
            tensor<fp16, [1, 188, 1024]> linear_25_cast_fp16 = linear(bias = module_layers_2_self_attn_linear_out_bias_to_fp16, weight = module_layers_2_self_attn_linear_out_weight_to_fp16_quantized, x = input_139_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_143_cast_fp16 = add(x = input_135_cast_fp16, y = linear_25_cast_fp16)[name = string("input_143_cast_fp16")];
            tensor<int32, [1]> x_61_axes_0 = const()[name = string("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_2_norm_conv_weight_to_fp16 = const()[name = string("module_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37448448)))];
            tensor<fp16, [1024]> module_layers_2_norm_conv_bias_to_fp16 = const()[name = string("module_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37450560)))];
            tensor<fp16, [1, 188, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = module_layers_2_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_conv_weight_to_fp16, x = input_143_cast_fp16)[name = string("x_61_cast_fp16")];
            tensor<int32, [3]> input_145_perm_0 = const()[name = string("input_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_147_pad_type_0 = const()[name = string("input_147_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_147_strides_0 = const()[name = string("input_147_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_147_pad_0 = const()[name = string("input_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_147_dilations_0 = const()[name = string("input_147_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_147_groups_0 = const()[name = string("input_147_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37452672))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38501312))))[name = string("module_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_2_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_2_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38632448)))];
            tensor<fp16, [1, 1024, 188]> input_145_cast_fp16 = transpose(perm = input_145_perm_0, x = x_61_cast_fp16)[name = string("transpose_397")];
            tensor<fp16, [1, 2048, 188]> input_147_cast_fp16 = conv(bias = module_layers_2_conv_pointwise_conv1_bias_to_fp16, dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = module_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_145_cast_fp16)[name = string("input_147_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, 188]> x_63_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_147_cast_fp16)[name = string("x_63_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_149_cast_fp16 = select(a = var_11_to_fp16, b = x_63_cast_fp16, cond = var_483)[name = string("input_149_cast_fp16")];
            tensor<int32, [6]> input_151_pad_0 = const()[name = string("input_151_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_151_mode_0 = const()[name = string("input_151_mode_0"), val = string("constant")];
            fp16 const_93_to_fp16 = const()[name = string("const_93_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_151_cast_fp16 = pad(constant_val = const_93_to_fp16, mode = input_151_mode_0, pad = input_151_pad_0, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
            string input_153_pad_type_0 = const()[name = string("input_153_pad_type_0"), val = string("valid")];
            int32 input_153_groups_0 = const()[name = string("input_153_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_153_strides_0 = const()[name = string("input_153_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_153_pad_0 = const()[name = string("input_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_153_dilations_0 = const()[name = string("input_153_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_388_to_fp16 = const()[name = string("const_388_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38636608)))];
            tensor<fp16, [1024]> const_389_to_fp16 = const()[name = string("const_389_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38655104)))];
            tensor<fp16, [1, 1024, 188]> input_155_cast_fp16 = conv(bias = const_389_to_fp16, dilations = input_153_dilations_0, groups = input_153_groups_0, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = input_153_strides_0, weight = const_388_to_fp16, x = input_151_cast_fp16)[name = string("input_155_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_157_cast_fp16 = silu(x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
            string x_65_pad_type_0 = const()[name = string("x_65_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_65_strides_0 = const()[name = string("x_65_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_65_pad_0 = const()[name = string("x_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_65_dilations_0 = const()[name = string("x_65_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_65_groups_0 = const()[name = string("x_65_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38657216))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39181568))))[name = string("module_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_2_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_2_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39247168)))];
            tensor<fp16, [1, 1024, 188]> x_65_cast_fp16 = conv(bias = module_layers_2_conv_pointwise_conv2_bias_to_fp16, dilations = x_65_dilations_0, groups = x_65_groups_0, pad = x_65_pad_0, pad_type = x_65_pad_type_0, strides = x_65_strides_0, weight = module_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_157_cast_fp16)[name = string("x_65_cast_fp16")];
            tensor<int32, [3]> input_159_perm_0 = const()[name = string("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_65_cast_fp16)[name = string("transpose_396")];
            tensor<fp16, [1, 188, 1024]> input_161_cast_fp16 = add(x = input_143_cast_fp16, y = input_159_cast_fp16)[name = string("input_161_cast_fp16")];
            tensor<int32, [1]> input_163_axes_0 = const()[name = string("input_163_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39249280)))];
            tensor<fp16, [1024]> module_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39251392)))];
            tensor<fp16, [1, 188, 1024]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = module_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_feed_forward2_weight_to_fp16, x = input_161_cast_fp16)[name = string("input_163_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_2_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39253504))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41350720))))[name = string("module_layers_2_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_2_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_2_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41612928)))];
            tensor<fp16, [1, 188, 4096]> linear_26_cast_fp16 = linear(bias = module_layers_2_feed_forward2_linear1_bias_to_fp16, weight = module_layers_2_feed_forward2_linear1_weight_to_fp16_quantized, x = input_163_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_167_cast_fp16 = silu(x = linear_26_cast_fp16)[name = string("input_167_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_2_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41621184))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43718400))))[name = string("module_layers_2_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_2_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_2_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43980608)))];
            tensor<fp16, [1, 188, 1024]> linear_27_cast_fp16 = linear(bias = module_layers_2_feed_forward2_linear2_bias_to_fp16, weight = module_layers_2_feed_forward2_linear2_weight_to_fp16_quantized, x = input_167_cast_fp16)[name = string("linear_27_cast_fp16")];
            fp16 var_873_to_fp16 = const()[name = string("op_873_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_874_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_873_to_fp16)[name = string("op_874_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_173_cast_fp16 = add(x = input_161_cast_fp16, y = var_874_cast_fp16)[name = string("input_173_cast_fp16")];
            tensor<int32, [1]> input_175_axes_0 = const()[name = string("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_2_norm_out_weight_to_fp16 = const()[name = string("module_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43982720)))];
            tensor<fp16, [1024]> module_layers_2_norm_out_bias_to_fp16 = const()[name = string("module_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43984832)))];
            tensor<fp16, [1, 188, 1024]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = module_layers_2_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_2_norm_out_weight_to_fp16, x = input_173_cast_fp16)[name = string("input_175_cast_fp16")];
            tensor<int32, [1]> input_177_axes_0 = const()[name = string("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43986944)))];
            tensor<fp16, [1024]> module_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43989056)))];
            tensor<fp16, [1, 188, 1024]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = module_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward1_weight_to_fp16, x = input_175_cast_fp16)[name = string("input_177_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_3_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43991168))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46088384))))[name = string("module_layers_3_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_3_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_3_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46350592)))];
            tensor<fp16, [1, 188, 4096]> linear_28_cast_fp16 = linear(bias = module_layers_3_feed_forward1_linear1_bias_to_fp16, weight = module_layers_3_feed_forward1_linear1_weight_to_fp16_quantized, x = input_177_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_181_cast_fp16 = silu(x = linear_28_cast_fp16)[name = string("input_181_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_3_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46358848))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48456064))))[name = string("module_layers_3_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_3_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_3_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48718272)))];
            tensor<fp16, [1, 188, 1024]> linear_29_cast_fp16 = linear(bias = module_layers_3_feed_forward1_linear2_bias_to_fp16, weight = module_layers_3_feed_forward1_linear2_weight_to_fp16_quantized, x = input_181_cast_fp16)[name = string("linear_29_cast_fp16")];
            fp16 var_904_to_fp16 = const()[name = string("op_904_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_905_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_904_to_fp16)[name = string("op_905_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_187_cast_fp16 = add(x = input_175_cast_fp16, y = var_905_cast_fp16)[name = string("input_187_cast_fp16")];
            tensor<int32, [1]> query_7_axes_0 = const()[name = string("query_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_3_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48720384)))];
            tensor<fp16, [1024]> module_layers_3_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48722496)))];
            tensor<fp16, [1, 188, 1024]> query_7_cast_fp16 = layer_norm(axes = query_7_axes_0, beta = module_layers_3_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_self_att_weight_to_fp16, x = input_187_cast_fp16)[name = string("query_7_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48724608))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49248960))))[name = string("module_layers_3_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_3_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_3_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49314560)))];
            tensor<fp16, [1, 188, 1024]> linear_30_cast_fp16 = linear(bias = module_layers_3_self_attn_linear_q_bias_to_fp16, weight = module_layers_3_self_attn_linear_q_weight_to_fp16_quantized, x = query_7_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> var_922 = const()[name = string("op_922"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_19_cast_fp16 = reshape(shape = var_922, x = linear_30_cast_fp16)[name = string("q_19_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49316672))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49841024))))[name = string("module_layers_3_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_3_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_3_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49906624)))];
            tensor<fp16, [1, 188, 1024]> linear_31_cast_fp16 = linear(bias = module_layers_3_self_attn_linear_k_bias_to_fp16, weight = module_layers_3_self_attn_linear_k_weight_to_fp16_quantized, x = query_7_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<int32, [4]> var_927 = const()[name = string("op_927"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_13_cast_fp16 = reshape(shape = var_927, x = linear_31_cast_fp16)[name = string("k_13_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49908736))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50433088))))[name = string("module_layers_3_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_3_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_3_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50498688)))];
            tensor<fp16, [1, 188, 1024]> linear_32_cast_fp16 = linear(bias = module_layers_3_self_attn_linear_v_bias_to_fp16, weight = module_layers_3_self_attn_linear_v_weight_to_fp16_quantized, x = query_7_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<int32, [4]> var_932 = const()[name = string("op_932"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_7_cast_fp16 = reshape(shape = var_932, x = linear_32_cast_fp16)[name = string("v_7_cast_fp16")];
            tensor<int32, [4]> value_9_perm_0 = const()[name = string("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50500800))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50501376))))[name = string("module_layers_3_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_944_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_944_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_3_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50501504))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50502080))))[name = string("module_layers_3_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_946_cast_fp16 = add(x = q_19_cast_fp16, y = module_layers_3_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_946_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_7_perm_0 = const()[name = string("q_with_bias_v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_73_transpose_x_0 = const()[name = string("x_73_transpose_x_0"), val = bool(false)];
            bool x_73_transpose_y_0 = const()[name = string("x_73_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_948_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50502208))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50694272))))[name = string("op_948_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_7_cast_fp16 = transpose(perm = q_with_bias_v_7_perm_0, x = var_946_cast_fp16)[name = string("transpose_395")];
            tensor<fp16, [1, 8, 188, 375]> x_73_cast_fp16 = matmul(transpose_x = x_73_transpose_x_0, transpose_y = x_73_transpose_y_0, x = q_with_bias_v_7_cast_fp16, y = op_948_to_fp16_quantized)[name = string("x_73_cast_fp16")];
            tensor<int32, [8]> x_75_pad_0 = const()[name = string("x_75_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_75_mode_0 = const()[name = string("x_75_mode_0"), val = string("constant")];
            fp16 const_100_to_fp16 = const()[name = string("const_100_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_75_cast_fp16 = pad(constant_val = const_100_to_fp16, mode = x_75_mode_0, pad = x_75_pad_0, x = x_73_cast_fp16)[name = string("x_75_cast_fp16")];
            tensor<int32, [4]> var_956 = const()[name = string("op_956"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_77_cast_fp16 = reshape(shape = var_956, x = x_75_cast_fp16)[name = string("x_77_cast_fp16")];
            tensor<int32, [4]> var_960_begin_0 = const()[name = string("op_960_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_960_end_0 = const()[name = string("op_960_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_960_end_mask_0 = const()[name = string("op_960_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_960_cast_fp16 = slice_by_index(begin = var_960_begin_0, end = var_960_end_0, end_mask = var_960_end_mask_0, x = x_77_cast_fp16)[name = string("op_960_cast_fp16")];
            tensor<int32, [4]> var_961 = const()[name = string("op_961"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_13_cast_fp16 = reshape(shape = var_961, x = var_960_cast_fp16)[name = string("matrix_bd_13_cast_fp16")];
            bool matrix_ac_7_transpose_x_0 = const()[name = string("matrix_ac_7_transpose_x_0"), val = bool(false)];
            bool matrix_ac_7_transpose_y_0 = const()[name = string("matrix_ac_7_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_134_perm_0 = const()[name = string("transpose_134_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_135_perm_0 = const()[name = string("transpose_135_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_135 = transpose(perm = transpose_135_perm_0, x = k_13_cast_fp16)[name = string("transpose_393")];
            tensor<fp16, [1, 8, 188, 128]> transpose_134 = transpose(perm = transpose_134_perm_0, x = var_944_cast_fp16)[name = string("transpose_394")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_7_cast_fp16 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_134, y = transpose_135)[name = string("matrix_ac_7_cast_fp16")];
            tensor<int32, [4]> matrix_bd_15_begin_0 = const()[name = string("matrix_bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_15_end_0 = const()[name = string("matrix_bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_15_end_mask_0 = const()[name = string("matrix_bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_15_cast_fp16 = slice_by_index(begin = matrix_bd_15_begin_0, end = matrix_bd_15_end_0, end_mask = matrix_bd_15_end_mask_0, x = matrix_bd_13_cast_fp16)[name = string("matrix_bd_15_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_970_cast_fp16 = add(x = matrix_ac_7_cast_fp16, y = matrix_bd_15_cast_fp16)[name = string("op_970_cast_fp16")];
            fp16 _inversed_scores_13_y_0_to_fp16 = const()[name = string("_inversed_scores_13_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_13_cast_fp16 = mul(x = var_970_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = string("_inversed_scores_13_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_15_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_13_cast_fp16, cond = mask_11)[name = string("scores_15_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_976_cast_fp16 = softmax(axis = var_23, x = scores_15_cast_fp16)[name = string("op_976_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_189_cast_fp16 = select(a = var_11_to_fp16, b = var_976_cast_fp16, cond = mask_11)[name = string("input_189_cast_fp16")];
            bool x_79_transpose_x_0 = const()[name = string("x_79_transpose_x_0"), val = bool(false)];
            bool x_79_transpose_y_0 = const()[name = string("x_79_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = v_7_cast_fp16)[name = string("transpose_392")];
            tensor<fp16, [1, 8, 188, 128]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = input_189_cast_fp16, y = value_9_cast_fp16)[name = string("x_79_cast_fp16")];
            tensor<int32, [4]> var_980_perm_0 = const()[name = string("op_980_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_981 = const()[name = string("op_981"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_980_cast_fp16 = transpose(perm = var_980_perm_0, x = x_79_cast_fp16)[name = string("transpose_391")];
            tensor<fp16, [1, 188, 1024]> input_191_cast_fp16 = reshape(shape = var_981, x = var_980_cast_fp16)[name = string("input_191_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_3_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(50697344))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51221696))))[name = string("module_layers_3_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_3_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_3_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51287296)))];
            tensor<fp16, [1, 188, 1024]> linear_34_cast_fp16 = linear(bias = module_layers_3_self_attn_linear_out_bias_to_fp16, weight = module_layers_3_self_attn_linear_out_weight_to_fp16_quantized, x = input_191_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_34_cast_fp16)[name = string("input_195_cast_fp16")];
            tensor<int32, [1]> x_83_axes_0 = const()[name = string("x_83_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_3_norm_conv_weight_to_fp16 = const()[name = string("module_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51289408)))];
            tensor<fp16, [1024]> module_layers_3_norm_conv_bias_to_fp16 = const()[name = string("module_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51291520)))];
            tensor<fp16, [1, 188, 1024]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = module_layers_3_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = string("x_83_cast_fp16")];
            tensor<int32, [3]> input_197_perm_0 = const()[name = string("input_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_199_pad_type_0 = const()[name = string("input_199_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_199_strides_0 = const()[name = string("input_199_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_199_pad_0 = const()[name = string("input_199_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_199_dilations_0 = const()[name = string("input_199_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_199_groups_0 = const()[name = string("input_199_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51293632))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52342272))))[name = string("module_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_3_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_3_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52473408)))];
            tensor<fp16, [1, 1024, 188]> input_197_cast_fp16 = transpose(perm = input_197_perm_0, x = x_83_cast_fp16)[name = string("transpose_390")];
            tensor<fp16, [1, 2048, 188]> input_199_cast_fp16 = conv(bias = module_layers_3_conv_pointwise_conv1_bias_to_fp16, dilations = input_199_dilations_0, groups = input_199_groups_0, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = input_199_strides_0, weight = module_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_197_cast_fp16)[name = string("input_199_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, 188]> x_85_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_199_cast_fp16)[name = string("x_85_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_201_cast_fp16 = select(a = var_11_to_fp16, b = x_85_cast_fp16, cond = var_483)[name = string("input_201_cast_fp16")];
            tensor<int32, [6]> input_203_pad_0 = const()[name = string("input_203_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_203_mode_0 = const()[name = string("input_203_mode_0"), val = string("constant")];
            fp16 const_103_to_fp16 = const()[name = string("const_103_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_203_cast_fp16 = pad(constant_val = const_103_to_fp16, mode = input_203_mode_0, pad = input_203_pad_0, x = input_201_cast_fp16)[name = string("input_203_cast_fp16")];
            string input_205_pad_type_0 = const()[name = string("input_205_pad_type_0"), val = string("valid")];
            int32 input_205_groups_0 = const()[name = string("input_205_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_205_strides_0 = const()[name = string("input_205_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_205_pad_0 = const()[name = string("input_205_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_205_dilations_0 = const()[name = string("input_205_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_390_to_fp16 = const()[name = string("const_390_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52477568)))];
            tensor<fp16, [1024]> const_391_to_fp16 = const()[name = string("const_391_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52496064)))];
            tensor<fp16, [1, 1024, 188]> input_207_cast_fp16 = conv(bias = const_391_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = const_390_to_fp16, x = input_203_cast_fp16)[name = string("input_207_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_209_cast_fp16 = silu(x = input_207_cast_fp16)[name = string("input_209_cast_fp16")];
            string x_87_pad_type_0 = const()[name = string("x_87_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_87_strides_0 = const()[name = string("x_87_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_87_pad_0 = const()[name = string("x_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_87_dilations_0 = const()[name = string("x_87_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_87_groups_0 = const()[name = string("x_87_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52498176))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53022528))))[name = string("module_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_3_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_3_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53088128)))];
            tensor<fp16, [1, 1024, 188]> x_87_cast_fp16 = conv(bias = module_layers_3_conv_pointwise_conv2_bias_to_fp16, dilations = x_87_dilations_0, groups = x_87_groups_0, pad = x_87_pad_0, pad_type = x_87_pad_type_0, strides = x_87_strides_0, weight = module_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_209_cast_fp16)[name = string("x_87_cast_fp16")];
            tensor<int32, [3]> input_211_perm_0 = const()[name = string("input_211_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_211_cast_fp16 = transpose(perm = input_211_perm_0, x = x_87_cast_fp16)[name = string("transpose_389")];
            tensor<fp16, [1, 188, 1024]> input_213_cast_fp16 = add(x = input_195_cast_fp16, y = input_211_cast_fp16)[name = string("input_213_cast_fp16")];
            tensor<int32, [1]> input_215_axes_0 = const()[name = string("input_215_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53090240)))];
            tensor<fp16, [1024]> module_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53092352)))];
            tensor<fp16, [1, 188, 1024]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = module_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_feed_forward2_weight_to_fp16, x = input_213_cast_fp16)[name = string("input_215_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_3_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53094464))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55191680))))[name = string("module_layers_3_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_3_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_3_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55453888)))];
            tensor<fp16, [1, 188, 4096]> linear_35_cast_fp16 = linear(bias = module_layers_3_feed_forward2_linear1_bias_to_fp16, weight = module_layers_3_feed_forward2_linear1_weight_to_fp16_quantized, x = input_215_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_219_cast_fp16 = silu(x = linear_35_cast_fp16)[name = string("input_219_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_3_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(55462144))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57559360))))[name = string("module_layers_3_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_3_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_3_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57821568)))];
            tensor<fp16, [1, 188, 1024]> linear_36_cast_fp16 = linear(bias = module_layers_3_feed_forward2_linear2_bias_to_fp16, weight = module_layers_3_feed_forward2_linear2_weight_to_fp16_quantized, x = input_219_cast_fp16)[name = string("linear_36_cast_fp16")];
            fp16 var_1047_to_fp16 = const()[name = string("op_1047_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1048_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_1047_to_fp16)[name = string("op_1048_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_225_cast_fp16 = add(x = input_213_cast_fp16, y = var_1048_cast_fp16)[name = string("input_225_cast_fp16")];
            tensor<int32, [1]> input_227_axes_0 = const()[name = string("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_3_norm_out_weight_to_fp16 = const()[name = string("module_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57823680)))];
            tensor<fp16, [1024]> module_layers_3_norm_out_bias_to_fp16 = const()[name = string("module_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57825792)))];
            tensor<fp16, [1, 188, 1024]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = module_layers_3_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_3_norm_out_weight_to_fp16, x = input_225_cast_fp16)[name = string("input_227_cast_fp16")];
            tensor<int32, [1]> input_229_axes_0 = const()[name = string("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57827904)))];
            tensor<fp16, [1024]> module_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57830016)))];
            tensor<fp16, [1, 188, 1024]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = module_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward1_weight_to_fp16, x = input_227_cast_fp16)[name = string("input_229_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_4_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(57832128))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(59929344))))[name = string("module_layers_4_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_4_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_4_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60191552)))];
            tensor<fp16, [1, 188, 4096]> linear_37_cast_fp16 = linear(bias = module_layers_4_feed_forward1_linear1_bias_to_fp16, weight = module_layers_4_feed_forward1_linear1_weight_to_fp16_quantized, x = input_229_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_233_cast_fp16 = silu(x = linear_37_cast_fp16)[name = string("input_233_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_4_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60199808))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62297024))))[name = string("module_layers_4_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_4_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_4_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62559232)))];
            tensor<fp16, [1, 188, 1024]> linear_38_cast_fp16 = linear(bias = module_layers_4_feed_forward1_linear2_bias_to_fp16, weight = module_layers_4_feed_forward1_linear2_weight_to_fp16_quantized, x = input_233_cast_fp16)[name = string("linear_38_cast_fp16")];
            fp16 var_1078_to_fp16 = const()[name = string("op_1078_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1079_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_1078_to_fp16)[name = string("op_1079_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_239_cast_fp16 = add(x = input_227_cast_fp16, y = var_1079_cast_fp16)[name = string("input_239_cast_fp16")];
            tensor<int32, [1]> query_9_axes_0 = const()[name = string("query_9_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_4_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62561344)))];
            tensor<fp16, [1024]> module_layers_4_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62563456)))];
            tensor<fp16, [1, 188, 1024]> query_9_cast_fp16 = layer_norm(axes = query_9_axes_0, beta = module_layers_4_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = string("query_9_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62565568))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63089920))))[name = string("module_layers_4_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_4_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_4_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63155520)))];
            tensor<fp16, [1, 188, 1024]> linear_39_cast_fp16 = linear(bias = module_layers_4_self_attn_linear_q_bias_to_fp16, weight = module_layers_4_self_attn_linear_q_weight_to_fp16_quantized, x = query_9_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<int32, [4]> var_1096 = const()[name = string("op_1096"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_25_cast_fp16 = reshape(shape = var_1096, x = linear_39_cast_fp16)[name = string("q_25_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63157632))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63681984))))[name = string("module_layers_4_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_4_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_4_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63747584)))];
            tensor<fp16, [1, 188, 1024]> linear_40_cast_fp16 = linear(bias = module_layers_4_self_attn_linear_k_bias_to_fp16, weight = module_layers_4_self_attn_linear_k_weight_to_fp16_quantized, x = query_9_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<int32, [4]> var_1101 = const()[name = string("op_1101"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_17_cast_fp16 = reshape(shape = var_1101, x = linear_40_cast_fp16)[name = string("k_17_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63749696))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64274048))))[name = string("module_layers_4_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_4_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_4_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64339648)))];
            tensor<fp16, [1, 188, 1024]> linear_41_cast_fp16 = linear(bias = module_layers_4_self_attn_linear_v_bias_to_fp16, weight = module_layers_4_self_attn_linear_v_weight_to_fp16_quantized, x = query_9_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<int32, [4]> var_1106 = const()[name = string("op_1106"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_9_cast_fp16 = reshape(shape = var_1106, x = linear_41_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> value_11_perm_0 = const()[name = string("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64341760))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64342336))))[name = string("module_layers_4_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1118_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_1118_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_4_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64342464))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64343040))))[name = string("module_layers_4_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1120_cast_fp16 = add(x = q_25_cast_fp16, y = module_layers_4_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_1120_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_9_perm_0 = const()[name = string("q_with_bias_v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_95_transpose_x_0 = const()[name = string("x_95_transpose_x_0"), val = bool(false)];
            bool x_95_transpose_y_0 = const()[name = string("x_95_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_1122_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64343168))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64535232))))[name = string("op_1122_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_9_cast_fp16 = transpose(perm = q_with_bias_v_9_perm_0, x = var_1120_cast_fp16)[name = string("transpose_388")];
            tensor<fp16, [1, 8, 188, 375]> x_95_cast_fp16 = matmul(transpose_x = x_95_transpose_x_0, transpose_y = x_95_transpose_y_0, x = q_with_bias_v_9_cast_fp16, y = op_1122_to_fp16_quantized)[name = string("x_95_cast_fp16")];
            tensor<int32, [8]> x_97_pad_0 = const()[name = string("x_97_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_97_mode_0 = const()[name = string("x_97_mode_0"), val = string("constant")];
            fp16 const_110_to_fp16 = const()[name = string("const_110_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_97_cast_fp16 = pad(constant_val = const_110_to_fp16, mode = x_97_mode_0, pad = x_97_pad_0, x = x_95_cast_fp16)[name = string("x_97_cast_fp16")];
            tensor<int32, [4]> var_1130 = const()[name = string("op_1130"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_99_cast_fp16 = reshape(shape = var_1130, x = x_97_cast_fp16)[name = string("x_99_cast_fp16")];
            tensor<int32, [4]> var_1134_begin_0 = const()[name = string("op_1134_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1134_end_0 = const()[name = string("op_1134_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_1134_end_mask_0 = const()[name = string("op_1134_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_1134_cast_fp16 = slice_by_index(begin = var_1134_begin_0, end = var_1134_end_0, end_mask = var_1134_end_mask_0, x = x_99_cast_fp16)[name = string("op_1134_cast_fp16")];
            tensor<int32, [4]> var_1135 = const()[name = string("op_1135"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_17_cast_fp16 = reshape(shape = var_1135, x = var_1134_cast_fp16)[name = string("matrix_bd_17_cast_fp16")];
            bool matrix_ac_9_transpose_x_0 = const()[name = string("matrix_ac_9_transpose_x_0"), val = bool(false)];
            bool matrix_ac_9_transpose_y_0 = const()[name = string("matrix_ac_9_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_136_perm_0 = const()[name = string("transpose_136_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_137_perm_0 = const()[name = string("transpose_137_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_137 = transpose(perm = transpose_137_perm_0, x = k_17_cast_fp16)[name = string("transpose_386")];
            tensor<fp16, [1, 8, 188, 128]> transpose_136 = transpose(perm = transpose_136_perm_0, x = var_1118_cast_fp16)[name = string("transpose_387")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_9_cast_fp16 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_136, y = transpose_137)[name = string("matrix_ac_9_cast_fp16")];
            tensor<int32, [4]> matrix_bd_19_begin_0 = const()[name = string("matrix_bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_19_end_0 = const()[name = string("matrix_bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_19_end_mask_0 = const()[name = string("matrix_bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_19_cast_fp16 = slice_by_index(begin = matrix_bd_19_begin_0, end = matrix_bd_19_end_0, end_mask = matrix_bd_19_end_mask_0, x = matrix_bd_17_cast_fp16)[name = string("matrix_bd_19_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1144_cast_fp16 = add(x = matrix_ac_9_cast_fp16, y = matrix_bd_19_cast_fp16)[name = string("op_1144_cast_fp16")];
            fp16 _inversed_scores_17_y_0_to_fp16 = const()[name = string("_inversed_scores_17_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_17_cast_fp16 = mul(x = var_1144_cast_fp16, y = _inversed_scores_17_y_0_to_fp16)[name = string("_inversed_scores_17_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_19_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_17_cast_fp16, cond = mask_11)[name = string("scores_19_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1150_cast_fp16 = softmax(axis = var_23, x = scores_19_cast_fp16)[name = string("op_1150_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_241_cast_fp16 = select(a = var_11_to_fp16, b = var_1150_cast_fp16, cond = mask_11)[name = string("input_241_cast_fp16")];
            bool x_101_transpose_x_0 = const()[name = string("x_101_transpose_x_0"), val = bool(false)];
            bool x_101_transpose_y_0 = const()[name = string("x_101_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = v_9_cast_fp16)[name = string("transpose_385")];
            tensor<fp16, [1, 8, 188, 128]> x_101_cast_fp16 = matmul(transpose_x = x_101_transpose_x_0, transpose_y = x_101_transpose_y_0, x = input_241_cast_fp16, y = value_11_cast_fp16)[name = string("x_101_cast_fp16")];
            tensor<int32, [4]> var_1154_perm_0 = const()[name = string("op_1154_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1155 = const()[name = string("op_1155"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_1154_cast_fp16 = transpose(perm = var_1154_perm_0, x = x_101_cast_fp16)[name = string("transpose_384")];
            tensor<fp16, [1, 188, 1024]> input_243_cast_fp16 = reshape(shape = var_1155, x = var_1154_cast_fp16)[name = string("input_243_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_4_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64538304))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65062656))))[name = string("module_layers_4_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_4_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_4_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65128256)))];
            tensor<fp16, [1, 188, 1024]> linear_43_cast_fp16 = linear(bias = module_layers_4_self_attn_linear_out_bias_to_fp16, weight = module_layers_4_self_attn_linear_out_weight_to_fp16_quantized, x = input_243_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_43_cast_fp16)[name = string("input_247_cast_fp16")];
            tensor<int32, [1]> x_105_axes_0 = const()[name = string("x_105_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_4_norm_conv_weight_to_fp16 = const()[name = string("module_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65130368)))];
            tensor<fp16, [1024]> module_layers_4_norm_conv_bias_to_fp16 = const()[name = string("module_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65132480)))];
            tensor<fp16, [1, 188, 1024]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = module_layers_4_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_conv_weight_to_fp16, x = input_247_cast_fp16)[name = string("x_105_cast_fp16")];
            tensor<int32, [3]> input_249_perm_0 = const()[name = string("input_249_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_251_pad_type_0 = const()[name = string("input_251_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_251_strides_0 = const()[name = string("input_251_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_251_pad_0 = const()[name = string("input_251_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_251_dilations_0 = const()[name = string("input_251_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_251_groups_0 = const()[name = string("input_251_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65134592))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66183232))))[name = string("module_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_4_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_4_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66314368)))];
            tensor<fp16, [1, 1024, 188]> input_249_cast_fp16 = transpose(perm = input_249_perm_0, x = x_105_cast_fp16)[name = string("transpose_383")];
            tensor<fp16, [1, 2048, 188]> input_251_cast_fp16 = conv(bias = module_layers_4_conv_pointwise_conv1_bias_to_fp16, dilations = input_251_dilations_0, groups = input_251_groups_0, pad = input_251_pad_0, pad_type = input_251_pad_type_0, strides = input_251_strides_0, weight = module_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_249_cast_fp16)[name = string("input_251_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, 188]> x_107_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_251_cast_fp16)[name = string("x_107_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_253_cast_fp16 = select(a = var_11_to_fp16, b = x_107_cast_fp16, cond = var_483)[name = string("input_253_cast_fp16")];
            tensor<int32, [6]> input_255_pad_0 = const()[name = string("input_255_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_255_mode_0 = const()[name = string("input_255_mode_0"), val = string("constant")];
            fp16 const_113_to_fp16 = const()[name = string("const_113_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_255_cast_fp16 = pad(constant_val = const_113_to_fp16, mode = input_255_mode_0, pad = input_255_pad_0, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
            string input_257_pad_type_0 = const()[name = string("input_257_pad_type_0"), val = string("valid")];
            int32 input_257_groups_0 = const()[name = string("input_257_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_257_strides_0 = const()[name = string("input_257_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_257_pad_0 = const()[name = string("input_257_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_257_dilations_0 = const()[name = string("input_257_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_392_to_fp16 = const()[name = string("const_392_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66318528)))];
            tensor<fp16, [1024]> const_393_to_fp16 = const()[name = string("const_393_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66337024)))];
            tensor<fp16, [1, 1024, 188]> input_259_cast_fp16 = conv(bias = const_393_to_fp16, dilations = input_257_dilations_0, groups = input_257_groups_0, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = input_257_strides_0, weight = const_392_to_fp16, x = input_255_cast_fp16)[name = string("input_259_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_261_cast_fp16 = silu(x = input_259_cast_fp16)[name = string("input_261_cast_fp16")];
            string x_109_pad_type_0 = const()[name = string("x_109_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_109_strides_0 = const()[name = string("x_109_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_109_pad_0 = const()[name = string("x_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_109_dilations_0 = const()[name = string("x_109_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_109_groups_0 = const()[name = string("x_109_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66339136))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66863488))))[name = string("module_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_4_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_4_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66929088)))];
            tensor<fp16, [1, 1024, 188]> x_109_cast_fp16 = conv(bias = module_layers_4_conv_pointwise_conv2_bias_to_fp16, dilations = x_109_dilations_0, groups = x_109_groups_0, pad = x_109_pad_0, pad_type = x_109_pad_type_0, strides = x_109_strides_0, weight = module_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_261_cast_fp16)[name = string("x_109_cast_fp16")];
            tensor<int32, [3]> input_263_perm_0 = const()[name = string("input_263_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_263_cast_fp16 = transpose(perm = input_263_perm_0, x = x_109_cast_fp16)[name = string("transpose_382")];
            tensor<fp16, [1, 188, 1024]> input_265_cast_fp16 = add(x = input_247_cast_fp16, y = input_263_cast_fp16)[name = string("input_265_cast_fp16")];
            tensor<int32, [1]> input_267_axes_0 = const()[name = string("input_267_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66931200)))];
            tensor<fp16, [1024]> module_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66933312)))];
            tensor<fp16, [1, 188, 1024]> input_267_cast_fp16 = layer_norm(axes = input_267_axes_0, beta = module_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_feed_forward2_weight_to_fp16, x = input_265_cast_fp16)[name = string("input_267_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_4_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66935424))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69032640))))[name = string("module_layers_4_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_4_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_4_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69294848)))];
            tensor<fp16, [1, 188, 4096]> linear_44_cast_fp16 = linear(bias = module_layers_4_feed_forward2_linear1_bias_to_fp16, weight = module_layers_4_feed_forward2_linear1_weight_to_fp16_quantized, x = input_267_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_271_cast_fp16 = silu(x = linear_44_cast_fp16)[name = string("input_271_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_4_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69303104))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71400320))))[name = string("module_layers_4_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_4_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_4_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71662528)))];
            tensor<fp16, [1, 188, 1024]> linear_45_cast_fp16 = linear(bias = module_layers_4_feed_forward2_linear2_bias_to_fp16, weight = module_layers_4_feed_forward2_linear2_weight_to_fp16_quantized, x = input_271_cast_fp16)[name = string("linear_45_cast_fp16")];
            fp16 var_1221_to_fp16 = const()[name = string("op_1221_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1222_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_1221_to_fp16)[name = string("op_1222_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_277_cast_fp16 = add(x = input_265_cast_fp16, y = var_1222_cast_fp16)[name = string("input_277_cast_fp16")];
            tensor<int32, [1]> input_279_axes_0 = const()[name = string("input_279_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_4_norm_out_weight_to_fp16 = const()[name = string("module_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71664640)))];
            tensor<fp16, [1024]> module_layers_4_norm_out_bias_to_fp16 = const()[name = string("module_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71666752)))];
            tensor<fp16, [1, 188, 1024]> input_279_cast_fp16 = layer_norm(axes = input_279_axes_0, beta = module_layers_4_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_4_norm_out_weight_to_fp16, x = input_277_cast_fp16)[name = string("input_279_cast_fp16")];
            tensor<int32, [1]> input_281_axes_0 = const()[name = string("input_281_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71668864)))];
            tensor<fp16, [1024]> module_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71670976)))];
            tensor<fp16, [1, 188, 1024]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = module_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward1_weight_to_fp16, x = input_279_cast_fp16)[name = string("input_281_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_5_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71673088))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73770304))))[name = string("module_layers_5_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_5_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_5_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74032512)))];
            tensor<fp16, [1, 188, 4096]> linear_46_cast_fp16 = linear(bias = module_layers_5_feed_forward1_linear1_bias_to_fp16, weight = module_layers_5_feed_forward1_linear1_weight_to_fp16_quantized, x = input_281_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_285_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("input_285_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_5_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(74040768))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76137984))))[name = string("module_layers_5_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_5_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_5_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76400192)))];
            tensor<fp16, [1, 188, 1024]> linear_47_cast_fp16 = linear(bias = module_layers_5_feed_forward1_linear2_bias_to_fp16, weight = module_layers_5_feed_forward1_linear2_weight_to_fp16_quantized, x = input_285_cast_fp16)[name = string("linear_47_cast_fp16")];
            fp16 var_1252_to_fp16 = const()[name = string("op_1252_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1253_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1252_to_fp16)[name = string("op_1253_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_291_cast_fp16 = add(x = input_279_cast_fp16, y = var_1253_cast_fp16)[name = string("input_291_cast_fp16")];
            tensor<int32, [1]> query_11_axes_0 = const()[name = string("query_11_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_5_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76402304)))];
            tensor<fp16, [1024]> module_layers_5_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76404416)))];
            tensor<fp16, [1, 188, 1024]> query_11_cast_fp16 = layer_norm(axes = query_11_axes_0, beta = module_layers_5_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_self_att_weight_to_fp16, x = input_291_cast_fp16)[name = string("query_11_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76406528))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76930880))))[name = string("module_layers_5_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_5_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_5_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76996480)))];
            tensor<fp16, [1, 188, 1024]> linear_48_cast_fp16 = linear(bias = module_layers_5_self_attn_linear_q_bias_to_fp16, weight = module_layers_5_self_attn_linear_q_weight_to_fp16_quantized, x = query_11_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<int32, [4]> var_1270 = const()[name = string("op_1270"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_31_cast_fp16 = reshape(shape = var_1270, x = linear_48_cast_fp16)[name = string("q_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(76998592))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77522944))))[name = string("module_layers_5_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_5_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_5_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77588544)))];
            tensor<fp16, [1, 188, 1024]> linear_49_cast_fp16 = linear(bias = module_layers_5_self_attn_linear_k_bias_to_fp16, weight = module_layers_5_self_attn_linear_k_weight_to_fp16_quantized, x = query_11_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<int32, [4]> var_1275 = const()[name = string("op_1275"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_21_cast_fp16 = reshape(shape = var_1275, x = linear_49_cast_fp16)[name = string("k_21_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77590656))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78115008))))[name = string("module_layers_5_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_5_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_5_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78180608)))];
            tensor<fp16, [1, 188, 1024]> linear_50_cast_fp16 = linear(bias = module_layers_5_self_attn_linear_v_bias_to_fp16, weight = module_layers_5_self_attn_linear_v_weight_to_fp16_quantized, x = query_11_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<int32, [4]> var_1280 = const()[name = string("op_1280"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_11_cast_fp16 = reshape(shape = var_1280, x = linear_50_cast_fp16)[name = string("v_11_cast_fp16")];
            tensor<int32, [4]> value_13_perm_0 = const()[name = string("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78182720))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78183296))))[name = string("module_layers_5_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1292_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_1292_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_5_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78183424))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78184000))))[name = string("module_layers_5_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1294_cast_fp16 = add(x = q_31_cast_fp16, y = module_layers_5_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_1294_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_11_perm_0 = const()[name = string("q_with_bias_v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_117_transpose_x_0 = const()[name = string("x_117_transpose_x_0"), val = bool(false)];
            bool x_117_transpose_y_0 = const()[name = string("x_117_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_1296_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78184128))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78376192))))[name = string("op_1296_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_11_cast_fp16 = transpose(perm = q_with_bias_v_11_perm_0, x = var_1294_cast_fp16)[name = string("transpose_381")];
            tensor<fp16, [1, 8, 188, 375]> x_117_cast_fp16 = matmul(transpose_x = x_117_transpose_x_0, transpose_y = x_117_transpose_y_0, x = q_with_bias_v_11_cast_fp16, y = op_1296_to_fp16_quantized)[name = string("x_117_cast_fp16")];
            tensor<int32, [8]> x_119_pad_0 = const()[name = string("x_119_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_119_mode_0 = const()[name = string("x_119_mode_0"), val = string("constant")];
            fp16 const_120_to_fp16 = const()[name = string("const_120_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_119_cast_fp16 = pad(constant_val = const_120_to_fp16, mode = x_119_mode_0, pad = x_119_pad_0, x = x_117_cast_fp16)[name = string("x_119_cast_fp16")];
            tensor<int32, [4]> var_1304 = const()[name = string("op_1304"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_121_cast_fp16 = reshape(shape = var_1304, x = x_119_cast_fp16)[name = string("x_121_cast_fp16")];
            tensor<int32, [4]> var_1308_begin_0 = const()[name = string("op_1308_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1308_end_0 = const()[name = string("op_1308_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_1308_end_mask_0 = const()[name = string("op_1308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_1308_cast_fp16 = slice_by_index(begin = var_1308_begin_0, end = var_1308_end_0, end_mask = var_1308_end_mask_0, x = x_121_cast_fp16)[name = string("op_1308_cast_fp16")];
            tensor<int32, [4]> var_1309 = const()[name = string("op_1309"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_21_cast_fp16 = reshape(shape = var_1309, x = var_1308_cast_fp16)[name = string("matrix_bd_21_cast_fp16")];
            bool matrix_ac_11_transpose_x_0 = const()[name = string("matrix_ac_11_transpose_x_0"), val = bool(false)];
            bool matrix_ac_11_transpose_y_0 = const()[name = string("matrix_ac_11_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_138_perm_0 = const()[name = string("transpose_138_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_139_perm_0 = const()[name = string("transpose_139_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_139 = transpose(perm = transpose_139_perm_0, x = k_21_cast_fp16)[name = string("transpose_379")];
            tensor<fp16, [1, 8, 188, 128]> transpose_138 = transpose(perm = transpose_138_perm_0, x = var_1292_cast_fp16)[name = string("transpose_380")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_11_cast_fp16 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_138, y = transpose_139)[name = string("matrix_ac_11_cast_fp16")];
            tensor<int32, [4]> matrix_bd_23_begin_0 = const()[name = string("matrix_bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_23_end_0 = const()[name = string("matrix_bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_23_end_mask_0 = const()[name = string("matrix_bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_23_cast_fp16 = slice_by_index(begin = matrix_bd_23_begin_0, end = matrix_bd_23_end_0, end_mask = matrix_bd_23_end_mask_0, x = matrix_bd_21_cast_fp16)[name = string("matrix_bd_23_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1318_cast_fp16 = add(x = matrix_ac_11_cast_fp16, y = matrix_bd_23_cast_fp16)[name = string("op_1318_cast_fp16")];
            fp16 _inversed_scores_21_y_0_to_fp16 = const()[name = string("_inversed_scores_21_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_21_cast_fp16 = mul(x = var_1318_cast_fp16, y = _inversed_scores_21_y_0_to_fp16)[name = string("_inversed_scores_21_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_23_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_21_cast_fp16, cond = mask_11)[name = string("scores_23_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1324_cast_fp16 = softmax(axis = var_23, x = scores_23_cast_fp16)[name = string("op_1324_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_293_cast_fp16 = select(a = var_11_to_fp16, b = var_1324_cast_fp16, cond = mask_11)[name = string("input_293_cast_fp16")];
            bool x_123_transpose_x_0 = const()[name = string("x_123_transpose_x_0"), val = bool(false)];
            bool x_123_transpose_y_0 = const()[name = string("x_123_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = v_11_cast_fp16)[name = string("transpose_378")];
            tensor<fp16, [1, 8, 188, 128]> x_123_cast_fp16 = matmul(transpose_x = x_123_transpose_x_0, transpose_y = x_123_transpose_y_0, x = input_293_cast_fp16, y = value_13_cast_fp16)[name = string("x_123_cast_fp16")];
            tensor<int32, [4]> var_1328_perm_0 = const()[name = string("op_1328_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1329 = const()[name = string("op_1329"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_1328_cast_fp16 = transpose(perm = var_1328_perm_0, x = x_123_cast_fp16)[name = string("transpose_377")];
            tensor<fp16, [1, 188, 1024]> input_295_cast_fp16 = reshape(shape = var_1329, x = var_1328_cast_fp16)[name = string("input_295_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_5_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78379264))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78903616))))[name = string("module_layers_5_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_5_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_5_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78969216)))];
            tensor<fp16, [1, 188, 1024]> linear_52_cast_fp16 = linear(bias = module_layers_5_self_attn_linear_out_bias_to_fp16, weight = module_layers_5_self_attn_linear_out_weight_to_fp16_quantized, x = input_295_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_52_cast_fp16)[name = string("input_299_cast_fp16")];
            tensor<int32, [1]> x_127_axes_0 = const()[name = string("x_127_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_5_norm_conv_weight_to_fp16 = const()[name = string("module_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78971328)))];
            tensor<fp16, [1024]> module_layers_5_norm_conv_bias_to_fp16 = const()[name = string("module_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78973440)))];
            tensor<fp16, [1, 188, 1024]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = module_layers_5_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_conv_weight_to_fp16, x = input_299_cast_fp16)[name = string("x_127_cast_fp16")];
            tensor<int32, [3]> input_301_perm_0 = const()[name = string("input_301_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_303_pad_type_0 = const()[name = string("input_303_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_303_strides_0 = const()[name = string("input_303_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_303_pad_0 = const()[name = string("input_303_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_303_dilations_0 = const()[name = string("input_303_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_303_groups_0 = const()[name = string("input_303_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78975552))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80024192))))[name = string("module_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_5_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_5_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80155328)))];
            tensor<fp16, [1, 1024, 188]> input_301_cast_fp16 = transpose(perm = input_301_perm_0, x = x_127_cast_fp16)[name = string("transpose_376")];
            tensor<fp16, [1, 2048, 188]> input_303_cast_fp16 = conv(bias = module_layers_5_conv_pointwise_conv1_bias_to_fp16, dilations = input_303_dilations_0, groups = input_303_groups_0, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = input_303_strides_0, weight = module_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_301_cast_fp16)[name = string("input_303_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, 188]> x_129_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_303_cast_fp16)[name = string("x_129_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_305_cast_fp16 = select(a = var_11_to_fp16, b = x_129_cast_fp16, cond = var_483)[name = string("input_305_cast_fp16")];
            tensor<int32, [6]> input_307_pad_0 = const()[name = string("input_307_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_307_mode_0 = const()[name = string("input_307_mode_0"), val = string("constant")];
            fp16 const_123_to_fp16 = const()[name = string("const_123_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_307_cast_fp16 = pad(constant_val = const_123_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = string("input_307_cast_fp16")];
            string input_309_pad_type_0 = const()[name = string("input_309_pad_type_0"), val = string("valid")];
            int32 input_309_groups_0 = const()[name = string("input_309_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_309_strides_0 = const()[name = string("input_309_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_309_pad_0 = const()[name = string("input_309_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_309_dilations_0 = const()[name = string("input_309_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_394_to_fp16 = const()[name = string("const_394_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80159488)))];
            tensor<fp16, [1024]> const_395_to_fp16 = const()[name = string("const_395_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80177984)))];
            tensor<fp16, [1, 1024, 188]> input_311_cast_fp16 = conv(bias = const_395_to_fp16, dilations = input_309_dilations_0, groups = input_309_groups_0, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = input_309_strides_0, weight = const_394_to_fp16, x = input_307_cast_fp16)[name = string("input_311_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_313_cast_fp16 = silu(x = input_311_cast_fp16)[name = string("input_313_cast_fp16")];
            string x_131_pad_type_0 = const()[name = string("x_131_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_131_strides_0 = const()[name = string("x_131_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_131_pad_0 = const()[name = string("x_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_131_dilations_0 = const()[name = string("x_131_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_131_groups_0 = const()[name = string("x_131_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80180096))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80704448))))[name = string("module_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_5_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_5_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80770048)))];
            tensor<fp16, [1, 1024, 188]> x_131_cast_fp16 = conv(bias = module_layers_5_conv_pointwise_conv2_bias_to_fp16, dilations = x_131_dilations_0, groups = x_131_groups_0, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = x_131_strides_0, weight = module_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_313_cast_fp16)[name = string("x_131_cast_fp16")];
            tensor<int32, [3]> input_315_perm_0 = const()[name = string("input_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_315_cast_fp16 = transpose(perm = input_315_perm_0, x = x_131_cast_fp16)[name = string("transpose_375")];
            tensor<fp16, [1, 188, 1024]> input_317_cast_fp16 = add(x = input_299_cast_fp16, y = input_315_cast_fp16)[name = string("input_317_cast_fp16")];
            tensor<int32, [1]> input_319_axes_0 = const()[name = string("input_319_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80772160)))];
            tensor<fp16, [1024]> module_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80774272)))];
            tensor<fp16, [1, 188, 1024]> input_319_cast_fp16 = layer_norm(axes = input_319_axes_0, beta = module_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_feed_forward2_weight_to_fp16, x = input_317_cast_fp16)[name = string("input_319_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_5_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80776384))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(82873600))))[name = string("module_layers_5_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_5_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_5_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83135808)))];
            tensor<fp16, [1, 188, 4096]> linear_53_cast_fp16 = linear(bias = module_layers_5_feed_forward2_linear1_bias_to_fp16, weight = module_layers_5_feed_forward2_linear1_weight_to_fp16_quantized, x = input_319_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_323_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("input_323_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_5_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83144064))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85241280))))[name = string("module_layers_5_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_5_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_5_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85503488)))];
            tensor<fp16, [1, 188, 1024]> linear_54_cast_fp16 = linear(bias = module_layers_5_feed_forward2_linear2_bias_to_fp16, weight = module_layers_5_feed_forward2_linear2_weight_to_fp16_quantized, x = input_323_cast_fp16)[name = string("linear_54_cast_fp16")];
            fp16 var_1395_to_fp16 = const()[name = string("op_1395_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1396_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_1395_to_fp16)[name = string("op_1396_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_329_cast_fp16 = add(x = input_317_cast_fp16, y = var_1396_cast_fp16)[name = string("input_329_cast_fp16")];
            tensor<int32, [1]> input_331_axes_0 = const()[name = string("input_331_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_5_norm_out_weight_to_fp16 = const()[name = string("module_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85505600)))];
            tensor<fp16, [1024]> module_layers_5_norm_out_bias_to_fp16 = const()[name = string("module_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85507712)))];
            tensor<fp16, [1, 188, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = module_layers_5_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_5_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = string("input_331_cast_fp16")];
            tensor<int32, [1]> input_333_axes_0 = const()[name = string("input_333_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85509824)))];
            tensor<fp16, [1024]> module_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85511936)))];
            tensor<fp16, [1, 188, 1024]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = module_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = string("input_333_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_6_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85514048))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87611264))))[name = string("module_layers_6_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_6_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_6_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87873472)))];
            tensor<fp16, [1, 188, 4096]> linear_55_cast_fp16 = linear(bias = module_layers_6_feed_forward1_linear1_bias_to_fp16, weight = module_layers_6_feed_forward1_linear1_weight_to_fp16_quantized, x = input_333_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_337_cast_fp16 = silu(x = linear_55_cast_fp16)[name = string("input_337_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_6_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87881728))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89978944))))[name = string("module_layers_6_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_6_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_6_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90241152)))];
            tensor<fp16, [1, 188, 1024]> linear_56_cast_fp16 = linear(bias = module_layers_6_feed_forward1_linear2_bias_to_fp16, weight = module_layers_6_feed_forward1_linear2_weight_to_fp16_quantized, x = input_337_cast_fp16)[name = string("linear_56_cast_fp16")];
            fp16 var_1426_to_fp16 = const()[name = string("op_1426_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1427_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_1426_to_fp16)[name = string("op_1427_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_343_cast_fp16 = add(x = input_331_cast_fp16, y = var_1427_cast_fp16)[name = string("input_343_cast_fp16")];
            tensor<int32, [1]> query_13_axes_0 = const()[name = string("query_13_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_6_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90243264)))];
            tensor<fp16, [1024]> module_layers_6_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90245376)))];
            tensor<fp16, [1, 188, 1024]> query_13_cast_fp16 = layer_norm(axes = query_13_axes_0, beta = module_layers_6_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_self_att_weight_to_fp16, x = input_343_cast_fp16)[name = string("query_13_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90247488))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90771840))))[name = string("module_layers_6_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_6_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_6_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90837440)))];
            tensor<fp16, [1, 188, 1024]> linear_57_cast_fp16 = linear(bias = module_layers_6_self_attn_linear_q_bias_to_fp16, weight = module_layers_6_self_attn_linear_q_weight_to_fp16_quantized, x = query_13_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<int32, [4]> var_1444 = const()[name = string("op_1444"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_37_cast_fp16 = reshape(shape = var_1444, x = linear_57_cast_fp16)[name = string("q_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(90839552))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91363904))))[name = string("module_layers_6_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_6_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_6_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91429504)))];
            tensor<fp16, [1, 188, 1024]> linear_58_cast_fp16 = linear(bias = module_layers_6_self_attn_linear_k_bias_to_fp16, weight = module_layers_6_self_attn_linear_k_weight_to_fp16_quantized, x = query_13_cast_fp16)[name = string("linear_58_cast_fp16")];
            tensor<int32, [4]> var_1449 = const()[name = string("op_1449"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_25_cast_fp16 = reshape(shape = var_1449, x = linear_58_cast_fp16)[name = string("k_25_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91431616))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91955968))))[name = string("module_layers_6_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_6_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_6_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92021568)))];
            tensor<fp16, [1, 188, 1024]> linear_59_cast_fp16 = linear(bias = module_layers_6_self_attn_linear_v_bias_to_fp16, weight = module_layers_6_self_attn_linear_v_weight_to_fp16_quantized, x = query_13_cast_fp16)[name = string("linear_59_cast_fp16")];
            tensor<int32, [4]> var_1454 = const()[name = string("op_1454"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_13_cast_fp16 = reshape(shape = var_1454, x = linear_59_cast_fp16)[name = string("v_13_cast_fp16")];
            tensor<int32, [4]> value_15_perm_0 = const()[name = string("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92023680))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92024256))))[name = string("module_layers_6_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1466_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_1466_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_6_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92024384))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92024960))))[name = string("module_layers_6_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1468_cast_fp16 = add(x = q_37_cast_fp16, y = module_layers_6_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_1468_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_13_perm_0 = const()[name = string("q_with_bias_v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_139_transpose_x_0 = const()[name = string("x_139_transpose_x_0"), val = bool(false)];
            bool x_139_transpose_y_0 = const()[name = string("x_139_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_1470_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92025088))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92217152))))[name = string("op_1470_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_13_cast_fp16 = transpose(perm = q_with_bias_v_13_perm_0, x = var_1468_cast_fp16)[name = string("transpose_374")];
            tensor<fp16, [1, 8, 188, 375]> x_139_cast_fp16 = matmul(transpose_x = x_139_transpose_x_0, transpose_y = x_139_transpose_y_0, x = q_with_bias_v_13_cast_fp16, y = op_1470_to_fp16_quantized)[name = string("x_139_cast_fp16")];
            tensor<int32, [8]> x_141_pad_0 = const()[name = string("x_141_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_141_mode_0 = const()[name = string("x_141_mode_0"), val = string("constant")];
            fp16 const_130_to_fp16 = const()[name = string("const_130_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_141_cast_fp16 = pad(constant_val = const_130_to_fp16, mode = x_141_mode_0, pad = x_141_pad_0, x = x_139_cast_fp16)[name = string("x_141_cast_fp16")];
            tensor<int32, [4]> var_1478 = const()[name = string("op_1478"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_143_cast_fp16 = reshape(shape = var_1478, x = x_141_cast_fp16)[name = string("x_143_cast_fp16")];
            tensor<int32, [4]> var_1482_begin_0 = const()[name = string("op_1482_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1482_end_0 = const()[name = string("op_1482_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_1482_end_mask_0 = const()[name = string("op_1482_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_1482_cast_fp16 = slice_by_index(begin = var_1482_begin_0, end = var_1482_end_0, end_mask = var_1482_end_mask_0, x = x_143_cast_fp16)[name = string("op_1482_cast_fp16")];
            tensor<int32, [4]> var_1483 = const()[name = string("op_1483"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_25_cast_fp16 = reshape(shape = var_1483, x = var_1482_cast_fp16)[name = string("matrix_bd_25_cast_fp16")];
            bool matrix_ac_13_transpose_x_0 = const()[name = string("matrix_ac_13_transpose_x_0"), val = bool(false)];
            bool matrix_ac_13_transpose_y_0 = const()[name = string("matrix_ac_13_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_140_perm_0 = const()[name = string("transpose_140_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_141_perm_0 = const()[name = string("transpose_141_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_141 = transpose(perm = transpose_141_perm_0, x = k_25_cast_fp16)[name = string("transpose_372")];
            tensor<fp16, [1, 8, 188, 128]> transpose_140 = transpose(perm = transpose_140_perm_0, x = var_1466_cast_fp16)[name = string("transpose_373")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_13_cast_fp16 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_140, y = transpose_141)[name = string("matrix_ac_13_cast_fp16")];
            tensor<int32, [4]> matrix_bd_27_begin_0 = const()[name = string("matrix_bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_27_end_0 = const()[name = string("matrix_bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_27_end_mask_0 = const()[name = string("matrix_bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_27_cast_fp16 = slice_by_index(begin = matrix_bd_27_begin_0, end = matrix_bd_27_end_0, end_mask = matrix_bd_27_end_mask_0, x = matrix_bd_25_cast_fp16)[name = string("matrix_bd_27_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1492_cast_fp16 = add(x = matrix_ac_13_cast_fp16, y = matrix_bd_27_cast_fp16)[name = string("op_1492_cast_fp16")];
            fp16 _inversed_scores_25_y_0_to_fp16 = const()[name = string("_inversed_scores_25_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_25_cast_fp16 = mul(x = var_1492_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = string("_inversed_scores_25_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_27_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_25_cast_fp16, cond = mask_11)[name = string("scores_27_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1498_cast_fp16 = softmax(axis = var_23, x = scores_27_cast_fp16)[name = string("op_1498_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_345_cast_fp16 = select(a = var_11_to_fp16, b = var_1498_cast_fp16, cond = mask_11)[name = string("input_345_cast_fp16")];
            bool x_145_transpose_x_0 = const()[name = string("x_145_transpose_x_0"), val = bool(false)];
            bool x_145_transpose_y_0 = const()[name = string("x_145_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = v_13_cast_fp16)[name = string("transpose_371")];
            tensor<fp16, [1, 8, 188, 128]> x_145_cast_fp16 = matmul(transpose_x = x_145_transpose_x_0, transpose_y = x_145_transpose_y_0, x = input_345_cast_fp16, y = value_15_cast_fp16)[name = string("x_145_cast_fp16")];
            tensor<int32, [4]> var_1502_perm_0 = const()[name = string("op_1502_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1503 = const()[name = string("op_1503"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_1502_cast_fp16 = transpose(perm = var_1502_perm_0, x = x_145_cast_fp16)[name = string("transpose_370")];
            tensor<fp16, [1, 188, 1024]> input_347_cast_fp16 = reshape(shape = var_1503, x = var_1502_cast_fp16)[name = string("input_347_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_6_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92220224))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92744576))))[name = string("module_layers_6_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_6_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_6_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92810176)))];
            tensor<fp16, [1, 188, 1024]> linear_61_cast_fp16 = linear(bias = module_layers_6_self_attn_linear_out_bias_to_fp16, weight = module_layers_6_self_attn_linear_out_weight_to_fp16_quantized, x = input_347_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_61_cast_fp16)[name = string("input_351_cast_fp16")];
            tensor<int32, [1]> x_149_axes_0 = const()[name = string("x_149_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_6_norm_conv_weight_to_fp16 = const()[name = string("module_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92812288)))];
            tensor<fp16, [1024]> module_layers_6_norm_conv_bias_to_fp16 = const()[name = string("module_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92814400)))];
            tensor<fp16, [1, 188, 1024]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = module_layers_6_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_conv_weight_to_fp16, x = input_351_cast_fp16)[name = string("x_149_cast_fp16")];
            tensor<int32, [3]> input_353_perm_0 = const()[name = string("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_355_pad_type_0 = const()[name = string("input_355_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_355_strides_0 = const()[name = string("input_355_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_355_pad_0 = const()[name = string("input_355_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_355_dilations_0 = const()[name = string("input_355_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_355_groups_0 = const()[name = string("input_355_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92816512))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93865152))))[name = string("module_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_6_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_6_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93996288)))];
            tensor<fp16, [1, 1024, 188]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = x_149_cast_fp16)[name = string("transpose_369")];
            tensor<fp16, [1, 2048, 188]> input_355_cast_fp16 = conv(bias = module_layers_6_conv_pointwise_conv1_bias_to_fp16, dilations = input_355_dilations_0, groups = input_355_groups_0, pad = input_355_pad_0, pad_type = input_355_pad_type_0, strides = input_355_strides_0, weight = module_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_353_cast_fp16)[name = string("input_355_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, 188]> x_151_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_355_cast_fp16)[name = string("x_151_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_357_cast_fp16 = select(a = var_11_to_fp16, b = x_151_cast_fp16, cond = var_483)[name = string("input_357_cast_fp16")];
            tensor<int32, [6]> input_359_pad_0 = const()[name = string("input_359_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_359_mode_0 = const()[name = string("input_359_mode_0"), val = string("constant")];
            fp16 const_133_to_fp16 = const()[name = string("const_133_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_359_cast_fp16 = pad(constant_val = const_133_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = string("input_359_cast_fp16")];
            string input_361_pad_type_0 = const()[name = string("input_361_pad_type_0"), val = string("valid")];
            int32 input_361_groups_0 = const()[name = string("input_361_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_361_strides_0 = const()[name = string("input_361_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_361_pad_0 = const()[name = string("input_361_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_361_dilations_0 = const()[name = string("input_361_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_396_to_fp16 = const()[name = string("const_396_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94000448)))];
            tensor<fp16, [1024]> const_397_to_fp16 = const()[name = string("const_397_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94018944)))];
            tensor<fp16, [1, 1024, 188]> input_363_cast_fp16 = conv(bias = const_397_to_fp16, dilations = input_361_dilations_0, groups = input_361_groups_0, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = input_361_strides_0, weight = const_396_to_fp16, x = input_359_cast_fp16)[name = string("input_363_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_365_cast_fp16 = silu(x = input_363_cast_fp16)[name = string("input_365_cast_fp16")];
            string x_153_pad_type_0 = const()[name = string("x_153_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_153_strides_0 = const()[name = string("x_153_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_153_pad_0 = const()[name = string("x_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_153_dilations_0 = const()[name = string("x_153_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_153_groups_0 = const()[name = string("x_153_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94021056))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94545408))))[name = string("module_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_6_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_6_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94611008)))];
            tensor<fp16, [1, 1024, 188]> x_153_cast_fp16 = conv(bias = module_layers_6_conv_pointwise_conv2_bias_to_fp16, dilations = x_153_dilations_0, groups = x_153_groups_0, pad = x_153_pad_0, pad_type = x_153_pad_type_0, strides = x_153_strides_0, weight = module_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_365_cast_fp16)[name = string("x_153_cast_fp16")];
            tensor<int32, [3]> input_367_perm_0 = const()[name = string("input_367_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_367_cast_fp16 = transpose(perm = input_367_perm_0, x = x_153_cast_fp16)[name = string("transpose_368")];
            tensor<fp16, [1, 188, 1024]> input_369_cast_fp16 = add(x = input_351_cast_fp16, y = input_367_cast_fp16)[name = string("input_369_cast_fp16")];
            tensor<int32, [1]> input_371_axes_0 = const()[name = string("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94613120)))];
            tensor<fp16, [1024]> module_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94615232)))];
            tensor<fp16, [1, 188, 1024]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = module_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = string("input_371_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_6_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94617344))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96714560))))[name = string("module_layers_6_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_6_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_6_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96976768)))];
            tensor<fp16, [1, 188, 4096]> linear_62_cast_fp16 = linear(bias = module_layers_6_feed_forward2_linear1_bias_to_fp16, weight = module_layers_6_feed_forward2_linear1_weight_to_fp16_quantized, x = input_371_cast_fp16)[name = string("linear_62_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_375_cast_fp16 = silu(x = linear_62_cast_fp16)[name = string("input_375_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_6_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(96985024))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99082240))))[name = string("module_layers_6_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_6_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_6_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99344448)))];
            tensor<fp16, [1, 188, 1024]> linear_63_cast_fp16 = linear(bias = module_layers_6_feed_forward2_linear2_bias_to_fp16, weight = module_layers_6_feed_forward2_linear2_weight_to_fp16_quantized, x = input_375_cast_fp16)[name = string("linear_63_cast_fp16")];
            fp16 var_1569_to_fp16 = const()[name = string("op_1569_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1570_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_1569_to_fp16)[name = string("op_1570_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_381_cast_fp16 = add(x = input_369_cast_fp16, y = var_1570_cast_fp16)[name = string("input_381_cast_fp16")];
            tensor<int32, [1]> input_383_axes_0 = const()[name = string("input_383_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_6_norm_out_weight_to_fp16 = const()[name = string("module_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99346560)))];
            tensor<fp16, [1024]> module_layers_6_norm_out_bias_to_fp16 = const()[name = string("module_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99348672)))];
            tensor<fp16, [1, 188, 1024]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = module_layers_6_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_6_norm_out_weight_to_fp16, x = input_381_cast_fp16)[name = string("input_383_cast_fp16")];
            tensor<int32, [1]> input_385_axes_0 = const()[name = string("input_385_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99350784)))];
            tensor<fp16, [1024]> module_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99352896)))];
            tensor<fp16, [1, 188, 1024]> input_385_cast_fp16 = layer_norm(axes = input_385_axes_0, beta = module_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward1_weight_to_fp16, x = input_383_cast_fp16)[name = string("input_385_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_7_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99355008))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101452224))))[name = string("module_layers_7_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_7_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_7_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101714432)))];
            tensor<fp16, [1, 188, 4096]> linear_64_cast_fp16 = linear(bias = module_layers_7_feed_forward1_linear1_bias_to_fp16, weight = module_layers_7_feed_forward1_linear1_weight_to_fp16_quantized, x = input_385_cast_fp16)[name = string("linear_64_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_389_cast_fp16 = silu(x = linear_64_cast_fp16)[name = string("input_389_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_7_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(101722688))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(103819904))))[name = string("module_layers_7_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_7_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_7_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104082112)))];
            tensor<fp16, [1, 188, 1024]> linear_65_cast_fp16 = linear(bias = module_layers_7_feed_forward1_linear2_bias_to_fp16, weight = module_layers_7_feed_forward1_linear2_weight_to_fp16_quantized, x = input_389_cast_fp16)[name = string("linear_65_cast_fp16")];
            fp16 var_1600_to_fp16 = const()[name = string("op_1600_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1601_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_1600_to_fp16)[name = string("op_1601_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_395_cast_fp16 = add(x = input_383_cast_fp16, y = var_1601_cast_fp16)[name = string("input_395_cast_fp16")];
            tensor<int32, [1]> query_15_axes_0 = const()[name = string("query_15_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_7_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104084224)))];
            tensor<fp16, [1024]> module_layers_7_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104086336)))];
            tensor<fp16, [1, 188, 1024]> query_15_cast_fp16 = layer_norm(axes = query_15_axes_0, beta = module_layers_7_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_self_att_weight_to_fp16, x = input_395_cast_fp16)[name = string("query_15_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104088448))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104612800))))[name = string("module_layers_7_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_7_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_7_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104678400)))];
            tensor<fp16, [1, 188, 1024]> linear_66_cast_fp16 = linear(bias = module_layers_7_self_attn_linear_q_bias_to_fp16, weight = module_layers_7_self_attn_linear_q_weight_to_fp16_quantized, x = query_15_cast_fp16)[name = string("linear_66_cast_fp16")];
            tensor<int32, [4]> var_1618 = const()[name = string("op_1618"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_43_cast_fp16 = reshape(shape = var_1618, x = linear_66_cast_fp16)[name = string("q_43_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104680512))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105204864))))[name = string("module_layers_7_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_7_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_7_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105270464)))];
            tensor<fp16, [1, 188, 1024]> linear_67_cast_fp16 = linear(bias = module_layers_7_self_attn_linear_k_bias_to_fp16, weight = module_layers_7_self_attn_linear_k_weight_to_fp16_quantized, x = query_15_cast_fp16)[name = string("linear_67_cast_fp16")];
            tensor<int32, [4]> var_1623 = const()[name = string("op_1623"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_29_cast_fp16 = reshape(shape = var_1623, x = linear_67_cast_fp16)[name = string("k_29_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105272576))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105796928))))[name = string("module_layers_7_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_7_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_7_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105862528)))];
            tensor<fp16, [1, 188, 1024]> linear_68_cast_fp16 = linear(bias = module_layers_7_self_attn_linear_v_bias_to_fp16, weight = module_layers_7_self_attn_linear_v_weight_to_fp16_quantized, x = query_15_cast_fp16)[name = string("linear_68_cast_fp16")];
            tensor<int32, [4]> var_1628 = const()[name = string("op_1628"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_15_cast_fp16 = reshape(shape = var_1628, x = linear_68_cast_fp16)[name = string("v_15_cast_fp16")];
            tensor<int32, [4]> value_17_perm_0 = const()[name = string("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105864640))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105865216))))[name = string("module_layers_7_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1640_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_1640_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_7_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105865344))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105865920))))[name = string("module_layers_7_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1642_cast_fp16 = add(x = q_43_cast_fp16, y = module_layers_7_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_1642_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_15_perm_0 = const()[name = string("q_with_bias_v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_161_transpose_x_0 = const()[name = string("x_161_transpose_x_0"), val = bool(false)];
            bool x_161_transpose_y_0 = const()[name = string("x_161_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_1644_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105866048))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106058112))))[name = string("op_1644_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_15_cast_fp16 = transpose(perm = q_with_bias_v_15_perm_0, x = var_1642_cast_fp16)[name = string("transpose_367")];
            tensor<fp16, [1, 8, 188, 375]> x_161_cast_fp16 = matmul(transpose_x = x_161_transpose_x_0, transpose_y = x_161_transpose_y_0, x = q_with_bias_v_15_cast_fp16, y = op_1644_to_fp16_quantized)[name = string("x_161_cast_fp16")];
            tensor<int32, [8]> x_163_pad_0 = const()[name = string("x_163_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_163_mode_0 = const()[name = string("x_163_mode_0"), val = string("constant")];
            fp16 const_140_to_fp16 = const()[name = string("const_140_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_163_cast_fp16 = pad(constant_val = const_140_to_fp16, mode = x_163_mode_0, pad = x_163_pad_0, x = x_161_cast_fp16)[name = string("x_163_cast_fp16")];
            tensor<int32, [4]> var_1652 = const()[name = string("op_1652"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_165_cast_fp16 = reshape(shape = var_1652, x = x_163_cast_fp16)[name = string("x_165_cast_fp16")];
            tensor<int32, [4]> var_1656_begin_0 = const()[name = string("op_1656_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1656_end_0 = const()[name = string("op_1656_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_1656_end_mask_0 = const()[name = string("op_1656_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_1656_cast_fp16 = slice_by_index(begin = var_1656_begin_0, end = var_1656_end_0, end_mask = var_1656_end_mask_0, x = x_165_cast_fp16)[name = string("op_1656_cast_fp16")];
            tensor<int32, [4]> var_1657 = const()[name = string("op_1657"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_29_cast_fp16 = reshape(shape = var_1657, x = var_1656_cast_fp16)[name = string("matrix_bd_29_cast_fp16")];
            bool matrix_ac_15_transpose_x_0 = const()[name = string("matrix_ac_15_transpose_x_0"), val = bool(false)];
            bool matrix_ac_15_transpose_y_0 = const()[name = string("matrix_ac_15_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_142_perm_0 = const()[name = string("transpose_142_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_143_perm_0 = const()[name = string("transpose_143_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_143 = transpose(perm = transpose_143_perm_0, x = k_29_cast_fp16)[name = string("transpose_365")];
            tensor<fp16, [1, 8, 188, 128]> transpose_142 = transpose(perm = transpose_142_perm_0, x = var_1640_cast_fp16)[name = string("transpose_366")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_15_cast_fp16 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_142, y = transpose_143)[name = string("matrix_ac_15_cast_fp16")];
            tensor<int32, [4]> matrix_bd_31_begin_0 = const()[name = string("matrix_bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_31_end_0 = const()[name = string("matrix_bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_31_end_mask_0 = const()[name = string("matrix_bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_31_cast_fp16 = slice_by_index(begin = matrix_bd_31_begin_0, end = matrix_bd_31_end_0, end_mask = matrix_bd_31_end_mask_0, x = matrix_bd_29_cast_fp16)[name = string("matrix_bd_31_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1666_cast_fp16 = add(x = matrix_ac_15_cast_fp16, y = matrix_bd_31_cast_fp16)[name = string("op_1666_cast_fp16")];
            fp16 _inversed_scores_29_y_0_to_fp16 = const()[name = string("_inversed_scores_29_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_29_cast_fp16 = mul(x = var_1666_cast_fp16, y = _inversed_scores_29_y_0_to_fp16)[name = string("_inversed_scores_29_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_31_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_29_cast_fp16, cond = mask_11)[name = string("scores_31_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1672_cast_fp16 = softmax(axis = var_23, x = scores_31_cast_fp16)[name = string("op_1672_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_397_cast_fp16 = select(a = var_11_to_fp16, b = var_1672_cast_fp16, cond = mask_11)[name = string("input_397_cast_fp16")];
            bool x_167_transpose_x_0 = const()[name = string("x_167_transpose_x_0"), val = bool(false)];
            bool x_167_transpose_y_0 = const()[name = string("x_167_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = v_15_cast_fp16)[name = string("transpose_364")];
            tensor<fp16, [1, 8, 188, 128]> x_167_cast_fp16 = matmul(transpose_x = x_167_transpose_x_0, transpose_y = x_167_transpose_y_0, x = input_397_cast_fp16, y = value_17_cast_fp16)[name = string("x_167_cast_fp16")];
            tensor<int32, [4]> var_1676_perm_0 = const()[name = string("op_1676_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1677 = const()[name = string("op_1677"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_1676_cast_fp16 = transpose(perm = var_1676_perm_0, x = x_167_cast_fp16)[name = string("transpose_363")];
            tensor<fp16, [1, 188, 1024]> input_399_cast_fp16 = reshape(shape = var_1677, x = var_1676_cast_fp16)[name = string("input_399_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_7_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106061184))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106585536))))[name = string("module_layers_7_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_7_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_7_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106651136)))];
            tensor<fp16, [1, 188, 1024]> linear_70_cast_fp16 = linear(bias = module_layers_7_self_attn_linear_out_bias_to_fp16, weight = module_layers_7_self_attn_linear_out_weight_to_fp16_quantized, x = input_399_cast_fp16)[name = string("linear_70_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_403_cast_fp16 = add(x = input_395_cast_fp16, y = linear_70_cast_fp16)[name = string("input_403_cast_fp16")];
            tensor<int32, [1]> x_171_axes_0 = const()[name = string("x_171_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_7_norm_conv_weight_to_fp16 = const()[name = string("module_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106653248)))];
            tensor<fp16, [1024]> module_layers_7_norm_conv_bias_to_fp16 = const()[name = string("module_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106655360)))];
            tensor<fp16, [1, 188, 1024]> x_171_cast_fp16 = layer_norm(axes = x_171_axes_0, beta = module_layers_7_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_conv_weight_to_fp16, x = input_403_cast_fp16)[name = string("x_171_cast_fp16")];
            tensor<int32, [3]> input_405_perm_0 = const()[name = string("input_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_407_pad_type_0 = const()[name = string("input_407_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_407_strides_0 = const()[name = string("input_407_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_407_pad_0 = const()[name = string("input_407_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_407_dilations_0 = const()[name = string("input_407_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_407_groups_0 = const()[name = string("input_407_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106657472))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107706112))))[name = string("module_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_7_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_7_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107837248)))];
            tensor<fp16, [1, 1024, 188]> input_405_cast_fp16 = transpose(perm = input_405_perm_0, x = x_171_cast_fp16)[name = string("transpose_362")];
            tensor<fp16, [1, 2048, 188]> input_407_cast_fp16 = conv(bias = module_layers_7_conv_pointwise_conv1_bias_to_fp16, dilations = input_407_dilations_0, groups = input_407_groups_0, pad = input_407_pad_0, pad_type = input_407_pad_type_0, strides = input_407_strides_0, weight = module_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_405_cast_fp16)[name = string("input_407_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, 188]> x_173_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_407_cast_fp16)[name = string("x_173_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_409_cast_fp16 = select(a = var_11_to_fp16, b = x_173_cast_fp16, cond = var_483)[name = string("input_409_cast_fp16")];
            tensor<int32, [6]> input_411_pad_0 = const()[name = string("input_411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_411_mode_0 = const()[name = string("input_411_mode_0"), val = string("constant")];
            fp16 const_143_to_fp16 = const()[name = string("const_143_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_411_cast_fp16 = pad(constant_val = const_143_to_fp16, mode = input_411_mode_0, pad = input_411_pad_0, x = input_409_cast_fp16)[name = string("input_411_cast_fp16")];
            string input_413_pad_type_0 = const()[name = string("input_413_pad_type_0"), val = string("valid")];
            int32 input_413_groups_0 = const()[name = string("input_413_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_413_strides_0 = const()[name = string("input_413_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_413_pad_0 = const()[name = string("input_413_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_413_dilations_0 = const()[name = string("input_413_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_398_to_fp16 = const()[name = string("const_398_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107841408)))];
            tensor<fp16, [1024]> const_399_to_fp16 = const()[name = string("const_399_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107859904)))];
            tensor<fp16, [1, 1024, 188]> input_415_cast_fp16 = conv(bias = const_399_to_fp16, dilations = input_413_dilations_0, groups = input_413_groups_0, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = input_413_strides_0, weight = const_398_to_fp16, x = input_411_cast_fp16)[name = string("input_415_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_417_cast_fp16 = silu(x = input_415_cast_fp16)[name = string("input_417_cast_fp16")];
            string x_175_pad_type_0 = const()[name = string("x_175_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_175_strides_0 = const()[name = string("x_175_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_175_pad_0 = const()[name = string("x_175_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_175_dilations_0 = const()[name = string("x_175_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_175_groups_0 = const()[name = string("x_175_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107862016))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108386368))))[name = string("module_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_7_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_7_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108451968)))];
            tensor<fp16, [1, 1024, 188]> x_175_cast_fp16 = conv(bias = module_layers_7_conv_pointwise_conv2_bias_to_fp16, dilations = x_175_dilations_0, groups = x_175_groups_0, pad = x_175_pad_0, pad_type = x_175_pad_type_0, strides = x_175_strides_0, weight = module_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_417_cast_fp16)[name = string("x_175_cast_fp16")];
            tensor<int32, [3]> input_419_perm_0 = const()[name = string("input_419_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_419_cast_fp16 = transpose(perm = input_419_perm_0, x = x_175_cast_fp16)[name = string("transpose_361")];
            tensor<fp16, [1, 188, 1024]> input_421_cast_fp16 = add(x = input_403_cast_fp16, y = input_419_cast_fp16)[name = string("input_421_cast_fp16")];
            tensor<int32, [1]> input_423_axes_0 = const()[name = string("input_423_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108454080)))];
            tensor<fp16, [1024]> module_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108456192)))];
            tensor<fp16, [1, 188, 1024]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = module_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = string("input_423_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_7_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108458304))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110555520))))[name = string("module_layers_7_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_7_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_7_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110817728)))];
            tensor<fp16, [1, 188, 4096]> linear_71_cast_fp16 = linear(bias = module_layers_7_feed_forward2_linear1_bias_to_fp16, weight = module_layers_7_feed_forward2_linear1_weight_to_fp16_quantized, x = input_423_cast_fp16)[name = string("linear_71_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_427_cast_fp16 = silu(x = linear_71_cast_fp16)[name = string("input_427_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_7_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110825984))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112923200))))[name = string("module_layers_7_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_7_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_7_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113185408)))];
            tensor<fp16, [1, 188, 1024]> linear_72_cast_fp16 = linear(bias = module_layers_7_feed_forward2_linear2_bias_to_fp16, weight = module_layers_7_feed_forward2_linear2_weight_to_fp16_quantized, x = input_427_cast_fp16)[name = string("linear_72_cast_fp16")];
            fp16 var_1743_to_fp16 = const()[name = string("op_1743_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1744_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1743_to_fp16)[name = string("op_1744_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_433_cast_fp16 = add(x = input_421_cast_fp16, y = var_1744_cast_fp16)[name = string("input_433_cast_fp16")];
            tensor<int32, [1]> input_435_axes_0 = const()[name = string("input_435_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_7_norm_out_weight_to_fp16 = const()[name = string("module_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113187520)))];
            tensor<fp16, [1024]> module_layers_7_norm_out_bias_to_fp16 = const()[name = string("module_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113189632)))];
            tensor<fp16, [1, 188, 1024]> input_435_cast_fp16 = layer_norm(axes = input_435_axes_0, beta = module_layers_7_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_7_norm_out_weight_to_fp16, x = input_433_cast_fp16)[name = string("input_435_cast_fp16")];
            tensor<int32, [1]> input_437_axes_0 = const()[name = string("input_437_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113191744)))];
            tensor<fp16, [1024]> module_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113193856)))];
            tensor<fp16, [1, 188, 1024]> input_437_cast_fp16 = layer_norm(axes = input_437_axes_0, beta = module_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward1_weight_to_fp16, x = input_435_cast_fp16)[name = string("input_437_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_8_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(113195968))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115293184))))[name = string("module_layers_8_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_8_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_8_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115555392)))];
            tensor<fp16, [1, 188, 4096]> linear_73_cast_fp16 = linear(bias = module_layers_8_feed_forward1_linear1_bias_to_fp16, weight = module_layers_8_feed_forward1_linear1_weight_to_fp16_quantized, x = input_437_cast_fp16)[name = string("linear_73_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_441_cast_fp16 = silu(x = linear_73_cast_fp16)[name = string("input_441_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_8_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(115563648))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117660864))))[name = string("module_layers_8_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_8_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_8_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117923072)))];
            tensor<fp16, [1, 188, 1024]> linear_74_cast_fp16 = linear(bias = module_layers_8_feed_forward1_linear2_bias_to_fp16, weight = module_layers_8_feed_forward1_linear2_weight_to_fp16_quantized, x = input_441_cast_fp16)[name = string("linear_74_cast_fp16")];
            fp16 var_1774_to_fp16 = const()[name = string("op_1774_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1775_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_1774_to_fp16)[name = string("op_1775_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_447_cast_fp16 = add(x = input_435_cast_fp16, y = var_1775_cast_fp16)[name = string("input_447_cast_fp16")];
            tensor<int32, [1]> query_17_axes_0 = const()[name = string("query_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_8_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117925184)))];
            tensor<fp16, [1024]> module_layers_8_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117927296)))];
            tensor<fp16, [1, 188, 1024]> query_17_cast_fp16 = layer_norm(axes = query_17_axes_0, beta = module_layers_8_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_self_att_weight_to_fp16, x = input_447_cast_fp16)[name = string("query_17_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(117929408))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118453760))))[name = string("module_layers_8_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_8_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_8_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118519360)))];
            tensor<fp16, [1, 188, 1024]> linear_75_cast_fp16 = linear(bias = module_layers_8_self_attn_linear_q_bias_to_fp16, weight = module_layers_8_self_attn_linear_q_weight_to_fp16_quantized, x = query_17_cast_fp16)[name = string("linear_75_cast_fp16")];
            tensor<int32, [4]> var_1792 = const()[name = string("op_1792"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_49_cast_fp16 = reshape(shape = var_1792, x = linear_75_cast_fp16)[name = string("q_49_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118521472))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119045824))))[name = string("module_layers_8_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_8_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_8_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119111424)))];
            tensor<fp16, [1, 188, 1024]> linear_76_cast_fp16 = linear(bias = module_layers_8_self_attn_linear_k_bias_to_fp16, weight = module_layers_8_self_attn_linear_k_weight_to_fp16_quantized, x = query_17_cast_fp16)[name = string("linear_76_cast_fp16")];
            tensor<int32, [4]> var_1797 = const()[name = string("op_1797"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_33_cast_fp16 = reshape(shape = var_1797, x = linear_76_cast_fp16)[name = string("k_33_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119113536))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119637888))))[name = string("module_layers_8_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_8_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_8_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119703488)))];
            tensor<fp16, [1, 188, 1024]> linear_77_cast_fp16 = linear(bias = module_layers_8_self_attn_linear_v_bias_to_fp16, weight = module_layers_8_self_attn_linear_v_weight_to_fp16_quantized, x = query_17_cast_fp16)[name = string("linear_77_cast_fp16")];
            tensor<int32, [4]> var_1802 = const()[name = string("op_1802"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_17_cast_fp16 = reshape(shape = var_1802, x = linear_77_cast_fp16)[name = string("v_17_cast_fp16")];
            tensor<int32, [4]> value_19_perm_0 = const()[name = string("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119705600))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119706176))))[name = string("module_layers_8_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1814_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_1814_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_8_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119706304))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119706880))))[name = string("module_layers_8_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1816_cast_fp16 = add(x = q_49_cast_fp16, y = module_layers_8_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_1816_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_17_perm_0 = const()[name = string("q_with_bias_v_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_183_transpose_x_0 = const()[name = string("x_183_transpose_x_0"), val = bool(false)];
            bool x_183_transpose_y_0 = const()[name = string("x_183_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_1818_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119707008))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119899072))))[name = string("op_1818_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_17_cast_fp16 = transpose(perm = q_with_bias_v_17_perm_0, x = var_1816_cast_fp16)[name = string("transpose_360")];
            tensor<fp16, [1, 8, 188, 375]> x_183_cast_fp16 = matmul(transpose_x = x_183_transpose_x_0, transpose_y = x_183_transpose_y_0, x = q_with_bias_v_17_cast_fp16, y = op_1818_to_fp16_quantized)[name = string("x_183_cast_fp16")];
            tensor<int32, [8]> x_185_pad_0 = const()[name = string("x_185_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("constant")];
            fp16 const_150_to_fp16 = const()[name = string("const_150_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_185_cast_fp16 = pad(constant_val = const_150_to_fp16, mode = x_185_mode_0, pad = x_185_pad_0, x = x_183_cast_fp16)[name = string("x_185_cast_fp16")];
            tensor<int32, [4]> var_1826 = const()[name = string("op_1826"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_187_cast_fp16 = reshape(shape = var_1826, x = x_185_cast_fp16)[name = string("x_187_cast_fp16")];
            tensor<int32, [4]> var_1830_begin_0 = const()[name = string("op_1830_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1830_end_0 = const()[name = string("op_1830_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_1830_end_mask_0 = const()[name = string("op_1830_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_1830_cast_fp16 = slice_by_index(begin = var_1830_begin_0, end = var_1830_end_0, end_mask = var_1830_end_mask_0, x = x_187_cast_fp16)[name = string("op_1830_cast_fp16")];
            tensor<int32, [4]> var_1831 = const()[name = string("op_1831"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_33_cast_fp16 = reshape(shape = var_1831, x = var_1830_cast_fp16)[name = string("matrix_bd_33_cast_fp16")];
            bool matrix_ac_17_transpose_x_0 = const()[name = string("matrix_ac_17_transpose_x_0"), val = bool(false)];
            bool matrix_ac_17_transpose_y_0 = const()[name = string("matrix_ac_17_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_144_perm_0 = const()[name = string("transpose_144_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_145_perm_0 = const()[name = string("transpose_145_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_145 = transpose(perm = transpose_145_perm_0, x = k_33_cast_fp16)[name = string("transpose_358")];
            tensor<fp16, [1, 8, 188, 128]> transpose_144 = transpose(perm = transpose_144_perm_0, x = var_1814_cast_fp16)[name = string("transpose_359")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_17_cast_fp16 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_144, y = transpose_145)[name = string("matrix_ac_17_cast_fp16")];
            tensor<int32, [4]> matrix_bd_35_begin_0 = const()[name = string("matrix_bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_35_end_0 = const()[name = string("matrix_bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_35_end_mask_0 = const()[name = string("matrix_bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_35_cast_fp16 = slice_by_index(begin = matrix_bd_35_begin_0, end = matrix_bd_35_end_0, end_mask = matrix_bd_35_end_mask_0, x = matrix_bd_33_cast_fp16)[name = string("matrix_bd_35_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1840_cast_fp16 = add(x = matrix_ac_17_cast_fp16, y = matrix_bd_35_cast_fp16)[name = string("op_1840_cast_fp16")];
            fp16 _inversed_scores_33_y_0_to_fp16 = const()[name = string("_inversed_scores_33_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_33_cast_fp16 = mul(x = var_1840_cast_fp16, y = _inversed_scores_33_y_0_to_fp16)[name = string("_inversed_scores_33_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_35_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_33_cast_fp16, cond = mask_11)[name = string("scores_35_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_1846_cast_fp16 = softmax(axis = var_23, x = scores_35_cast_fp16)[name = string("op_1846_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_449_cast_fp16 = select(a = var_11_to_fp16, b = var_1846_cast_fp16, cond = mask_11)[name = string("input_449_cast_fp16")];
            bool x_189_transpose_x_0 = const()[name = string("x_189_transpose_x_0"), val = bool(false)];
            bool x_189_transpose_y_0 = const()[name = string("x_189_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = v_17_cast_fp16)[name = string("transpose_357")];
            tensor<fp16, [1, 8, 188, 128]> x_189_cast_fp16 = matmul(transpose_x = x_189_transpose_x_0, transpose_y = x_189_transpose_y_0, x = input_449_cast_fp16, y = value_19_cast_fp16)[name = string("x_189_cast_fp16")];
            tensor<int32, [4]> var_1850_perm_0 = const()[name = string("op_1850_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1851 = const()[name = string("op_1851"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_1850_cast_fp16 = transpose(perm = var_1850_perm_0, x = x_189_cast_fp16)[name = string("transpose_356")];
            tensor<fp16, [1, 188, 1024]> input_451_cast_fp16 = reshape(shape = var_1851, x = var_1850_cast_fp16)[name = string("input_451_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_8_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119902144))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120426496))))[name = string("module_layers_8_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_8_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_8_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120492096)))];
            tensor<fp16, [1, 188, 1024]> linear_79_cast_fp16 = linear(bias = module_layers_8_self_attn_linear_out_bias_to_fp16, weight = module_layers_8_self_attn_linear_out_weight_to_fp16_quantized, x = input_451_cast_fp16)[name = string("linear_79_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_455_cast_fp16 = add(x = input_447_cast_fp16, y = linear_79_cast_fp16)[name = string("input_455_cast_fp16")];
            tensor<int32, [1]> x_193_axes_0 = const()[name = string("x_193_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_8_norm_conv_weight_to_fp16 = const()[name = string("module_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120494208)))];
            tensor<fp16, [1024]> module_layers_8_norm_conv_bias_to_fp16 = const()[name = string("module_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120496320)))];
            tensor<fp16, [1, 188, 1024]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = module_layers_8_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_conv_weight_to_fp16, x = input_455_cast_fp16)[name = string("x_193_cast_fp16")];
            tensor<int32, [3]> input_457_perm_0 = const()[name = string("input_457_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_459_pad_type_0 = const()[name = string("input_459_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_459_strides_0 = const()[name = string("input_459_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_459_pad_0 = const()[name = string("input_459_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_459_dilations_0 = const()[name = string("input_459_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_459_groups_0 = const()[name = string("input_459_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120498432))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121547072))))[name = string("module_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_8_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_8_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121678208)))];
            tensor<fp16, [1, 1024, 188]> input_457_cast_fp16 = transpose(perm = input_457_perm_0, x = x_193_cast_fp16)[name = string("transpose_355")];
            tensor<fp16, [1, 2048, 188]> input_459_cast_fp16 = conv(bias = module_layers_8_conv_pointwise_conv1_bias_to_fp16, dilations = input_459_dilations_0, groups = input_459_groups_0, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = input_459_strides_0, weight = module_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_457_cast_fp16)[name = string("input_459_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, 188]> x_195_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_459_cast_fp16)[name = string("x_195_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_461_cast_fp16 = select(a = var_11_to_fp16, b = x_195_cast_fp16, cond = var_483)[name = string("input_461_cast_fp16")];
            tensor<int32, [6]> input_463_pad_0 = const()[name = string("input_463_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_463_mode_0 = const()[name = string("input_463_mode_0"), val = string("constant")];
            fp16 const_153_to_fp16 = const()[name = string("const_153_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_463_cast_fp16 = pad(constant_val = const_153_to_fp16, mode = input_463_mode_0, pad = input_463_pad_0, x = input_461_cast_fp16)[name = string("input_463_cast_fp16")];
            string input_465_pad_type_0 = const()[name = string("input_465_pad_type_0"), val = string("valid")];
            int32 input_465_groups_0 = const()[name = string("input_465_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_465_strides_0 = const()[name = string("input_465_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_465_pad_0 = const()[name = string("input_465_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_465_dilations_0 = const()[name = string("input_465_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_400_to_fp16 = const()[name = string("const_400_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121682368)))];
            tensor<fp16, [1024]> const_401_to_fp16 = const()[name = string("const_401_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121700864)))];
            tensor<fp16, [1, 1024, 188]> input_467_cast_fp16 = conv(bias = const_401_to_fp16, dilations = input_465_dilations_0, groups = input_465_groups_0, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = input_465_strides_0, weight = const_400_to_fp16, x = input_463_cast_fp16)[name = string("input_467_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_469_cast_fp16 = silu(x = input_467_cast_fp16)[name = string("input_469_cast_fp16")];
            string x_197_pad_type_0 = const()[name = string("x_197_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_197_strides_0 = const()[name = string("x_197_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_197_pad_0 = const()[name = string("x_197_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_197_dilations_0 = const()[name = string("x_197_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_197_groups_0 = const()[name = string("x_197_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121702976))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122227328))))[name = string("module_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_8_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_8_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122292928)))];
            tensor<fp16, [1, 1024, 188]> x_197_cast_fp16 = conv(bias = module_layers_8_conv_pointwise_conv2_bias_to_fp16, dilations = x_197_dilations_0, groups = x_197_groups_0, pad = x_197_pad_0, pad_type = x_197_pad_type_0, strides = x_197_strides_0, weight = module_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_469_cast_fp16)[name = string("x_197_cast_fp16")];
            tensor<int32, [3]> input_471_perm_0 = const()[name = string("input_471_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_471_cast_fp16 = transpose(perm = input_471_perm_0, x = x_197_cast_fp16)[name = string("transpose_354")];
            tensor<fp16, [1, 188, 1024]> input_473_cast_fp16 = add(x = input_455_cast_fp16, y = input_471_cast_fp16)[name = string("input_473_cast_fp16")];
            tensor<int32, [1]> input_475_axes_0 = const()[name = string("input_475_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122295040)))];
            tensor<fp16, [1024]> module_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122297152)))];
            tensor<fp16, [1, 188, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = module_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_feed_forward2_weight_to_fp16, x = input_473_cast_fp16)[name = string("input_475_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_8_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(122299264))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124396480))))[name = string("module_layers_8_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_8_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_8_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124658688)))];
            tensor<fp16, [1, 188, 4096]> linear_80_cast_fp16 = linear(bias = module_layers_8_feed_forward2_linear1_bias_to_fp16, weight = module_layers_8_feed_forward2_linear1_weight_to_fp16_quantized, x = input_475_cast_fp16)[name = string("linear_80_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_479_cast_fp16 = silu(x = linear_80_cast_fp16)[name = string("input_479_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_8_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124666944))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(126764160))))[name = string("module_layers_8_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_8_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_8_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127026368)))];
            tensor<fp16, [1, 188, 1024]> linear_81_cast_fp16 = linear(bias = module_layers_8_feed_forward2_linear2_bias_to_fp16, weight = module_layers_8_feed_forward2_linear2_weight_to_fp16_quantized, x = input_479_cast_fp16)[name = string("linear_81_cast_fp16")];
            fp16 var_1917_to_fp16 = const()[name = string("op_1917_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1918_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_1917_to_fp16)[name = string("op_1918_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_485_cast_fp16 = add(x = input_473_cast_fp16, y = var_1918_cast_fp16)[name = string("input_485_cast_fp16")];
            tensor<int32, [1]> input_487_axes_0 = const()[name = string("input_487_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_8_norm_out_weight_to_fp16 = const()[name = string("module_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127028480)))];
            tensor<fp16, [1024]> module_layers_8_norm_out_bias_to_fp16 = const()[name = string("module_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127030592)))];
            tensor<fp16, [1, 188, 1024]> input_487_cast_fp16 = layer_norm(axes = input_487_axes_0, beta = module_layers_8_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_8_norm_out_weight_to_fp16, x = input_485_cast_fp16)[name = string("input_487_cast_fp16")];
            tensor<int32, [1]> input_489_axes_0 = const()[name = string("input_489_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127032704)))];
            tensor<fp16, [1024]> module_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127034816)))];
            tensor<fp16, [1, 188, 1024]> input_489_cast_fp16 = layer_norm(axes = input_489_axes_0, beta = module_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward1_weight_to_fp16, x = input_487_cast_fp16)[name = string("input_489_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_9_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127036928))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129134144))))[name = string("module_layers_9_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_9_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_9_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129396352)))];
            tensor<fp16, [1, 188, 4096]> linear_82_cast_fp16 = linear(bias = module_layers_9_feed_forward1_linear1_bias_to_fp16, weight = module_layers_9_feed_forward1_linear1_weight_to_fp16_quantized, x = input_489_cast_fp16)[name = string("linear_82_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_493_cast_fp16 = silu(x = linear_82_cast_fp16)[name = string("input_493_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_9_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129404608))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131501824))))[name = string("module_layers_9_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_9_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_9_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131764032)))];
            tensor<fp16, [1, 188, 1024]> linear_83_cast_fp16 = linear(bias = module_layers_9_feed_forward1_linear2_bias_to_fp16, weight = module_layers_9_feed_forward1_linear2_weight_to_fp16_quantized, x = input_493_cast_fp16)[name = string("linear_83_cast_fp16")];
            fp16 var_1948_to_fp16 = const()[name = string("op_1948_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_1949_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_1948_to_fp16)[name = string("op_1949_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_499_cast_fp16 = add(x = input_487_cast_fp16, y = var_1949_cast_fp16)[name = string("input_499_cast_fp16")];
            tensor<int32, [1]> query_19_axes_0 = const()[name = string("query_19_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_9_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131766144)))];
            tensor<fp16, [1024]> module_layers_9_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131768256)))];
            tensor<fp16, [1, 188, 1024]> query_19_cast_fp16 = layer_norm(axes = query_19_axes_0, beta = module_layers_9_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_self_att_weight_to_fp16, x = input_499_cast_fp16)[name = string("query_19_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131770368))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132294720))))[name = string("module_layers_9_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_9_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_9_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132360320)))];
            tensor<fp16, [1, 188, 1024]> linear_84_cast_fp16 = linear(bias = module_layers_9_self_attn_linear_q_bias_to_fp16, weight = module_layers_9_self_attn_linear_q_weight_to_fp16_quantized, x = query_19_cast_fp16)[name = string("linear_84_cast_fp16")];
            tensor<int32, [4]> var_1966 = const()[name = string("op_1966"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_55_cast_fp16 = reshape(shape = var_1966, x = linear_84_cast_fp16)[name = string("q_55_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132362432))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132886784))))[name = string("module_layers_9_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_9_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_9_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132952384)))];
            tensor<fp16, [1, 188, 1024]> linear_85_cast_fp16 = linear(bias = module_layers_9_self_attn_linear_k_bias_to_fp16, weight = module_layers_9_self_attn_linear_k_weight_to_fp16_quantized, x = query_19_cast_fp16)[name = string("linear_85_cast_fp16")];
            tensor<int32, [4]> var_1971 = const()[name = string("op_1971"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_37_cast_fp16 = reshape(shape = var_1971, x = linear_85_cast_fp16)[name = string("k_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(132954496))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133478848))))[name = string("module_layers_9_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_9_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_9_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133544448)))];
            tensor<fp16, [1, 188, 1024]> linear_86_cast_fp16 = linear(bias = module_layers_9_self_attn_linear_v_bias_to_fp16, weight = module_layers_9_self_attn_linear_v_weight_to_fp16_quantized, x = query_19_cast_fp16)[name = string("linear_86_cast_fp16")];
            tensor<int32, [4]> var_1976 = const()[name = string("op_1976"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_19_cast_fp16 = reshape(shape = var_1976, x = linear_86_cast_fp16)[name = string("v_19_cast_fp16")];
            tensor<int32, [4]> value_21_perm_0 = const()[name = string("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133546560))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133547136))))[name = string("module_layers_9_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1988_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_1988_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_9_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133547264))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133547840))))[name = string("module_layers_9_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_1990_cast_fp16 = add(x = q_55_cast_fp16, y = module_layers_9_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_1990_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_19_perm_0 = const()[name = string("q_with_bias_v_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_205_transpose_x_0 = const()[name = string("x_205_transpose_x_0"), val = bool(false)];
            bool x_205_transpose_y_0 = const()[name = string("x_205_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_1992_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133547968))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133740032))))[name = string("op_1992_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_19_cast_fp16 = transpose(perm = q_with_bias_v_19_perm_0, x = var_1990_cast_fp16)[name = string("transpose_353")];
            tensor<fp16, [1, 8, 188, 375]> x_205_cast_fp16 = matmul(transpose_x = x_205_transpose_x_0, transpose_y = x_205_transpose_y_0, x = q_with_bias_v_19_cast_fp16, y = op_1992_to_fp16_quantized)[name = string("x_205_cast_fp16")];
            tensor<int32, [8]> x_207_pad_0 = const()[name = string("x_207_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_207_mode_0 = const()[name = string("x_207_mode_0"), val = string("constant")];
            fp16 const_160_to_fp16 = const()[name = string("const_160_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_207_cast_fp16 = pad(constant_val = const_160_to_fp16, mode = x_207_mode_0, pad = x_207_pad_0, x = x_205_cast_fp16)[name = string("x_207_cast_fp16")];
            tensor<int32, [4]> var_2000 = const()[name = string("op_2000"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_209_cast_fp16 = reshape(shape = var_2000, x = x_207_cast_fp16)[name = string("x_209_cast_fp16")];
            tensor<int32, [4]> var_2004_begin_0 = const()[name = string("op_2004_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2004_end_0 = const()[name = string("op_2004_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2004_end_mask_0 = const()[name = string("op_2004_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2004_cast_fp16 = slice_by_index(begin = var_2004_begin_0, end = var_2004_end_0, end_mask = var_2004_end_mask_0, x = x_209_cast_fp16)[name = string("op_2004_cast_fp16")];
            tensor<int32, [4]> var_2005 = const()[name = string("op_2005"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_37_cast_fp16 = reshape(shape = var_2005, x = var_2004_cast_fp16)[name = string("matrix_bd_37_cast_fp16")];
            bool matrix_ac_19_transpose_x_0 = const()[name = string("matrix_ac_19_transpose_x_0"), val = bool(false)];
            bool matrix_ac_19_transpose_y_0 = const()[name = string("matrix_ac_19_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_146_perm_0 = const()[name = string("transpose_146_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_147_perm_0 = const()[name = string("transpose_147_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_147 = transpose(perm = transpose_147_perm_0, x = k_37_cast_fp16)[name = string("transpose_351")];
            tensor<fp16, [1, 8, 188, 128]> transpose_146 = transpose(perm = transpose_146_perm_0, x = var_1988_cast_fp16)[name = string("transpose_352")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_19_cast_fp16 = matmul(transpose_x = matrix_ac_19_transpose_x_0, transpose_y = matrix_ac_19_transpose_y_0, x = transpose_146, y = transpose_147)[name = string("matrix_ac_19_cast_fp16")];
            tensor<int32, [4]> matrix_bd_39_begin_0 = const()[name = string("matrix_bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_39_end_0 = const()[name = string("matrix_bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_39_end_mask_0 = const()[name = string("matrix_bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_39_cast_fp16 = slice_by_index(begin = matrix_bd_39_begin_0, end = matrix_bd_39_end_0, end_mask = matrix_bd_39_end_mask_0, x = matrix_bd_37_cast_fp16)[name = string("matrix_bd_39_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2014_cast_fp16 = add(x = matrix_ac_19_cast_fp16, y = matrix_bd_39_cast_fp16)[name = string("op_2014_cast_fp16")];
            fp16 _inversed_scores_37_y_0_to_fp16 = const()[name = string("_inversed_scores_37_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_37_cast_fp16 = mul(x = var_2014_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = string("_inversed_scores_37_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_39_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_37_cast_fp16, cond = mask_11)[name = string("scores_39_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2020_cast_fp16 = softmax(axis = var_23, x = scores_39_cast_fp16)[name = string("op_2020_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_501_cast_fp16 = select(a = var_11_to_fp16, b = var_2020_cast_fp16, cond = mask_11)[name = string("input_501_cast_fp16")];
            bool x_211_transpose_x_0 = const()[name = string("x_211_transpose_x_0"), val = bool(false)];
            bool x_211_transpose_y_0 = const()[name = string("x_211_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = v_19_cast_fp16)[name = string("transpose_350")];
            tensor<fp16, [1, 8, 188, 128]> x_211_cast_fp16 = matmul(transpose_x = x_211_transpose_x_0, transpose_y = x_211_transpose_y_0, x = input_501_cast_fp16, y = value_21_cast_fp16)[name = string("x_211_cast_fp16")];
            tensor<int32, [4]> var_2024_perm_0 = const()[name = string("op_2024_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2025 = const()[name = string("op_2025"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2024_cast_fp16 = transpose(perm = var_2024_perm_0, x = x_211_cast_fp16)[name = string("transpose_349")];
            tensor<fp16, [1, 188, 1024]> input_503_cast_fp16 = reshape(shape = var_2025, x = var_2024_cast_fp16)[name = string("input_503_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_9_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133743104))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134267456))))[name = string("module_layers_9_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_9_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_9_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134333056)))];
            tensor<fp16, [1, 188, 1024]> linear_88_cast_fp16 = linear(bias = module_layers_9_self_attn_linear_out_bias_to_fp16, weight = module_layers_9_self_attn_linear_out_weight_to_fp16_quantized, x = input_503_cast_fp16)[name = string("linear_88_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_507_cast_fp16 = add(x = input_499_cast_fp16, y = linear_88_cast_fp16)[name = string("input_507_cast_fp16")];
            tensor<int32, [1]> x_215_axes_0 = const()[name = string("x_215_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_9_norm_conv_weight_to_fp16 = const()[name = string("module_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134335168)))];
            tensor<fp16, [1024]> module_layers_9_norm_conv_bias_to_fp16 = const()[name = string("module_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134337280)))];
            tensor<fp16, [1, 188, 1024]> x_215_cast_fp16 = layer_norm(axes = x_215_axes_0, beta = module_layers_9_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = string("x_215_cast_fp16")];
            tensor<int32, [3]> input_509_perm_0 = const()[name = string("input_509_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_511_pad_type_0 = const()[name = string("input_511_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_511_strides_0 = const()[name = string("input_511_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_511_pad_0 = const()[name = string("input_511_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_511_dilations_0 = const()[name = string("input_511_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_511_groups_0 = const()[name = string("input_511_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134339392))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135388032))))[name = string("module_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_9_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_9_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135519168)))];
            tensor<fp16, [1, 1024, 188]> input_509_cast_fp16 = transpose(perm = input_509_perm_0, x = x_215_cast_fp16)[name = string("transpose_348")];
            tensor<fp16, [1, 2048, 188]> input_511_cast_fp16 = conv(bias = module_layers_9_conv_pointwise_conv1_bias_to_fp16, dilations = input_511_dilations_0, groups = input_511_groups_0, pad = input_511_pad_0, pad_type = input_511_pad_type_0, strides = input_511_strides_0, weight = module_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_509_cast_fp16)[name = string("input_511_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, 188]> x_217_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_511_cast_fp16)[name = string("x_217_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_513_cast_fp16 = select(a = var_11_to_fp16, b = x_217_cast_fp16, cond = var_483)[name = string("input_513_cast_fp16")];
            tensor<int32, [6]> input_515_pad_0 = const()[name = string("input_515_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_515_mode_0 = const()[name = string("input_515_mode_0"), val = string("constant")];
            fp16 const_163_to_fp16 = const()[name = string("const_163_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_515_cast_fp16 = pad(constant_val = const_163_to_fp16, mode = input_515_mode_0, pad = input_515_pad_0, x = input_513_cast_fp16)[name = string("input_515_cast_fp16")];
            string input_517_pad_type_0 = const()[name = string("input_517_pad_type_0"), val = string("valid")];
            int32 input_517_groups_0 = const()[name = string("input_517_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_517_strides_0 = const()[name = string("input_517_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_517_pad_0 = const()[name = string("input_517_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_517_dilations_0 = const()[name = string("input_517_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_402_to_fp16 = const()[name = string("const_402_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135523328)))];
            tensor<fp16, [1024]> const_403_to_fp16 = const()[name = string("const_403_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135541824)))];
            tensor<fp16, [1, 1024, 188]> input_519_cast_fp16 = conv(bias = const_403_to_fp16, dilations = input_517_dilations_0, groups = input_517_groups_0, pad = input_517_pad_0, pad_type = input_517_pad_type_0, strides = input_517_strides_0, weight = const_402_to_fp16, x = input_515_cast_fp16)[name = string("input_519_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_521_cast_fp16 = silu(x = input_519_cast_fp16)[name = string("input_521_cast_fp16")];
            string x_219_pad_type_0 = const()[name = string("x_219_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_219_strides_0 = const()[name = string("x_219_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_219_pad_0 = const()[name = string("x_219_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_219_dilations_0 = const()[name = string("x_219_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_219_groups_0 = const()[name = string("x_219_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135543936))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136068288))))[name = string("module_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_9_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_9_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136133888)))];
            tensor<fp16, [1, 1024, 188]> x_219_cast_fp16 = conv(bias = module_layers_9_conv_pointwise_conv2_bias_to_fp16, dilations = x_219_dilations_0, groups = x_219_groups_0, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = x_219_strides_0, weight = module_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_521_cast_fp16)[name = string("x_219_cast_fp16")];
            tensor<int32, [3]> input_523_perm_0 = const()[name = string("input_523_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_523_cast_fp16 = transpose(perm = input_523_perm_0, x = x_219_cast_fp16)[name = string("transpose_347")];
            tensor<fp16, [1, 188, 1024]> input_525_cast_fp16 = add(x = input_507_cast_fp16, y = input_523_cast_fp16)[name = string("input_525_cast_fp16")];
            tensor<int32, [1]> input_527_axes_0 = const()[name = string("input_527_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136136000)))];
            tensor<fp16, [1024]> module_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136138112)))];
            tensor<fp16, [1, 188, 1024]> input_527_cast_fp16 = layer_norm(axes = input_527_axes_0, beta = module_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_feed_forward2_weight_to_fp16, x = input_525_cast_fp16)[name = string("input_527_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_9_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136140224))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138237440))))[name = string("module_layers_9_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_9_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_9_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138499648)))];
            tensor<fp16, [1, 188, 4096]> linear_89_cast_fp16 = linear(bias = module_layers_9_feed_forward2_linear1_bias_to_fp16, weight = module_layers_9_feed_forward2_linear1_weight_to_fp16_quantized, x = input_527_cast_fp16)[name = string("linear_89_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_531_cast_fp16 = silu(x = linear_89_cast_fp16)[name = string("input_531_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_9_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138507904))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140605120))))[name = string("module_layers_9_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_9_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_9_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140867328)))];
            tensor<fp16, [1, 188, 1024]> linear_90_cast_fp16 = linear(bias = module_layers_9_feed_forward2_linear2_bias_to_fp16, weight = module_layers_9_feed_forward2_linear2_weight_to_fp16_quantized, x = input_531_cast_fp16)[name = string("linear_90_cast_fp16")];
            fp16 var_2091_to_fp16 = const()[name = string("op_2091_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2092_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_2091_to_fp16)[name = string("op_2092_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_537_cast_fp16 = add(x = input_525_cast_fp16, y = var_2092_cast_fp16)[name = string("input_537_cast_fp16")];
            tensor<int32, [1]> input_539_axes_0 = const()[name = string("input_539_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_9_norm_out_weight_to_fp16 = const()[name = string("module_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140869440)))];
            tensor<fp16, [1024]> module_layers_9_norm_out_bias_to_fp16 = const()[name = string("module_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140871552)))];
            tensor<fp16, [1, 188, 1024]> input_539_cast_fp16 = layer_norm(axes = input_539_axes_0, beta = module_layers_9_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_9_norm_out_weight_to_fp16, x = input_537_cast_fp16)[name = string("input_539_cast_fp16")];
            tensor<int32, [1]> input_541_axes_0 = const()[name = string("input_541_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140873664)))];
            tensor<fp16, [1024]> module_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140875776)))];
            tensor<fp16, [1, 188, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = module_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward1_weight_to_fp16, x = input_539_cast_fp16)[name = string("input_541_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_10_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140877888))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142975104))))[name = string("module_layers_10_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_10_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_10_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143237312)))];
            tensor<fp16, [1, 188, 4096]> linear_91_cast_fp16 = linear(bias = module_layers_10_feed_forward1_linear1_bias_to_fp16, weight = module_layers_10_feed_forward1_linear1_weight_to_fp16_quantized, x = input_541_cast_fp16)[name = string("linear_91_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_545_cast_fp16 = silu(x = linear_91_cast_fp16)[name = string("input_545_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_10_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143245568))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145342784))))[name = string("module_layers_10_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_10_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_10_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145604992)))];
            tensor<fp16, [1, 188, 1024]> linear_92_cast_fp16 = linear(bias = module_layers_10_feed_forward1_linear2_bias_to_fp16, weight = module_layers_10_feed_forward1_linear2_weight_to_fp16_quantized, x = input_545_cast_fp16)[name = string("linear_92_cast_fp16")];
            fp16 var_2122_to_fp16 = const()[name = string("op_2122_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2123_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_2122_to_fp16)[name = string("op_2123_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_551_cast_fp16 = add(x = input_539_cast_fp16, y = var_2123_cast_fp16)[name = string("input_551_cast_fp16")];
            tensor<int32, [1]> query_21_axes_0 = const()[name = string("query_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_10_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145607104)))];
            tensor<fp16, [1024]> module_layers_10_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145609216)))];
            tensor<fp16, [1, 188, 1024]> query_21_cast_fp16 = layer_norm(axes = query_21_axes_0, beta = module_layers_10_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = string("query_21_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(145611328))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146135680))))[name = string("module_layers_10_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_10_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_10_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146201280)))];
            tensor<fp16, [1, 188, 1024]> linear_93_cast_fp16 = linear(bias = module_layers_10_self_attn_linear_q_bias_to_fp16, weight = module_layers_10_self_attn_linear_q_weight_to_fp16_quantized, x = query_21_cast_fp16)[name = string("linear_93_cast_fp16")];
            tensor<int32, [4]> var_2140 = const()[name = string("op_2140"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_61_cast_fp16 = reshape(shape = var_2140, x = linear_93_cast_fp16)[name = string("q_61_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146203392))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146727744))))[name = string("module_layers_10_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_10_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_10_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146793344)))];
            tensor<fp16, [1, 188, 1024]> linear_94_cast_fp16 = linear(bias = module_layers_10_self_attn_linear_k_bias_to_fp16, weight = module_layers_10_self_attn_linear_k_weight_to_fp16_quantized, x = query_21_cast_fp16)[name = string("linear_94_cast_fp16")];
            tensor<int32, [4]> var_2145 = const()[name = string("op_2145"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_41_cast_fp16 = reshape(shape = var_2145, x = linear_94_cast_fp16)[name = string("k_41_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(146795456))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147319808))))[name = string("module_layers_10_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_10_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_10_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147385408)))];
            tensor<fp16, [1, 188, 1024]> linear_95_cast_fp16 = linear(bias = module_layers_10_self_attn_linear_v_bias_to_fp16, weight = module_layers_10_self_attn_linear_v_weight_to_fp16_quantized, x = query_21_cast_fp16)[name = string("linear_95_cast_fp16")];
            tensor<int32, [4]> var_2150 = const()[name = string("op_2150"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_21_cast_fp16 = reshape(shape = var_2150, x = linear_95_cast_fp16)[name = string("v_21_cast_fp16")];
            tensor<int32, [4]> value_23_perm_0 = const()[name = string("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147387520))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147388096))))[name = string("module_layers_10_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2162_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_2162_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_10_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147388224))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147388800))))[name = string("module_layers_10_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2164_cast_fp16 = add(x = q_61_cast_fp16, y = module_layers_10_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_2164_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_21_perm_0 = const()[name = string("q_with_bias_v_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_227_transpose_x_0 = const()[name = string("x_227_transpose_x_0"), val = bool(false)];
            bool x_227_transpose_y_0 = const()[name = string("x_227_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_2166_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147388928))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147580992))))[name = string("op_2166_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_21_cast_fp16 = transpose(perm = q_with_bias_v_21_perm_0, x = var_2164_cast_fp16)[name = string("transpose_346")];
            tensor<fp16, [1, 8, 188, 375]> x_227_cast_fp16 = matmul(transpose_x = x_227_transpose_x_0, transpose_y = x_227_transpose_y_0, x = q_with_bias_v_21_cast_fp16, y = op_2166_to_fp16_quantized)[name = string("x_227_cast_fp16")];
            tensor<int32, [8]> x_229_pad_0 = const()[name = string("x_229_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_229_mode_0 = const()[name = string("x_229_mode_0"), val = string("constant")];
            fp16 const_170_to_fp16 = const()[name = string("const_170_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_229_cast_fp16 = pad(constant_val = const_170_to_fp16, mode = x_229_mode_0, pad = x_229_pad_0, x = x_227_cast_fp16)[name = string("x_229_cast_fp16")];
            tensor<int32, [4]> var_2174 = const()[name = string("op_2174"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_231_cast_fp16 = reshape(shape = var_2174, x = x_229_cast_fp16)[name = string("x_231_cast_fp16")];
            tensor<int32, [4]> var_2178_begin_0 = const()[name = string("op_2178_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2178_end_0 = const()[name = string("op_2178_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2178_end_mask_0 = const()[name = string("op_2178_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2178_cast_fp16 = slice_by_index(begin = var_2178_begin_0, end = var_2178_end_0, end_mask = var_2178_end_mask_0, x = x_231_cast_fp16)[name = string("op_2178_cast_fp16")];
            tensor<int32, [4]> var_2179 = const()[name = string("op_2179"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_41_cast_fp16 = reshape(shape = var_2179, x = var_2178_cast_fp16)[name = string("matrix_bd_41_cast_fp16")];
            bool matrix_ac_21_transpose_x_0 = const()[name = string("matrix_ac_21_transpose_x_0"), val = bool(false)];
            bool matrix_ac_21_transpose_y_0 = const()[name = string("matrix_ac_21_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_148_perm_0 = const()[name = string("transpose_148_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_149_perm_0 = const()[name = string("transpose_149_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_149 = transpose(perm = transpose_149_perm_0, x = k_41_cast_fp16)[name = string("transpose_344")];
            tensor<fp16, [1, 8, 188, 128]> transpose_148 = transpose(perm = transpose_148_perm_0, x = var_2162_cast_fp16)[name = string("transpose_345")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_21_cast_fp16 = matmul(transpose_x = matrix_ac_21_transpose_x_0, transpose_y = matrix_ac_21_transpose_y_0, x = transpose_148, y = transpose_149)[name = string("matrix_ac_21_cast_fp16")];
            tensor<int32, [4]> matrix_bd_43_begin_0 = const()[name = string("matrix_bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_43_end_0 = const()[name = string("matrix_bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_43_end_mask_0 = const()[name = string("matrix_bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_43_cast_fp16 = slice_by_index(begin = matrix_bd_43_begin_0, end = matrix_bd_43_end_0, end_mask = matrix_bd_43_end_mask_0, x = matrix_bd_41_cast_fp16)[name = string("matrix_bd_43_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2188_cast_fp16 = add(x = matrix_ac_21_cast_fp16, y = matrix_bd_43_cast_fp16)[name = string("op_2188_cast_fp16")];
            fp16 _inversed_scores_41_y_0_to_fp16 = const()[name = string("_inversed_scores_41_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_41_cast_fp16 = mul(x = var_2188_cast_fp16, y = _inversed_scores_41_y_0_to_fp16)[name = string("_inversed_scores_41_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_43_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_41_cast_fp16, cond = mask_11)[name = string("scores_43_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2194_cast_fp16 = softmax(axis = var_23, x = scores_43_cast_fp16)[name = string("op_2194_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_553_cast_fp16 = select(a = var_11_to_fp16, b = var_2194_cast_fp16, cond = mask_11)[name = string("input_553_cast_fp16")];
            bool x_233_transpose_x_0 = const()[name = string("x_233_transpose_x_0"), val = bool(false)];
            bool x_233_transpose_y_0 = const()[name = string("x_233_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = v_21_cast_fp16)[name = string("transpose_343")];
            tensor<fp16, [1, 8, 188, 128]> x_233_cast_fp16 = matmul(transpose_x = x_233_transpose_x_0, transpose_y = x_233_transpose_y_0, x = input_553_cast_fp16, y = value_23_cast_fp16)[name = string("x_233_cast_fp16")];
            tensor<int32, [4]> var_2198_perm_0 = const()[name = string("op_2198_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2199 = const()[name = string("op_2199"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2198_cast_fp16 = transpose(perm = var_2198_perm_0, x = x_233_cast_fp16)[name = string("transpose_342")];
            tensor<fp16, [1, 188, 1024]> input_555_cast_fp16 = reshape(shape = var_2199, x = var_2198_cast_fp16)[name = string("input_555_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_10_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147584064))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148108416))))[name = string("module_layers_10_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_10_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_10_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148174016)))];
            tensor<fp16, [1, 188, 1024]> linear_97_cast_fp16 = linear(bias = module_layers_10_self_attn_linear_out_bias_to_fp16, weight = module_layers_10_self_attn_linear_out_weight_to_fp16_quantized, x = input_555_cast_fp16)[name = string("linear_97_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_97_cast_fp16)[name = string("input_559_cast_fp16")];
            tensor<int32, [1]> x_237_axes_0 = const()[name = string("x_237_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_10_norm_conv_weight_to_fp16 = const()[name = string("module_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148176128)))];
            tensor<fp16, [1024]> module_layers_10_norm_conv_bias_to_fp16 = const()[name = string("module_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148178240)))];
            tensor<fp16, [1, 188, 1024]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = module_layers_10_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = string("x_237_cast_fp16")];
            tensor<int32, [3]> input_561_perm_0 = const()[name = string("input_561_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_563_pad_type_0 = const()[name = string("input_563_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_563_strides_0 = const()[name = string("input_563_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_563_pad_0 = const()[name = string("input_563_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_563_dilations_0 = const()[name = string("input_563_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_563_groups_0 = const()[name = string("input_563_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(148180352))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149228992))))[name = string("module_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_10_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_10_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149360128)))];
            tensor<fp16, [1, 1024, 188]> input_561_cast_fp16 = transpose(perm = input_561_perm_0, x = x_237_cast_fp16)[name = string("transpose_341")];
            tensor<fp16, [1, 2048, 188]> input_563_cast_fp16 = conv(bias = module_layers_10_conv_pointwise_conv1_bias_to_fp16, dilations = input_563_dilations_0, groups = input_563_groups_0, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = input_563_strides_0, weight = module_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_561_cast_fp16)[name = string("input_563_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, 188]> x_239_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_563_cast_fp16)[name = string("x_239_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_565_cast_fp16 = select(a = var_11_to_fp16, b = x_239_cast_fp16, cond = var_483)[name = string("input_565_cast_fp16")];
            tensor<int32, [6]> input_567_pad_0 = const()[name = string("input_567_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_567_mode_0 = const()[name = string("input_567_mode_0"), val = string("constant")];
            fp16 const_173_to_fp16 = const()[name = string("const_173_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_567_cast_fp16 = pad(constant_val = const_173_to_fp16, mode = input_567_mode_0, pad = input_567_pad_0, x = input_565_cast_fp16)[name = string("input_567_cast_fp16")];
            string input_569_pad_type_0 = const()[name = string("input_569_pad_type_0"), val = string("valid")];
            int32 input_569_groups_0 = const()[name = string("input_569_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_569_strides_0 = const()[name = string("input_569_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_569_pad_0 = const()[name = string("input_569_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_569_dilations_0 = const()[name = string("input_569_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_404_to_fp16 = const()[name = string("const_404_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149364288)))];
            tensor<fp16, [1024]> const_405_to_fp16 = const()[name = string("const_405_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149382784)))];
            tensor<fp16, [1, 1024, 188]> input_571_cast_fp16 = conv(bias = const_405_to_fp16, dilations = input_569_dilations_0, groups = input_569_groups_0, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = input_569_strides_0, weight = const_404_to_fp16, x = input_567_cast_fp16)[name = string("input_571_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_573_cast_fp16 = silu(x = input_571_cast_fp16)[name = string("input_573_cast_fp16")];
            string x_241_pad_type_0 = const()[name = string("x_241_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_241_strides_0 = const()[name = string("x_241_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_241_pad_0 = const()[name = string("x_241_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_241_dilations_0 = const()[name = string("x_241_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_241_groups_0 = const()[name = string("x_241_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149384896))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149909248))))[name = string("module_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_10_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_10_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149974848)))];
            tensor<fp16, [1, 1024, 188]> x_241_cast_fp16 = conv(bias = module_layers_10_conv_pointwise_conv2_bias_to_fp16, dilations = x_241_dilations_0, groups = x_241_groups_0, pad = x_241_pad_0, pad_type = x_241_pad_type_0, strides = x_241_strides_0, weight = module_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_573_cast_fp16)[name = string("x_241_cast_fp16")];
            tensor<int32, [3]> input_575_perm_0 = const()[name = string("input_575_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_575_cast_fp16 = transpose(perm = input_575_perm_0, x = x_241_cast_fp16)[name = string("transpose_340")];
            tensor<fp16, [1, 188, 1024]> input_577_cast_fp16 = add(x = input_559_cast_fp16, y = input_575_cast_fp16)[name = string("input_577_cast_fp16")];
            tensor<int32, [1]> input_579_axes_0 = const()[name = string("input_579_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149976960)))];
            tensor<fp16, [1024]> module_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149979072)))];
            tensor<fp16, [1, 188, 1024]> input_579_cast_fp16 = layer_norm(axes = input_579_axes_0, beta = module_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_feed_forward2_weight_to_fp16, x = input_577_cast_fp16)[name = string("input_579_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_10_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149981184))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152078400))))[name = string("module_layers_10_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_10_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_10_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152340608)))];
            tensor<fp16, [1, 188, 4096]> linear_98_cast_fp16 = linear(bias = module_layers_10_feed_forward2_linear1_bias_to_fp16, weight = module_layers_10_feed_forward2_linear1_weight_to_fp16_quantized, x = input_579_cast_fp16)[name = string("linear_98_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_583_cast_fp16 = silu(x = linear_98_cast_fp16)[name = string("input_583_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_10_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152348864))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154446080))))[name = string("module_layers_10_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_10_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_10_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154708288)))];
            tensor<fp16, [1, 188, 1024]> linear_99_cast_fp16 = linear(bias = module_layers_10_feed_forward2_linear2_bias_to_fp16, weight = module_layers_10_feed_forward2_linear2_weight_to_fp16_quantized, x = input_583_cast_fp16)[name = string("linear_99_cast_fp16")];
            fp16 var_2265_to_fp16 = const()[name = string("op_2265_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2266_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_2265_to_fp16)[name = string("op_2266_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_589_cast_fp16 = add(x = input_577_cast_fp16, y = var_2266_cast_fp16)[name = string("input_589_cast_fp16")];
            tensor<int32, [1]> input_591_axes_0 = const()[name = string("input_591_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_10_norm_out_weight_to_fp16 = const()[name = string("module_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154710400)))];
            tensor<fp16, [1024]> module_layers_10_norm_out_bias_to_fp16 = const()[name = string("module_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154712512)))];
            tensor<fp16, [1, 188, 1024]> input_591_cast_fp16 = layer_norm(axes = input_591_axes_0, beta = module_layers_10_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_10_norm_out_weight_to_fp16, x = input_589_cast_fp16)[name = string("input_591_cast_fp16")];
            tensor<int32, [1]> input_593_axes_0 = const()[name = string("input_593_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154714624)))];
            tensor<fp16, [1024]> module_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154716736)))];
            tensor<fp16, [1, 188, 1024]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = module_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward1_weight_to_fp16, x = input_591_cast_fp16)[name = string("input_593_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_11_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(154718848))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(156816064))))[name = string("module_layers_11_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_11_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_11_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157078272)))];
            tensor<fp16, [1, 188, 4096]> linear_100_cast_fp16 = linear(bias = module_layers_11_feed_forward1_linear1_bias_to_fp16, weight = module_layers_11_feed_forward1_linear1_weight_to_fp16_quantized, x = input_593_cast_fp16)[name = string("linear_100_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_597_cast_fp16 = silu(x = linear_100_cast_fp16)[name = string("input_597_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_11_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157086528))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159183744))))[name = string("module_layers_11_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_11_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_11_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159445952)))];
            tensor<fp16, [1, 188, 1024]> linear_101_cast_fp16 = linear(bias = module_layers_11_feed_forward1_linear2_bias_to_fp16, weight = module_layers_11_feed_forward1_linear2_weight_to_fp16_quantized, x = input_597_cast_fp16)[name = string("linear_101_cast_fp16")];
            fp16 var_2296_to_fp16 = const()[name = string("op_2296_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2297_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_2296_to_fp16)[name = string("op_2297_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_603_cast_fp16 = add(x = input_591_cast_fp16, y = var_2297_cast_fp16)[name = string("input_603_cast_fp16")];
            tensor<int32, [1]> query_23_axes_0 = const()[name = string("query_23_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_11_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159448064)))];
            tensor<fp16, [1024]> module_layers_11_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159450176)))];
            tensor<fp16, [1, 188, 1024]> query_23_cast_fp16 = layer_norm(axes = query_23_axes_0, beta = module_layers_11_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_self_att_weight_to_fp16, x = input_603_cast_fp16)[name = string("query_23_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159452288))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(159976640))))[name = string("module_layers_11_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_11_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_11_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160042240)))];
            tensor<fp16, [1, 188, 1024]> linear_102_cast_fp16 = linear(bias = module_layers_11_self_attn_linear_q_bias_to_fp16, weight = module_layers_11_self_attn_linear_q_weight_to_fp16_quantized, x = query_23_cast_fp16)[name = string("linear_102_cast_fp16")];
            tensor<int32, [4]> var_2314 = const()[name = string("op_2314"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_67_cast_fp16 = reshape(shape = var_2314, x = linear_102_cast_fp16)[name = string("q_67_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160044352))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160568704))))[name = string("module_layers_11_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_11_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_11_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160634304)))];
            tensor<fp16, [1, 188, 1024]> linear_103_cast_fp16 = linear(bias = module_layers_11_self_attn_linear_k_bias_to_fp16, weight = module_layers_11_self_attn_linear_k_weight_to_fp16_quantized, x = query_23_cast_fp16)[name = string("linear_103_cast_fp16")];
            tensor<int32, [4]> var_2319 = const()[name = string("op_2319"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_45_cast_fp16 = reshape(shape = var_2319, x = linear_103_cast_fp16)[name = string("k_45_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160636416))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161160768))))[name = string("module_layers_11_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_11_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_11_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161226368)))];
            tensor<fp16, [1, 188, 1024]> linear_104_cast_fp16 = linear(bias = module_layers_11_self_attn_linear_v_bias_to_fp16, weight = module_layers_11_self_attn_linear_v_weight_to_fp16_quantized, x = query_23_cast_fp16)[name = string("linear_104_cast_fp16")];
            tensor<int32, [4]> var_2324 = const()[name = string("op_2324"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_23_cast_fp16 = reshape(shape = var_2324, x = linear_104_cast_fp16)[name = string("v_23_cast_fp16")];
            tensor<int32, [4]> value_25_perm_0 = const()[name = string("value_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161228480))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161229056))))[name = string("module_layers_11_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2336_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_2336_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_11_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161229184))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161229760))))[name = string("module_layers_11_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2338_cast_fp16 = add(x = q_67_cast_fp16, y = module_layers_11_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_2338_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_23_perm_0 = const()[name = string("q_with_bias_v_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_249_transpose_x_0 = const()[name = string("x_249_transpose_x_0"), val = bool(false)];
            bool x_249_transpose_y_0 = const()[name = string("x_249_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_2340_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161229888))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161421952))))[name = string("op_2340_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_23_cast_fp16 = transpose(perm = q_with_bias_v_23_perm_0, x = var_2338_cast_fp16)[name = string("transpose_339")];
            tensor<fp16, [1, 8, 188, 375]> x_249_cast_fp16 = matmul(transpose_x = x_249_transpose_x_0, transpose_y = x_249_transpose_y_0, x = q_with_bias_v_23_cast_fp16, y = op_2340_to_fp16_quantized)[name = string("x_249_cast_fp16")];
            tensor<int32, [8]> x_251_pad_0 = const()[name = string("x_251_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_251_mode_0 = const()[name = string("x_251_mode_0"), val = string("constant")];
            fp16 const_180_to_fp16 = const()[name = string("const_180_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_251_cast_fp16 = pad(constant_val = const_180_to_fp16, mode = x_251_mode_0, pad = x_251_pad_0, x = x_249_cast_fp16)[name = string("x_251_cast_fp16")];
            tensor<int32, [4]> var_2348 = const()[name = string("op_2348"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_253_cast_fp16 = reshape(shape = var_2348, x = x_251_cast_fp16)[name = string("x_253_cast_fp16")];
            tensor<int32, [4]> var_2352_begin_0 = const()[name = string("op_2352_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2352_end_0 = const()[name = string("op_2352_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2352_end_mask_0 = const()[name = string("op_2352_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2352_cast_fp16 = slice_by_index(begin = var_2352_begin_0, end = var_2352_end_0, end_mask = var_2352_end_mask_0, x = x_253_cast_fp16)[name = string("op_2352_cast_fp16")];
            tensor<int32, [4]> var_2353 = const()[name = string("op_2353"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_45_cast_fp16 = reshape(shape = var_2353, x = var_2352_cast_fp16)[name = string("matrix_bd_45_cast_fp16")];
            bool matrix_ac_23_transpose_x_0 = const()[name = string("matrix_ac_23_transpose_x_0"), val = bool(false)];
            bool matrix_ac_23_transpose_y_0 = const()[name = string("matrix_ac_23_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_150_perm_0 = const()[name = string("transpose_150_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_151_perm_0 = const()[name = string("transpose_151_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_151 = transpose(perm = transpose_151_perm_0, x = k_45_cast_fp16)[name = string("transpose_337")];
            tensor<fp16, [1, 8, 188, 128]> transpose_150 = transpose(perm = transpose_150_perm_0, x = var_2336_cast_fp16)[name = string("transpose_338")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_23_cast_fp16 = matmul(transpose_x = matrix_ac_23_transpose_x_0, transpose_y = matrix_ac_23_transpose_y_0, x = transpose_150, y = transpose_151)[name = string("matrix_ac_23_cast_fp16")];
            tensor<int32, [4]> matrix_bd_47_begin_0 = const()[name = string("matrix_bd_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_47_end_0 = const()[name = string("matrix_bd_47_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_47_end_mask_0 = const()[name = string("matrix_bd_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_47_cast_fp16 = slice_by_index(begin = matrix_bd_47_begin_0, end = matrix_bd_47_end_0, end_mask = matrix_bd_47_end_mask_0, x = matrix_bd_45_cast_fp16)[name = string("matrix_bd_47_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2362_cast_fp16 = add(x = matrix_ac_23_cast_fp16, y = matrix_bd_47_cast_fp16)[name = string("op_2362_cast_fp16")];
            fp16 _inversed_scores_45_y_0_to_fp16 = const()[name = string("_inversed_scores_45_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_45_cast_fp16 = mul(x = var_2362_cast_fp16, y = _inversed_scores_45_y_0_to_fp16)[name = string("_inversed_scores_45_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_47_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_45_cast_fp16, cond = mask_11)[name = string("scores_47_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2368_cast_fp16 = softmax(axis = var_23, x = scores_47_cast_fp16)[name = string("op_2368_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_605_cast_fp16 = select(a = var_11_to_fp16, b = var_2368_cast_fp16, cond = mask_11)[name = string("input_605_cast_fp16")];
            bool x_255_transpose_x_0 = const()[name = string("x_255_transpose_x_0"), val = bool(false)];
            bool x_255_transpose_y_0 = const()[name = string("x_255_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_25_cast_fp16 = transpose(perm = value_25_perm_0, x = v_23_cast_fp16)[name = string("transpose_336")];
            tensor<fp16, [1, 8, 188, 128]> x_255_cast_fp16 = matmul(transpose_x = x_255_transpose_x_0, transpose_y = x_255_transpose_y_0, x = input_605_cast_fp16, y = value_25_cast_fp16)[name = string("x_255_cast_fp16")];
            tensor<int32, [4]> var_2372_perm_0 = const()[name = string("op_2372_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2373 = const()[name = string("op_2373"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2372_cast_fp16 = transpose(perm = var_2372_perm_0, x = x_255_cast_fp16)[name = string("transpose_335")];
            tensor<fp16, [1, 188, 1024]> input_607_cast_fp16 = reshape(shape = var_2373, x = var_2372_cast_fp16)[name = string("input_607_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_11_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161425024))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161949376))))[name = string("module_layers_11_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_11_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_11_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162014976)))];
            tensor<fp16, [1, 188, 1024]> linear_106_cast_fp16 = linear(bias = module_layers_11_self_attn_linear_out_bias_to_fp16, weight = module_layers_11_self_attn_linear_out_weight_to_fp16_quantized, x = input_607_cast_fp16)[name = string("linear_106_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_611_cast_fp16 = add(x = input_603_cast_fp16, y = linear_106_cast_fp16)[name = string("input_611_cast_fp16")];
            tensor<int32, [1]> x_259_axes_0 = const()[name = string("x_259_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_11_norm_conv_weight_to_fp16 = const()[name = string("module_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162017088)))];
            tensor<fp16, [1024]> module_layers_11_norm_conv_bias_to_fp16 = const()[name = string("module_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162019200)))];
            tensor<fp16, [1, 188, 1024]> x_259_cast_fp16 = layer_norm(axes = x_259_axes_0, beta = module_layers_11_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_conv_weight_to_fp16, x = input_611_cast_fp16)[name = string("x_259_cast_fp16")];
            tensor<int32, [3]> input_613_perm_0 = const()[name = string("input_613_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_615_pad_type_0 = const()[name = string("input_615_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_615_strides_0 = const()[name = string("input_615_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_615_pad_0 = const()[name = string("input_615_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_615_dilations_0 = const()[name = string("input_615_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_615_groups_0 = const()[name = string("input_615_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162021312))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163069952))))[name = string("module_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_11_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_11_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163201088)))];
            tensor<fp16, [1, 1024, 188]> input_613_cast_fp16 = transpose(perm = input_613_perm_0, x = x_259_cast_fp16)[name = string("transpose_334")];
            tensor<fp16, [1, 2048, 188]> input_615_cast_fp16 = conv(bias = module_layers_11_conv_pointwise_conv1_bias_to_fp16, dilations = input_615_dilations_0, groups = input_615_groups_0, pad = input_615_pad_0, pad_type = input_615_pad_type_0, strides = input_615_strides_0, weight = module_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_613_cast_fp16)[name = string("input_615_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, 188]> x_261_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_615_cast_fp16)[name = string("x_261_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_617_cast_fp16 = select(a = var_11_to_fp16, b = x_261_cast_fp16, cond = var_483)[name = string("input_617_cast_fp16")];
            tensor<int32, [6]> input_619_pad_0 = const()[name = string("input_619_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_619_mode_0 = const()[name = string("input_619_mode_0"), val = string("constant")];
            fp16 const_183_to_fp16 = const()[name = string("const_183_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_619_cast_fp16 = pad(constant_val = const_183_to_fp16, mode = input_619_mode_0, pad = input_619_pad_0, x = input_617_cast_fp16)[name = string("input_619_cast_fp16")];
            string input_621_pad_type_0 = const()[name = string("input_621_pad_type_0"), val = string("valid")];
            int32 input_621_groups_0 = const()[name = string("input_621_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_621_strides_0 = const()[name = string("input_621_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_621_pad_0 = const()[name = string("input_621_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_621_dilations_0 = const()[name = string("input_621_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_406_to_fp16 = const()[name = string("const_406_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163205248)))];
            tensor<fp16, [1024]> const_407_to_fp16 = const()[name = string("const_407_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163223744)))];
            tensor<fp16, [1, 1024, 188]> input_623_cast_fp16 = conv(bias = const_407_to_fp16, dilations = input_621_dilations_0, groups = input_621_groups_0, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = input_621_strides_0, weight = const_406_to_fp16, x = input_619_cast_fp16)[name = string("input_623_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_625_cast_fp16 = silu(x = input_623_cast_fp16)[name = string("input_625_cast_fp16")];
            string x_263_pad_type_0 = const()[name = string("x_263_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_263_strides_0 = const()[name = string("x_263_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_263_pad_0 = const()[name = string("x_263_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_263_dilations_0 = const()[name = string("x_263_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_263_groups_0 = const()[name = string("x_263_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163225856))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163750208))))[name = string("module_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_11_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_11_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163815808)))];
            tensor<fp16, [1, 1024, 188]> x_263_cast_fp16 = conv(bias = module_layers_11_conv_pointwise_conv2_bias_to_fp16, dilations = x_263_dilations_0, groups = x_263_groups_0, pad = x_263_pad_0, pad_type = x_263_pad_type_0, strides = x_263_strides_0, weight = module_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_625_cast_fp16)[name = string("x_263_cast_fp16")];
            tensor<int32, [3]> input_627_perm_0 = const()[name = string("input_627_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_627_cast_fp16 = transpose(perm = input_627_perm_0, x = x_263_cast_fp16)[name = string("transpose_333")];
            tensor<fp16, [1, 188, 1024]> input_629_cast_fp16 = add(x = input_611_cast_fp16, y = input_627_cast_fp16)[name = string("input_629_cast_fp16")];
            tensor<int32, [1]> input_631_axes_0 = const()[name = string("input_631_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163817920)))];
            tensor<fp16, [1024]> module_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163820032)))];
            tensor<fp16, [1, 188, 1024]> input_631_cast_fp16 = layer_norm(axes = input_631_axes_0, beta = module_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_feed_forward2_weight_to_fp16, x = input_629_cast_fp16)[name = string("input_631_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_11_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163822144))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165919360))))[name = string("module_layers_11_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_11_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_11_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166181568)))];
            tensor<fp16, [1, 188, 4096]> linear_107_cast_fp16 = linear(bias = module_layers_11_feed_forward2_linear1_bias_to_fp16, weight = module_layers_11_feed_forward2_linear1_weight_to_fp16_quantized, x = input_631_cast_fp16)[name = string("linear_107_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_635_cast_fp16 = silu(x = linear_107_cast_fp16)[name = string("input_635_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_11_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166189824))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168287040))))[name = string("module_layers_11_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_11_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_11_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168549248)))];
            tensor<fp16, [1, 188, 1024]> linear_108_cast_fp16 = linear(bias = module_layers_11_feed_forward2_linear2_bias_to_fp16, weight = module_layers_11_feed_forward2_linear2_weight_to_fp16_quantized, x = input_635_cast_fp16)[name = string("linear_108_cast_fp16")];
            fp16 var_2439_to_fp16 = const()[name = string("op_2439_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2440_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_2439_to_fp16)[name = string("op_2440_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_641_cast_fp16 = add(x = input_629_cast_fp16, y = var_2440_cast_fp16)[name = string("input_641_cast_fp16")];
            tensor<int32, [1]> input_643_axes_0 = const()[name = string("input_643_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_11_norm_out_weight_to_fp16 = const()[name = string("module_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168551360)))];
            tensor<fp16, [1024]> module_layers_11_norm_out_bias_to_fp16 = const()[name = string("module_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168553472)))];
            tensor<fp16, [1, 188, 1024]> input_643_cast_fp16 = layer_norm(axes = input_643_axes_0, beta = module_layers_11_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_11_norm_out_weight_to_fp16, x = input_641_cast_fp16)[name = string("input_643_cast_fp16")];
            tensor<int32, [1]> input_645_axes_0 = const()[name = string("input_645_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168555584)))];
            tensor<fp16, [1024]> module_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168557696)))];
            tensor<fp16, [1, 188, 1024]> input_645_cast_fp16 = layer_norm(axes = input_645_axes_0, beta = module_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward1_weight_to_fp16, x = input_643_cast_fp16)[name = string("input_645_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_12_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(168559808))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170657024))))[name = string("module_layers_12_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_12_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_12_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170919232)))];
            tensor<fp16, [1, 188, 4096]> linear_109_cast_fp16 = linear(bias = module_layers_12_feed_forward1_linear1_bias_to_fp16, weight = module_layers_12_feed_forward1_linear1_weight_to_fp16_quantized, x = input_645_cast_fp16)[name = string("linear_109_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_649_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("input_649_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_12_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170927488))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173024704))))[name = string("module_layers_12_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_12_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_12_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173286912)))];
            tensor<fp16, [1, 188, 1024]> linear_110_cast_fp16 = linear(bias = module_layers_12_feed_forward1_linear2_bias_to_fp16, weight = module_layers_12_feed_forward1_linear2_weight_to_fp16_quantized, x = input_649_cast_fp16)[name = string("linear_110_cast_fp16")];
            fp16 var_2470_to_fp16 = const()[name = string("op_2470_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2471_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_2470_to_fp16)[name = string("op_2471_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_655_cast_fp16 = add(x = input_643_cast_fp16, y = var_2471_cast_fp16)[name = string("input_655_cast_fp16")];
            tensor<int32, [1]> query_25_axes_0 = const()[name = string("query_25_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_12_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173289024)))];
            tensor<fp16, [1024]> module_layers_12_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173291136)))];
            tensor<fp16, [1, 188, 1024]> query_25_cast_fp16 = layer_norm(axes = query_25_axes_0, beta = module_layers_12_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_self_att_weight_to_fp16, x = input_655_cast_fp16)[name = string("query_25_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173293248))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173817600))))[name = string("module_layers_12_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_12_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_12_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173883200)))];
            tensor<fp16, [1, 188, 1024]> linear_111_cast_fp16 = linear(bias = module_layers_12_self_attn_linear_q_bias_to_fp16, weight = module_layers_12_self_attn_linear_q_weight_to_fp16_quantized, x = query_25_cast_fp16)[name = string("linear_111_cast_fp16")];
            tensor<int32, [4]> var_2488 = const()[name = string("op_2488"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_73_cast_fp16 = reshape(shape = var_2488, x = linear_111_cast_fp16)[name = string("q_73_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(173885312))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174409664))))[name = string("module_layers_12_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_12_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_12_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174475264)))];
            tensor<fp16, [1, 188, 1024]> linear_112_cast_fp16 = linear(bias = module_layers_12_self_attn_linear_k_bias_to_fp16, weight = module_layers_12_self_attn_linear_k_weight_to_fp16_quantized, x = query_25_cast_fp16)[name = string("linear_112_cast_fp16")];
            tensor<int32, [4]> var_2493 = const()[name = string("op_2493"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_49_cast_fp16 = reshape(shape = var_2493, x = linear_112_cast_fp16)[name = string("k_49_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174477376))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175001728))))[name = string("module_layers_12_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_12_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_12_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175067328)))];
            tensor<fp16, [1, 188, 1024]> linear_113_cast_fp16 = linear(bias = module_layers_12_self_attn_linear_v_bias_to_fp16, weight = module_layers_12_self_attn_linear_v_weight_to_fp16_quantized, x = query_25_cast_fp16)[name = string("linear_113_cast_fp16")];
            tensor<int32, [4]> var_2498 = const()[name = string("op_2498"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_25_cast_fp16 = reshape(shape = var_2498, x = linear_113_cast_fp16)[name = string("v_25_cast_fp16")];
            tensor<int32, [4]> value_27_perm_0 = const()[name = string("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175069440))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175070016))))[name = string("module_layers_12_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2510_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_2510_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_12_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175070144))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175070720))))[name = string("module_layers_12_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2512_cast_fp16 = add(x = q_73_cast_fp16, y = module_layers_12_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_2512_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_25_perm_0 = const()[name = string("q_with_bias_v_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_271_transpose_x_0 = const()[name = string("x_271_transpose_x_0"), val = bool(false)];
            bool x_271_transpose_y_0 = const()[name = string("x_271_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_2514_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175070848))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175262912))))[name = string("op_2514_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_25_cast_fp16 = transpose(perm = q_with_bias_v_25_perm_0, x = var_2512_cast_fp16)[name = string("transpose_332")];
            tensor<fp16, [1, 8, 188, 375]> x_271_cast_fp16 = matmul(transpose_x = x_271_transpose_x_0, transpose_y = x_271_transpose_y_0, x = q_with_bias_v_25_cast_fp16, y = op_2514_to_fp16_quantized)[name = string("x_271_cast_fp16")];
            tensor<int32, [8]> x_273_pad_0 = const()[name = string("x_273_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_273_mode_0 = const()[name = string("x_273_mode_0"), val = string("constant")];
            fp16 const_190_to_fp16 = const()[name = string("const_190_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_273_cast_fp16 = pad(constant_val = const_190_to_fp16, mode = x_273_mode_0, pad = x_273_pad_0, x = x_271_cast_fp16)[name = string("x_273_cast_fp16")];
            tensor<int32, [4]> var_2522 = const()[name = string("op_2522"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_275_cast_fp16 = reshape(shape = var_2522, x = x_273_cast_fp16)[name = string("x_275_cast_fp16")];
            tensor<int32, [4]> var_2526_begin_0 = const()[name = string("op_2526_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2526_end_0 = const()[name = string("op_2526_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2526_end_mask_0 = const()[name = string("op_2526_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2526_cast_fp16 = slice_by_index(begin = var_2526_begin_0, end = var_2526_end_0, end_mask = var_2526_end_mask_0, x = x_275_cast_fp16)[name = string("op_2526_cast_fp16")];
            tensor<int32, [4]> var_2527 = const()[name = string("op_2527"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_49_cast_fp16 = reshape(shape = var_2527, x = var_2526_cast_fp16)[name = string("matrix_bd_49_cast_fp16")];
            bool matrix_ac_25_transpose_x_0 = const()[name = string("matrix_ac_25_transpose_x_0"), val = bool(false)];
            bool matrix_ac_25_transpose_y_0 = const()[name = string("matrix_ac_25_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_152_perm_0 = const()[name = string("transpose_152_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_153_perm_0 = const()[name = string("transpose_153_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_153 = transpose(perm = transpose_153_perm_0, x = k_49_cast_fp16)[name = string("transpose_330")];
            tensor<fp16, [1, 8, 188, 128]> transpose_152 = transpose(perm = transpose_152_perm_0, x = var_2510_cast_fp16)[name = string("transpose_331")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_25_cast_fp16 = matmul(transpose_x = matrix_ac_25_transpose_x_0, transpose_y = matrix_ac_25_transpose_y_0, x = transpose_152, y = transpose_153)[name = string("matrix_ac_25_cast_fp16")];
            tensor<int32, [4]> matrix_bd_51_begin_0 = const()[name = string("matrix_bd_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_51_end_0 = const()[name = string("matrix_bd_51_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_51_end_mask_0 = const()[name = string("matrix_bd_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_51_cast_fp16 = slice_by_index(begin = matrix_bd_51_begin_0, end = matrix_bd_51_end_0, end_mask = matrix_bd_51_end_mask_0, x = matrix_bd_49_cast_fp16)[name = string("matrix_bd_51_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2536_cast_fp16 = add(x = matrix_ac_25_cast_fp16, y = matrix_bd_51_cast_fp16)[name = string("op_2536_cast_fp16")];
            fp16 _inversed_scores_49_y_0_to_fp16 = const()[name = string("_inversed_scores_49_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_49_cast_fp16 = mul(x = var_2536_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = string("_inversed_scores_49_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_51_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_49_cast_fp16, cond = mask_11)[name = string("scores_51_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2542_cast_fp16 = softmax(axis = var_23, x = scores_51_cast_fp16)[name = string("op_2542_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_657_cast_fp16 = select(a = var_11_to_fp16, b = var_2542_cast_fp16, cond = mask_11)[name = string("input_657_cast_fp16")];
            bool x_277_transpose_x_0 = const()[name = string("x_277_transpose_x_0"), val = bool(false)];
            bool x_277_transpose_y_0 = const()[name = string("x_277_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = v_25_cast_fp16)[name = string("transpose_329")];
            tensor<fp16, [1, 8, 188, 128]> x_277_cast_fp16 = matmul(transpose_x = x_277_transpose_x_0, transpose_y = x_277_transpose_y_0, x = input_657_cast_fp16, y = value_27_cast_fp16)[name = string("x_277_cast_fp16")];
            tensor<int32, [4]> var_2546_perm_0 = const()[name = string("op_2546_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2547 = const()[name = string("op_2547"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2546_cast_fp16 = transpose(perm = var_2546_perm_0, x = x_277_cast_fp16)[name = string("transpose_328")];
            tensor<fp16, [1, 188, 1024]> input_659_cast_fp16 = reshape(shape = var_2547, x = var_2546_cast_fp16)[name = string("input_659_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_12_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175265984))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175790336))))[name = string("module_layers_12_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_12_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_12_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175855936)))];
            tensor<fp16, [1, 188, 1024]> linear_115_cast_fp16 = linear(bias = module_layers_12_self_attn_linear_out_bias_to_fp16, weight = module_layers_12_self_attn_linear_out_weight_to_fp16_quantized, x = input_659_cast_fp16)[name = string("linear_115_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_663_cast_fp16 = add(x = input_655_cast_fp16, y = linear_115_cast_fp16)[name = string("input_663_cast_fp16")];
            tensor<int32, [1]> x_281_axes_0 = const()[name = string("x_281_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_12_norm_conv_weight_to_fp16 = const()[name = string("module_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175858048)))];
            tensor<fp16, [1024]> module_layers_12_norm_conv_bias_to_fp16 = const()[name = string("module_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175860160)))];
            tensor<fp16, [1, 188, 1024]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = module_layers_12_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_conv_weight_to_fp16, x = input_663_cast_fp16)[name = string("x_281_cast_fp16")];
            tensor<int32, [3]> input_665_perm_0 = const()[name = string("input_665_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_667_pad_type_0 = const()[name = string("input_667_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_667_strides_0 = const()[name = string("input_667_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_667_pad_0 = const()[name = string("input_667_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_667_dilations_0 = const()[name = string("input_667_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_667_groups_0 = const()[name = string("input_667_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(175862272))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176910912))))[name = string("module_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_12_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_12_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177042048)))];
            tensor<fp16, [1, 1024, 188]> input_665_cast_fp16 = transpose(perm = input_665_perm_0, x = x_281_cast_fp16)[name = string("transpose_327")];
            tensor<fp16, [1, 2048, 188]> input_667_cast_fp16 = conv(bias = module_layers_12_conv_pointwise_conv1_bias_to_fp16, dilations = input_667_dilations_0, groups = input_667_groups_0, pad = input_667_pad_0, pad_type = input_667_pad_type_0, strides = input_667_strides_0, weight = module_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_665_cast_fp16)[name = string("input_667_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, 188]> x_283_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_667_cast_fp16)[name = string("x_283_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_669_cast_fp16 = select(a = var_11_to_fp16, b = x_283_cast_fp16, cond = var_483)[name = string("input_669_cast_fp16")];
            tensor<int32, [6]> input_671_pad_0 = const()[name = string("input_671_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_671_mode_0 = const()[name = string("input_671_mode_0"), val = string("constant")];
            fp16 const_193_to_fp16 = const()[name = string("const_193_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_671_cast_fp16 = pad(constant_val = const_193_to_fp16, mode = input_671_mode_0, pad = input_671_pad_0, x = input_669_cast_fp16)[name = string("input_671_cast_fp16")];
            string input_673_pad_type_0 = const()[name = string("input_673_pad_type_0"), val = string("valid")];
            int32 input_673_groups_0 = const()[name = string("input_673_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_673_strides_0 = const()[name = string("input_673_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_673_pad_0 = const()[name = string("input_673_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_673_dilations_0 = const()[name = string("input_673_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_408_to_fp16 = const()[name = string("const_408_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177046208)))];
            tensor<fp16, [1024]> const_409_to_fp16 = const()[name = string("const_409_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177064704)))];
            tensor<fp16, [1, 1024, 188]> input_675_cast_fp16 = conv(bias = const_409_to_fp16, dilations = input_673_dilations_0, groups = input_673_groups_0, pad = input_673_pad_0, pad_type = input_673_pad_type_0, strides = input_673_strides_0, weight = const_408_to_fp16, x = input_671_cast_fp16)[name = string("input_675_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_677_cast_fp16 = silu(x = input_675_cast_fp16)[name = string("input_677_cast_fp16")];
            string x_285_pad_type_0 = const()[name = string("x_285_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_285_strides_0 = const()[name = string("x_285_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_285_pad_0 = const()[name = string("x_285_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_285_dilations_0 = const()[name = string("x_285_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_285_groups_0 = const()[name = string("x_285_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177066816))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177591168))))[name = string("module_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_12_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_12_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177656768)))];
            tensor<fp16, [1, 1024, 188]> x_285_cast_fp16 = conv(bias = module_layers_12_conv_pointwise_conv2_bias_to_fp16, dilations = x_285_dilations_0, groups = x_285_groups_0, pad = x_285_pad_0, pad_type = x_285_pad_type_0, strides = x_285_strides_0, weight = module_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_677_cast_fp16)[name = string("x_285_cast_fp16")];
            tensor<int32, [3]> input_679_perm_0 = const()[name = string("input_679_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_679_cast_fp16 = transpose(perm = input_679_perm_0, x = x_285_cast_fp16)[name = string("transpose_326")];
            tensor<fp16, [1, 188, 1024]> input_681_cast_fp16 = add(x = input_663_cast_fp16, y = input_679_cast_fp16)[name = string("input_681_cast_fp16")];
            tensor<int32, [1]> input_683_axes_0 = const()[name = string("input_683_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177658880)))];
            tensor<fp16, [1024]> module_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177660992)))];
            tensor<fp16, [1, 188, 1024]> input_683_cast_fp16 = layer_norm(axes = input_683_axes_0, beta = module_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_feed_forward2_weight_to_fp16, x = input_681_cast_fp16)[name = string("input_683_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_12_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177663104))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179760320))))[name = string("module_layers_12_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_12_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_12_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180022528)))];
            tensor<fp16, [1, 188, 4096]> linear_116_cast_fp16 = linear(bias = module_layers_12_feed_forward2_linear1_bias_to_fp16, weight = module_layers_12_feed_forward2_linear1_weight_to_fp16_quantized, x = input_683_cast_fp16)[name = string("linear_116_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_687_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("input_687_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_12_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(180030784))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182128000))))[name = string("module_layers_12_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_12_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_12_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182390208)))];
            tensor<fp16, [1, 188, 1024]> linear_117_cast_fp16 = linear(bias = module_layers_12_feed_forward2_linear2_bias_to_fp16, weight = module_layers_12_feed_forward2_linear2_weight_to_fp16_quantized, x = input_687_cast_fp16)[name = string("linear_117_cast_fp16")];
            fp16 var_2613_to_fp16 = const()[name = string("op_2613_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2614_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_2613_to_fp16)[name = string("op_2614_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_693_cast_fp16 = add(x = input_681_cast_fp16, y = var_2614_cast_fp16)[name = string("input_693_cast_fp16")];
            tensor<int32, [1]> input_695_axes_0 = const()[name = string("input_695_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_12_norm_out_weight_to_fp16 = const()[name = string("module_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182392320)))];
            tensor<fp16, [1024]> module_layers_12_norm_out_bias_to_fp16 = const()[name = string("module_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182394432)))];
            tensor<fp16, [1, 188, 1024]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = module_layers_12_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_12_norm_out_weight_to_fp16, x = input_693_cast_fp16)[name = string("input_695_cast_fp16")];
            tensor<int32, [1]> input_697_axes_0 = const()[name = string("input_697_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182396544)))];
            tensor<fp16, [1024]> module_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182398656)))];
            tensor<fp16, [1, 188, 1024]> input_697_cast_fp16 = layer_norm(axes = input_697_axes_0, beta = module_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward1_weight_to_fp16, x = input_695_cast_fp16)[name = string("input_697_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_13_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182400768))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184497984))))[name = string("module_layers_13_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_13_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_13_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184760192)))];
            tensor<fp16, [1, 188, 4096]> linear_118_cast_fp16 = linear(bias = module_layers_13_feed_forward1_linear1_bias_to_fp16, weight = module_layers_13_feed_forward1_linear1_weight_to_fp16_quantized, x = input_697_cast_fp16)[name = string("linear_118_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_701_cast_fp16 = silu(x = linear_118_cast_fp16)[name = string("input_701_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_13_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184768448))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186865664))))[name = string("module_layers_13_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_13_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_13_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187127872)))];
            tensor<fp16, [1, 188, 1024]> linear_119_cast_fp16 = linear(bias = module_layers_13_feed_forward1_linear2_bias_to_fp16, weight = module_layers_13_feed_forward1_linear2_weight_to_fp16_quantized, x = input_701_cast_fp16)[name = string("linear_119_cast_fp16")];
            fp16 var_2644_to_fp16 = const()[name = string("op_2644_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2645_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_2644_to_fp16)[name = string("op_2645_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_707_cast_fp16 = add(x = input_695_cast_fp16, y = var_2645_cast_fp16)[name = string("input_707_cast_fp16")];
            tensor<int32, [1]> query_27_axes_0 = const()[name = string("query_27_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_13_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187129984)))];
            tensor<fp16, [1024]> module_layers_13_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187132096)))];
            tensor<fp16, [1, 188, 1024]> query_27_cast_fp16 = layer_norm(axes = query_27_axes_0, beta = module_layers_13_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_self_att_weight_to_fp16, x = input_707_cast_fp16)[name = string("query_27_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187134208))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187658560))))[name = string("module_layers_13_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_13_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_13_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187724160)))];
            tensor<fp16, [1, 188, 1024]> linear_120_cast_fp16 = linear(bias = module_layers_13_self_attn_linear_q_bias_to_fp16, weight = module_layers_13_self_attn_linear_q_weight_to_fp16_quantized, x = query_27_cast_fp16)[name = string("linear_120_cast_fp16")];
            tensor<int32, [4]> var_2662 = const()[name = string("op_2662"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_79_cast_fp16 = reshape(shape = var_2662, x = linear_120_cast_fp16)[name = string("q_79_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(187726272))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188250624))))[name = string("module_layers_13_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_13_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_13_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188316224)))];
            tensor<fp16, [1, 188, 1024]> linear_121_cast_fp16 = linear(bias = module_layers_13_self_attn_linear_k_bias_to_fp16, weight = module_layers_13_self_attn_linear_k_weight_to_fp16_quantized, x = query_27_cast_fp16)[name = string("linear_121_cast_fp16")];
            tensor<int32, [4]> var_2667 = const()[name = string("op_2667"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_53_cast_fp16 = reshape(shape = var_2667, x = linear_121_cast_fp16)[name = string("k_53_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188318336))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188842688))))[name = string("module_layers_13_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_13_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_13_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188908288)))];
            tensor<fp16, [1, 188, 1024]> linear_122_cast_fp16 = linear(bias = module_layers_13_self_attn_linear_v_bias_to_fp16, weight = module_layers_13_self_attn_linear_v_weight_to_fp16_quantized, x = query_27_cast_fp16)[name = string("linear_122_cast_fp16")];
            tensor<int32, [4]> var_2672 = const()[name = string("op_2672"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_27_cast_fp16 = reshape(shape = var_2672, x = linear_122_cast_fp16)[name = string("v_27_cast_fp16")];
            tensor<int32, [4]> value_29_perm_0 = const()[name = string("value_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188910400))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188910976))))[name = string("module_layers_13_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2684_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_2684_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_13_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188911104))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188911680))))[name = string("module_layers_13_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2686_cast_fp16 = add(x = q_79_cast_fp16, y = module_layers_13_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_2686_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_27_perm_0 = const()[name = string("q_with_bias_v_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_293_transpose_x_0 = const()[name = string("x_293_transpose_x_0"), val = bool(false)];
            bool x_293_transpose_y_0 = const()[name = string("x_293_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_2688_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188911808))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189103872))))[name = string("op_2688_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_27_cast_fp16 = transpose(perm = q_with_bias_v_27_perm_0, x = var_2686_cast_fp16)[name = string("transpose_325")];
            tensor<fp16, [1, 8, 188, 375]> x_293_cast_fp16 = matmul(transpose_x = x_293_transpose_x_0, transpose_y = x_293_transpose_y_0, x = q_with_bias_v_27_cast_fp16, y = op_2688_to_fp16_quantized)[name = string("x_293_cast_fp16")];
            tensor<int32, [8]> x_295_pad_0 = const()[name = string("x_295_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_295_mode_0 = const()[name = string("x_295_mode_0"), val = string("constant")];
            fp16 const_200_to_fp16 = const()[name = string("const_200_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_295_cast_fp16 = pad(constant_val = const_200_to_fp16, mode = x_295_mode_0, pad = x_295_pad_0, x = x_293_cast_fp16)[name = string("x_295_cast_fp16")];
            tensor<int32, [4]> var_2696 = const()[name = string("op_2696"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_297_cast_fp16 = reshape(shape = var_2696, x = x_295_cast_fp16)[name = string("x_297_cast_fp16")];
            tensor<int32, [4]> var_2700_begin_0 = const()[name = string("op_2700_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2700_end_0 = const()[name = string("op_2700_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2700_end_mask_0 = const()[name = string("op_2700_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2700_cast_fp16 = slice_by_index(begin = var_2700_begin_0, end = var_2700_end_0, end_mask = var_2700_end_mask_0, x = x_297_cast_fp16)[name = string("op_2700_cast_fp16")];
            tensor<int32, [4]> var_2701 = const()[name = string("op_2701"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_53_cast_fp16 = reshape(shape = var_2701, x = var_2700_cast_fp16)[name = string("matrix_bd_53_cast_fp16")];
            bool matrix_ac_27_transpose_x_0 = const()[name = string("matrix_ac_27_transpose_x_0"), val = bool(false)];
            bool matrix_ac_27_transpose_y_0 = const()[name = string("matrix_ac_27_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_154_perm_0 = const()[name = string("transpose_154_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_155_perm_0 = const()[name = string("transpose_155_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_155 = transpose(perm = transpose_155_perm_0, x = k_53_cast_fp16)[name = string("transpose_323")];
            tensor<fp16, [1, 8, 188, 128]> transpose_154 = transpose(perm = transpose_154_perm_0, x = var_2684_cast_fp16)[name = string("transpose_324")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_27_cast_fp16 = matmul(transpose_x = matrix_ac_27_transpose_x_0, transpose_y = matrix_ac_27_transpose_y_0, x = transpose_154, y = transpose_155)[name = string("matrix_ac_27_cast_fp16")];
            tensor<int32, [4]> matrix_bd_55_begin_0 = const()[name = string("matrix_bd_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_55_end_0 = const()[name = string("matrix_bd_55_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_55_end_mask_0 = const()[name = string("matrix_bd_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_55_cast_fp16 = slice_by_index(begin = matrix_bd_55_begin_0, end = matrix_bd_55_end_0, end_mask = matrix_bd_55_end_mask_0, x = matrix_bd_53_cast_fp16)[name = string("matrix_bd_55_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2710_cast_fp16 = add(x = matrix_ac_27_cast_fp16, y = matrix_bd_55_cast_fp16)[name = string("op_2710_cast_fp16")];
            fp16 _inversed_scores_53_y_0_to_fp16 = const()[name = string("_inversed_scores_53_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_53_cast_fp16 = mul(x = var_2710_cast_fp16, y = _inversed_scores_53_y_0_to_fp16)[name = string("_inversed_scores_53_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_55_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_53_cast_fp16, cond = mask_11)[name = string("scores_55_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2716_cast_fp16 = softmax(axis = var_23, x = scores_55_cast_fp16)[name = string("op_2716_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_709_cast_fp16 = select(a = var_11_to_fp16, b = var_2716_cast_fp16, cond = mask_11)[name = string("input_709_cast_fp16")];
            bool x_299_transpose_x_0 = const()[name = string("x_299_transpose_x_0"), val = bool(false)];
            bool x_299_transpose_y_0 = const()[name = string("x_299_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_29_cast_fp16 = transpose(perm = value_29_perm_0, x = v_27_cast_fp16)[name = string("transpose_322")];
            tensor<fp16, [1, 8, 188, 128]> x_299_cast_fp16 = matmul(transpose_x = x_299_transpose_x_0, transpose_y = x_299_transpose_y_0, x = input_709_cast_fp16, y = value_29_cast_fp16)[name = string("x_299_cast_fp16")];
            tensor<int32, [4]> var_2720_perm_0 = const()[name = string("op_2720_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2721 = const()[name = string("op_2721"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2720_cast_fp16 = transpose(perm = var_2720_perm_0, x = x_299_cast_fp16)[name = string("transpose_321")];
            tensor<fp16, [1, 188, 1024]> input_711_cast_fp16 = reshape(shape = var_2721, x = var_2720_cast_fp16)[name = string("input_711_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_13_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189106944))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189631296))))[name = string("module_layers_13_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_13_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_13_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189696896)))];
            tensor<fp16, [1, 188, 1024]> linear_124_cast_fp16 = linear(bias = module_layers_13_self_attn_linear_out_bias_to_fp16, weight = module_layers_13_self_attn_linear_out_weight_to_fp16_quantized, x = input_711_cast_fp16)[name = string("linear_124_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_715_cast_fp16 = add(x = input_707_cast_fp16, y = linear_124_cast_fp16)[name = string("input_715_cast_fp16")];
            tensor<int32, [1]> x_303_axes_0 = const()[name = string("x_303_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_13_norm_conv_weight_to_fp16 = const()[name = string("module_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189699008)))];
            tensor<fp16, [1024]> module_layers_13_norm_conv_bias_to_fp16 = const()[name = string("module_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189701120)))];
            tensor<fp16, [1, 188, 1024]> x_303_cast_fp16 = layer_norm(axes = x_303_axes_0, beta = module_layers_13_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_conv_weight_to_fp16, x = input_715_cast_fp16)[name = string("x_303_cast_fp16")];
            tensor<int32, [3]> input_717_perm_0 = const()[name = string("input_717_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_719_pad_type_0 = const()[name = string("input_719_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_719_strides_0 = const()[name = string("input_719_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_719_pad_0 = const()[name = string("input_719_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_719_dilations_0 = const()[name = string("input_719_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_719_groups_0 = const()[name = string("input_719_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189703232))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190751872))))[name = string("module_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_13_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_13_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190883008)))];
            tensor<fp16, [1, 1024, 188]> input_717_cast_fp16 = transpose(perm = input_717_perm_0, x = x_303_cast_fp16)[name = string("transpose_320")];
            tensor<fp16, [1, 2048, 188]> input_719_cast_fp16 = conv(bias = module_layers_13_conv_pointwise_conv1_bias_to_fp16, dilations = input_719_dilations_0, groups = input_719_groups_0, pad = input_719_pad_0, pad_type = input_719_pad_type_0, strides = input_719_strides_0, weight = module_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_717_cast_fp16)[name = string("input_719_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, 188]> x_305_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_719_cast_fp16)[name = string("x_305_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_721_cast_fp16 = select(a = var_11_to_fp16, b = x_305_cast_fp16, cond = var_483)[name = string("input_721_cast_fp16")];
            tensor<int32, [6]> input_723_pad_0 = const()[name = string("input_723_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_723_mode_0 = const()[name = string("input_723_mode_0"), val = string("constant")];
            fp16 const_203_to_fp16 = const()[name = string("const_203_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_723_cast_fp16 = pad(constant_val = const_203_to_fp16, mode = input_723_mode_0, pad = input_723_pad_0, x = input_721_cast_fp16)[name = string("input_723_cast_fp16")];
            string input_725_pad_type_0 = const()[name = string("input_725_pad_type_0"), val = string("valid")];
            int32 input_725_groups_0 = const()[name = string("input_725_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_725_strides_0 = const()[name = string("input_725_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_725_pad_0 = const()[name = string("input_725_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_725_dilations_0 = const()[name = string("input_725_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_410_to_fp16 = const()[name = string("const_410_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190887168)))];
            tensor<fp16, [1024]> const_411_to_fp16 = const()[name = string("const_411_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190905664)))];
            tensor<fp16, [1, 1024, 188]> input_727_cast_fp16 = conv(bias = const_411_to_fp16, dilations = input_725_dilations_0, groups = input_725_groups_0, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = input_725_strides_0, weight = const_410_to_fp16, x = input_723_cast_fp16)[name = string("input_727_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_729_cast_fp16 = silu(x = input_727_cast_fp16)[name = string("input_729_cast_fp16")];
            string x_307_pad_type_0 = const()[name = string("x_307_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_307_strides_0 = const()[name = string("x_307_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_307_pad_0 = const()[name = string("x_307_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_307_dilations_0 = const()[name = string("x_307_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_307_groups_0 = const()[name = string("x_307_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190907776))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191432128))))[name = string("module_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_13_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_13_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191497728)))];
            tensor<fp16, [1, 1024, 188]> x_307_cast_fp16 = conv(bias = module_layers_13_conv_pointwise_conv2_bias_to_fp16, dilations = x_307_dilations_0, groups = x_307_groups_0, pad = x_307_pad_0, pad_type = x_307_pad_type_0, strides = x_307_strides_0, weight = module_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_729_cast_fp16)[name = string("x_307_cast_fp16")];
            tensor<int32, [3]> input_731_perm_0 = const()[name = string("input_731_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_731_cast_fp16 = transpose(perm = input_731_perm_0, x = x_307_cast_fp16)[name = string("transpose_319")];
            tensor<fp16, [1, 188, 1024]> input_733_cast_fp16 = add(x = input_715_cast_fp16, y = input_731_cast_fp16)[name = string("input_733_cast_fp16")];
            tensor<int32, [1]> input_735_axes_0 = const()[name = string("input_735_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191499840)))];
            tensor<fp16, [1024]> module_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191501952)))];
            tensor<fp16, [1, 188, 1024]> input_735_cast_fp16 = layer_norm(axes = input_735_axes_0, beta = module_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_feed_forward2_weight_to_fp16, x = input_733_cast_fp16)[name = string("input_735_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_13_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(191504064))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193601280))))[name = string("module_layers_13_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_13_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_13_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193863488)))];
            tensor<fp16, [1, 188, 4096]> linear_125_cast_fp16 = linear(bias = module_layers_13_feed_forward2_linear1_bias_to_fp16, weight = module_layers_13_feed_forward2_linear1_weight_to_fp16_quantized, x = input_735_cast_fp16)[name = string("linear_125_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_739_cast_fp16 = silu(x = linear_125_cast_fp16)[name = string("input_739_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_13_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(193871744))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195968960))))[name = string("module_layers_13_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_13_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_13_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196231168)))];
            tensor<fp16, [1, 188, 1024]> linear_126_cast_fp16 = linear(bias = module_layers_13_feed_forward2_linear2_bias_to_fp16, weight = module_layers_13_feed_forward2_linear2_weight_to_fp16_quantized, x = input_739_cast_fp16)[name = string("linear_126_cast_fp16")];
            fp16 var_2787_to_fp16 = const()[name = string("op_2787_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2788_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_2787_to_fp16)[name = string("op_2788_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_745_cast_fp16 = add(x = input_733_cast_fp16, y = var_2788_cast_fp16)[name = string("input_745_cast_fp16")];
            tensor<int32, [1]> input_747_axes_0 = const()[name = string("input_747_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_13_norm_out_weight_to_fp16 = const()[name = string("module_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196233280)))];
            tensor<fp16, [1024]> module_layers_13_norm_out_bias_to_fp16 = const()[name = string("module_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196235392)))];
            tensor<fp16, [1, 188, 1024]> input_747_cast_fp16 = layer_norm(axes = input_747_axes_0, beta = module_layers_13_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_13_norm_out_weight_to_fp16, x = input_745_cast_fp16)[name = string("input_747_cast_fp16")];
            tensor<int32, [1]> input_749_axes_0 = const()[name = string("input_749_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196237504)))];
            tensor<fp16, [1024]> module_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196239616)))];
            tensor<fp16, [1, 188, 1024]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = module_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward1_weight_to_fp16, x = input_747_cast_fp16)[name = string("input_749_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_14_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(196241728))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198338944))))[name = string("module_layers_14_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_14_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_14_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198601152)))];
            tensor<fp16, [1, 188, 4096]> linear_127_cast_fp16 = linear(bias = module_layers_14_feed_forward1_linear1_bias_to_fp16, weight = module_layers_14_feed_forward1_linear1_weight_to_fp16_quantized, x = input_749_cast_fp16)[name = string("linear_127_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_753_cast_fp16 = silu(x = linear_127_cast_fp16)[name = string("input_753_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_14_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198609408))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200706624))))[name = string("module_layers_14_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_14_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_14_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200968832)))];
            tensor<fp16, [1, 188, 1024]> linear_128_cast_fp16 = linear(bias = module_layers_14_feed_forward1_linear2_bias_to_fp16, weight = module_layers_14_feed_forward1_linear2_weight_to_fp16_quantized, x = input_753_cast_fp16)[name = string("linear_128_cast_fp16")];
            fp16 var_2818_to_fp16 = const()[name = string("op_2818_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2819_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_2818_to_fp16)[name = string("op_2819_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_759_cast_fp16 = add(x = input_747_cast_fp16, y = var_2819_cast_fp16)[name = string("input_759_cast_fp16")];
            tensor<int32, [1]> query_29_axes_0 = const()[name = string("query_29_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_14_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200970944)))];
            tensor<fp16, [1024]> module_layers_14_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200973056)))];
            tensor<fp16, [1, 188, 1024]> query_29_cast_fp16 = layer_norm(axes = query_29_axes_0, beta = module_layers_14_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_self_att_weight_to_fp16, x = input_759_cast_fp16)[name = string("query_29_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200975168))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201499520))))[name = string("module_layers_14_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_14_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_14_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201565120)))];
            tensor<fp16, [1, 188, 1024]> linear_129_cast_fp16 = linear(bias = module_layers_14_self_attn_linear_q_bias_to_fp16, weight = module_layers_14_self_attn_linear_q_weight_to_fp16_quantized, x = query_29_cast_fp16)[name = string("linear_129_cast_fp16")];
            tensor<int32, [4]> var_2836 = const()[name = string("op_2836"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_85_cast_fp16 = reshape(shape = var_2836, x = linear_129_cast_fp16)[name = string("q_85_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(201567232))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202091584))))[name = string("module_layers_14_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_14_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_14_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202157184)))];
            tensor<fp16, [1, 188, 1024]> linear_130_cast_fp16 = linear(bias = module_layers_14_self_attn_linear_k_bias_to_fp16, weight = module_layers_14_self_attn_linear_k_weight_to_fp16_quantized, x = query_29_cast_fp16)[name = string("linear_130_cast_fp16")];
            tensor<int32, [4]> var_2841 = const()[name = string("op_2841"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_57_cast_fp16 = reshape(shape = var_2841, x = linear_130_cast_fp16)[name = string("k_57_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202159296))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202683648))))[name = string("module_layers_14_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_14_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_14_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202749248)))];
            tensor<fp16, [1, 188, 1024]> linear_131_cast_fp16 = linear(bias = module_layers_14_self_attn_linear_v_bias_to_fp16, weight = module_layers_14_self_attn_linear_v_weight_to_fp16_quantized, x = query_29_cast_fp16)[name = string("linear_131_cast_fp16")];
            tensor<int32, [4]> var_2846 = const()[name = string("op_2846"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_29_cast_fp16 = reshape(shape = var_2846, x = linear_131_cast_fp16)[name = string("v_29_cast_fp16")];
            tensor<int32, [4]> value_31_perm_0 = const()[name = string("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202751360))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202751936))))[name = string("module_layers_14_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2858_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_2858_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_14_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202752064))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202752640))))[name = string("module_layers_14_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_2860_cast_fp16 = add(x = q_85_cast_fp16, y = module_layers_14_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_2860_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_29_perm_0 = const()[name = string("q_with_bias_v_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_315_transpose_x_0 = const()[name = string("x_315_transpose_x_0"), val = bool(false)];
            bool x_315_transpose_y_0 = const()[name = string("x_315_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_2862_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202752768))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202944832))))[name = string("op_2862_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_29_cast_fp16 = transpose(perm = q_with_bias_v_29_perm_0, x = var_2860_cast_fp16)[name = string("transpose_318")];
            tensor<fp16, [1, 8, 188, 375]> x_315_cast_fp16 = matmul(transpose_x = x_315_transpose_x_0, transpose_y = x_315_transpose_y_0, x = q_with_bias_v_29_cast_fp16, y = op_2862_to_fp16_quantized)[name = string("x_315_cast_fp16")];
            tensor<int32, [8]> x_317_pad_0 = const()[name = string("x_317_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_317_mode_0 = const()[name = string("x_317_mode_0"), val = string("constant")];
            fp16 const_210_to_fp16 = const()[name = string("const_210_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_317_cast_fp16 = pad(constant_val = const_210_to_fp16, mode = x_317_mode_0, pad = x_317_pad_0, x = x_315_cast_fp16)[name = string("x_317_cast_fp16")];
            tensor<int32, [4]> var_2870 = const()[name = string("op_2870"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_319_cast_fp16 = reshape(shape = var_2870, x = x_317_cast_fp16)[name = string("x_319_cast_fp16")];
            tensor<int32, [4]> var_2874_begin_0 = const()[name = string("op_2874_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2874_end_0 = const()[name = string("op_2874_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_2874_end_mask_0 = const()[name = string("op_2874_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_2874_cast_fp16 = slice_by_index(begin = var_2874_begin_0, end = var_2874_end_0, end_mask = var_2874_end_mask_0, x = x_319_cast_fp16)[name = string("op_2874_cast_fp16")];
            tensor<int32, [4]> var_2875 = const()[name = string("op_2875"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_57_cast_fp16 = reshape(shape = var_2875, x = var_2874_cast_fp16)[name = string("matrix_bd_57_cast_fp16")];
            bool matrix_ac_29_transpose_x_0 = const()[name = string("matrix_ac_29_transpose_x_0"), val = bool(false)];
            bool matrix_ac_29_transpose_y_0 = const()[name = string("matrix_ac_29_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_156_perm_0 = const()[name = string("transpose_156_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_157_perm_0 = const()[name = string("transpose_157_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_157 = transpose(perm = transpose_157_perm_0, x = k_57_cast_fp16)[name = string("transpose_316")];
            tensor<fp16, [1, 8, 188, 128]> transpose_156 = transpose(perm = transpose_156_perm_0, x = var_2858_cast_fp16)[name = string("transpose_317")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_29_cast_fp16 = matmul(transpose_x = matrix_ac_29_transpose_x_0, transpose_y = matrix_ac_29_transpose_y_0, x = transpose_156, y = transpose_157)[name = string("matrix_ac_29_cast_fp16")];
            tensor<int32, [4]> matrix_bd_59_begin_0 = const()[name = string("matrix_bd_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_59_end_0 = const()[name = string("matrix_bd_59_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_59_end_mask_0 = const()[name = string("matrix_bd_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_59_cast_fp16 = slice_by_index(begin = matrix_bd_59_begin_0, end = matrix_bd_59_end_0, end_mask = matrix_bd_59_end_mask_0, x = matrix_bd_57_cast_fp16)[name = string("matrix_bd_59_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2884_cast_fp16 = add(x = matrix_ac_29_cast_fp16, y = matrix_bd_59_cast_fp16)[name = string("op_2884_cast_fp16")];
            fp16 _inversed_scores_57_y_0_to_fp16 = const()[name = string("_inversed_scores_57_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_57_cast_fp16 = mul(x = var_2884_cast_fp16, y = _inversed_scores_57_y_0_to_fp16)[name = string("_inversed_scores_57_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_59_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_57_cast_fp16, cond = mask_11)[name = string("scores_59_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_2890_cast_fp16 = softmax(axis = var_23, x = scores_59_cast_fp16)[name = string("op_2890_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_761_cast_fp16 = select(a = var_11_to_fp16, b = var_2890_cast_fp16, cond = mask_11)[name = string("input_761_cast_fp16")];
            bool x_321_transpose_x_0 = const()[name = string("x_321_transpose_x_0"), val = bool(false)];
            bool x_321_transpose_y_0 = const()[name = string("x_321_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = v_29_cast_fp16)[name = string("transpose_315")];
            tensor<fp16, [1, 8, 188, 128]> x_321_cast_fp16 = matmul(transpose_x = x_321_transpose_x_0, transpose_y = x_321_transpose_y_0, x = input_761_cast_fp16, y = value_31_cast_fp16)[name = string("x_321_cast_fp16")];
            tensor<int32, [4]> var_2894_perm_0 = const()[name = string("op_2894_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2895 = const()[name = string("op_2895"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_2894_cast_fp16 = transpose(perm = var_2894_perm_0, x = x_321_cast_fp16)[name = string("transpose_314")];
            tensor<fp16, [1, 188, 1024]> input_763_cast_fp16 = reshape(shape = var_2895, x = var_2894_cast_fp16)[name = string("input_763_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_14_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202947904))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203472256))))[name = string("module_layers_14_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_14_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_14_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203537856)))];
            tensor<fp16, [1, 188, 1024]> linear_133_cast_fp16 = linear(bias = module_layers_14_self_attn_linear_out_bias_to_fp16, weight = module_layers_14_self_attn_linear_out_weight_to_fp16_quantized, x = input_763_cast_fp16)[name = string("linear_133_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_767_cast_fp16 = add(x = input_759_cast_fp16, y = linear_133_cast_fp16)[name = string("input_767_cast_fp16")];
            tensor<int32, [1]> x_325_axes_0 = const()[name = string("x_325_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_14_norm_conv_weight_to_fp16 = const()[name = string("module_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203539968)))];
            tensor<fp16, [1024]> module_layers_14_norm_conv_bias_to_fp16 = const()[name = string("module_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203542080)))];
            tensor<fp16, [1, 188, 1024]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = module_layers_14_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_conv_weight_to_fp16, x = input_767_cast_fp16)[name = string("x_325_cast_fp16")];
            tensor<int32, [3]> input_769_perm_0 = const()[name = string("input_769_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_771_pad_type_0 = const()[name = string("input_771_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_771_strides_0 = const()[name = string("input_771_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_771_pad_0 = const()[name = string("input_771_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_771_dilations_0 = const()[name = string("input_771_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_771_groups_0 = const()[name = string("input_771_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203544192))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204592832))))[name = string("module_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_14_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_14_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204723968)))];
            tensor<fp16, [1, 1024, 188]> input_769_cast_fp16 = transpose(perm = input_769_perm_0, x = x_325_cast_fp16)[name = string("transpose_313")];
            tensor<fp16, [1, 2048, 188]> input_771_cast_fp16 = conv(bias = module_layers_14_conv_pointwise_conv1_bias_to_fp16, dilations = input_771_dilations_0, groups = input_771_groups_0, pad = input_771_pad_0, pad_type = input_771_pad_type_0, strides = input_771_strides_0, weight = module_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_769_cast_fp16)[name = string("input_771_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, 188]> x_327_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_771_cast_fp16)[name = string("x_327_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_773_cast_fp16 = select(a = var_11_to_fp16, b = x_327_cast_fp16, cond = var_483)[name = string("input_773_cast_fp16")];
            tensor<int32, [6]> input_775_pad_0 = const()[name = string("input_775_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_775_mode_0 = const()[name = string("input_775_mode_0"), val = string("constant")];
            fp16 const_213_to_fp16 = const()[name = string("const_213_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_775_cast_fp16 = pad(constant_val = const_213_to_fp16, mode = input_775_mode_0, pad = input_775_pad_0, x = input_773_cast_fp16)[name = string("input_775_cast_fp16")];
            string input_777_pad_type_0 = const()[name = string("input_777_pad_type_0"), val = string("valid")];
            int32 input_777_groups_0 = const()[name = string("input_777_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_777_strides_0 = const()[name = string("input_777_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_777_pad_0 = const()[name = string("input_777_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_777_dilations_0 = const()[name = string("input_777_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_412_to_fp16 = const()[name = string("const_412_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204728128)))];
            tensor<fp16, [1024]> const_413_to_fp16 = const()[name = string("const_413_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204746624)))];
            tensor<fp16, [1, 1024, 188]> input_779_cast_fp16 = conv(bias = const_413_to_fp16, dilations = input_777_dilations_0, groups = input_777_groups_0, pad = input_777_pad_0, pad_type = input_777_pad_type_0, strides = input_777_strides_0, weight = const_412_to_fp16, x = input_775_cast_fp16)[name = string("input_779_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_781_cast_fp16 = silu(x = input_779_cast_fp16)[name = string("input_781_cast_fp16")];
            string x_329_pad_type_0 = const()[name = string("x_329_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_329_strides_0 = const()[name = string("x_329_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_329_pad_0 = const()[name = string("x_329_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_329_dilations_0 = const()[name = string("x_329_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_329_groups_0 = const()[name = string("x_329_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(204748736))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205273088))))[name = string("module_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_14_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_14_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205338688)))];
            tensor<fp16, [1, 1024, 188]> x_329_cast_fp16 = conv(bias = module_layers_14_conv_pointwise_conv2_bias_to_fp16, dilations = x_329_dilations_0, groups = x_329_groups_0, pad = x_329_pad_0, pad_type = x_329_pad_type_0, strides = x_329_strides_0, weight = module_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_781_cast_fp16)[name = string("x_329_cast_fp16")];
            tensor<int32, [3]> input_783_perm_0 = const()[name = string("input_783_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_783_cast_fp16 = transpose(perm = input_783_perm_0, x = x_329_cast_fp16)[name = string("transpose_312")];
            tensor<fp16, [1, 188, 1024]> input_785_cast_fp16 = add(x = input_767_cast_fp16, y = input_783_cast_fp16)[name = string("input_785_cast_fp16")];
            tensor<int32, [1]> input_787_axes_0 = const()[name = string("input_787_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205340800)))];
            tensor<fp16, [1024]> module_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205342912)))];
            tensor<fp16, [1, 188, 1024]> input_787_cast_fp16 = layer_norm(axes = input_787_axes_0, beta = module_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_feed_forward2_weight_to_fp16, x = input_785_cast_fp16)[name = string("input_787_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_14_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205345024))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207442240))))[name = string("module_layers_14_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_14_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_14_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207704448)))];
            tensor<fp16, [1, 188, 4096]> linear_134_cast_fp16 = linear(bias = module_layers_14_feed_forward2_linear1_bias_to_fp16, weight = module_layers_14_feed_forward2_linear1_weight_to_fp16_quantized, x = input_787_cast_fp16)[name = string("linear_134_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_791_cast_fp16 = silu(x = linear_134_cast_fp16)[name = string("input_791_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_14_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207712704))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209809920))))[name = string("module_layers_14_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_14_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_14_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210072128)))];
            tensor<fp16, [1, 188, 1024]> linear_135_cast_fp16 = linear(bias = module_layers_14_feed_forward2_linear2_bias_to_fp16, weight = module_layers_14_feed_forward2_linear2_weight_to_fp16_quantized, x = input_791_cast_fp16)[name = string("linear_135_cast_fp16")];
            fp16 var_2961_to_fp16 = const()[name = string("op_2961_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2962_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_2961_to_fp16)[name = string("op_2962_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_797_cast_fp16 = add(x = input_785_cast_fp16, y = var_2962_cast_fp16)[name = string("input_797_cast_fp16")];
            tensor<int32, [1]> input_799_axes_0 = const()[name = string("input_799_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_14_norm_out_weight_to_fp16 = const()[name = string("module_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210074240)))];
            tensor<fp16, [1024]> module_layers_14_norm_out_bias_to_fp16 = const()[name = string("module_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210076352)))];
            tensor<fp16, [1, 188, 1024]> input_799_cast_fp16 = layer_norm(axes = input_799_axes_0, beta = module_layers_14_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_14_norm_out_weight_to_fp16, x = input_797_cast_fp16)[name = string("input_799_cast_fp16")];
            tensor<int32, [1]> input_801_axes_0 = const()[name = string("input_801_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210078464)))];
            tensor<fp16, [1024]> module_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210080576)))];
            tensor<fp16, [1, 188, 1024]> input_801_cast_fp16 = layer_norm(axes = input_801_axes_0, beta = module_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward1_weight_to_fp16, x = input_799_cast_fp16)[name = string("input_801_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_15_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(210082688))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212179904))))[name = string("module_layers_15_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_15_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_15_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212442112)))];
            tensor<fp16, [1, 188, 4096]> linear_136_cast_fp16 = linear(bias = module_layers_15_feed_forward1_linear1_bias_to_fp16, weight = module_layers_15_feed_forward1_linear1_weight_to_fp16_quantized, x = input_801_cast_fp16)[name = string("linear_136_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_805_cast_fp16 = silu(x = linear_136_cast_fp16)[name = string("input_805_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_15_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(212450368))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214547584))))[name = string("module_layers_15_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_15_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_15_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214809792)))];
            tensor<fp16, [1, 188, 1024]> linear_137_cast_fp16 = linear(bias = module_layers_15_feed_forward1_linear2_bias_to_fp16, weight = module_layers_15_feed_forward1_linear2_weight_to_fp16_quantized, x = input_805_cast_fp16)[name = string("linear_137_cast_fp16")];
            fp16 var_2992_to_fp16 = const()[name = string("op_2992_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_2993_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_2992_to_fp16)[name = string("op_2993_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_811_cast_fp16 = add(x = input_799_cast_fp16, y = var_2993_cast_fp16)[name = string("input_811_cast_fp16")];
            tensor<int32, [1]> query_31_axes_0 = const()[name = string("query_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_15_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214811904)))];
            tensor<fp16, [1024]> module_layers_15_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214814016)))];
            tensor<fp16, [1, 188, 1024]> query_31_cast_fp16 = layer_norm(axes = query_31_axes_0, beta = module_layers_15_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_self_att_weight_to_fp16, x = input_811_cast_fp16)[name = string("query_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(214816128))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215340480))))[name = string("module_layers_15_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_15_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_15_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215406080)))];
            tensor<fp16, [1, 188, 1024]> linear_138_cast_fp16 = linear(bias = module_layers_15_self_attn_linear_q_bias_to_fp16, weight = module_layers_15_self_attn_linear_q_weight_to_fp16_quantized, x = query_31_cast_fp16)[name = string("linear_138_cast_fp16")];
            tensor<int32, [4]> var_3010 = const()[name = string("op_3010"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_91_cast_fp16 = reshape(shape = var_3010, x = linear_138_cast_fp16)[name = string("q_91_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215408192))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215932544))))[name = string("module_layers_15_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_15_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_15_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215998144)))];
            tensor<fp16, [1, 188, 1024]> linear_139_cast_fp16 = linear(bias = module_layers_15_self_attn_linear_k_bias_to_fp16, weight = module_layers_15_self_attn_linear_k_weight_to_fp16_quantized, x = query_31_cast_fp16)[name = string("linear_139_cast_fp16")];
            tensor<int32, [4]> var_3015 = const()[name = string("op_3015"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_61_cast_fp16 = reshape(shape = var_3015, x = linear_139_cast_fp16)[name = string("k_61_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216000256))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216524608))))[name = string("module_layers_15_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_15_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_15_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216590208)))];
            tensor<fp16, [1, 188, 1024]> linear_140_cast_fp16 = linear(bias = module_layers_15_self_attn_linear_v_bias_to_fp16, weight = module_layers_15_self_attn_linear_v_weight_to_fp16_quantized, x = query_31_cast_fp16)[name = string("linear_140_cast_fp16")];
            tensor<int32, [4]> var_3020 = const()[name = string("op_3020"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_31_cast_fp16 = reshape(shape = var_3020, x = linear_140_cast_fp16)[name = string("v_31_cast_fp16")];
            tensor<int32, [4]> value_33_perm_0 = const()[name = string("value_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216592320))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216592896))))[name = string("module_layers_15_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3032_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_3032_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_15_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216593024))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216593600))))[name = string("module_layers_15_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3034_cast_fp16 = add(x = q_91_cast_fp16, y = module_layers_15_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_3034_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_31_perm_0 = const()[name = string("q_with_bias_v_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_337_transpose_x_0 = const()[name = string("x_337_transpose_x_0"), val = bool(false)];
            bool x_337_transpose_y_0 = const()[name = string("x_337_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_3036_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216593728))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216785792))))[name = string("op_3036_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_31_cast_fp16 = transpose(perm = q_with_bias_v_31_perm_0, x = var_3034_cast_fp16)[name = string("transpose_311")];
            tensor<fp16, [1, 8, 188, 375]> x_337_cast_fp16 = matmul(transpose_x = x_337_transpose_x_0, transpose_y = x_337_transpose_y_0, x = q_with_bias_v_31_cast_fp16, y = op_3036_to_fp16_quantized)[name = string("x_337_cast_fp16")];
            tensor<int32, [8]> x_339_pad_0 = const()[name = string("x_339_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_339_mode_0 = const()[name = string("x_339_mode_0"), val = string("constant")];
            fp16 const_220_to_fp16 = const()[name = string("const_220_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_339_cast_fp16 = pad(constant_val = const_220_to_fp16, mode = x_339_mode_0, pad = x_339_pad_0, x = x_337_cast_fp16)[name = string("x_339_cast_fp16")];
            tensor<int32, [4]> var_3044 = const()[name = string("op_3044"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_341_cast_fp16 = reshape(shape = var_3044, x = x_339_cast_fp16)[name = string("x_341_cast_fp16")];
            tensor<int32, [4]> var_3048_begin_0 = const()[name = string("op_3048_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3048_end_0 = const()[name = string("op_3048_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3048_end_mask_0 = const()[name = string("op_3048_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3048_cast_fp16 = slice_by_index(begin = var_3048_begin_0, end = var_3048_end_0, end_mask = var_3048_end_mask_0, x = x_341_cast_fp16)[name = string("op_3048_cast_fp16")];
            tensor<int32, [4]> var_3049 = const()[name = string("op_3049"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_61_cast_fp16 = reshape(shape = var_3049, x = var_3048_cast_fp16)[name = string("matrix_bd_61_cast_fp16")];
            bool matrix_ac_31_transpose_x_0 = const()[name = string("matrix_ac_31_transpose_x_0"), val = bool(false)];
            bool matrix_ac_31_transpose_y_0 = const()[name = string("matrix_ac_31_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_158_perm_0 = const()[name = string("transpose_158_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_159_perm_0 = const()[name = string("transpose_159_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_159 = transpose(perm = transpose_159_perm_0, x = k_61_cast_fp16)[name = string("transpose_309")];
            tensor<fp16, [1, 8, 188, 128]> transpose_158 = transpose(perm = transpose_158_perm_0, x = var_3032_cast_fp16)[name = string("transpose_310")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_31_cast_fp16 = matmul(transpose_x = matrix_ac_31_transpose_x_0, transpose_y = matrix_ac_31_transpose_y_0, x = transpose_158, y = transpose_159)[name = string("matrix_ac_31_cast_fp16")];
            tensor<int32, [4]> matrix_bd_63_begin_0 = const()[name = string("matrix_bd_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_63_end_0 = const()[name = string("matrix_bd_63_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_63_end_mask_0 = const()[name = string("matrix_bd_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_63_cast_fp16 = slice_by_index(begin = matrix_bd_63_begin_0, end = matrix_bd_63_end_0, end_mask = matrix_bd_63_end_mask_0, x = matrix_bd_61_cast_fp16)[name = string("matrix_bd_63_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3058_cast_fp16 = add(x = matrix_ac_31_cast_fp16, y = matrix_bd_63_cast_fp16)[name = string("op_3058_cast_fp16")];
            fp16 _inversed_scores_61_y_0_to_fp16 = const()[name = string("_inversed_scores_61_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_61_cast_fp16 = mul(x = var_3058_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = string("_inversed_scores_61_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_63_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_61_cast_fp16, cond = mask_11)[name = string("scores_63_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3064_cast_fp16 = softmax(axis = var_23, x = scores_63_cast_fp16)[name = string("op_3064_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_813_cast_fp16 = select(a = var_11_to_fp16, b = var_3064_cast_fp16, cond = mask_11)[name = string("input_813_cast_fp16")];
            bool x_343_transpose_x_0 = const()[name = string("x_343_transpose_x_0"), val = bool(false)];
            bool x_343_transpose_y_0 = const()[name = string("x_343_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_33_cast_fp16 = transpose(perm = value_33_perm_0, x = v_31_cast_fp16)[name = string("transpose_308")];
            tensor<fp16, [1, 8, 188, 128]> x_343_cast_fp16 = matmul(transpose_x = x_343_transpose_x_0, transpose_y = x_343_transpose_y_0, x = input_813_cast_fp16, y = value_33_cast_fp16)[name = string("x_343_cast_fp16")];
            tensor<int32, [4]> var_3068_perm_0 = const()[name = string("op_3068_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3069 = const()[name = string("op_3069"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3068_cast_fp16 = transpose(perm = var_3068_perm_0, x = x_343_cast_fp16)[name = string("transpose_307")];
            tensor<fp16, [1, 188, 1024]> input_815_cast_fp16 = reshape(shape = var_3069, x = var_3068_cast_fp16)[name = string("input_815_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_15_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216788864))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217313216))))[name = string("module_layers_15_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_15_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_15_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217378816)))];
            tensor<fp16, [1, 188, 1024]> linear_142_cast_fp16 = linear(bias = module_layers_15_self_attn_linear_out_bias_to_fp16, weight = module_layers_15_self_attn_linear_out_weight_to_fp16_quantized, x = input_815_cast_fp16)[name = string("linear_142_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_819_cast_fp16 = add(x = input_811_cast_fp16, y = linear_142_cast_fp16)[name = string("input_819_cast_fp16")];
            tensor<int32, [1]> x_347_axes_0 = const()[name = string("x_347_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_15_norm_conv_weight_to_fp16 = const()[name = string("module_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217380928)))];
            tensor<fp16, [1024]> module_layers_15_norm_conv_bias_to_fp16 = const()[name = string("module_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217383040)))];
            tensor<fp16, [1, 188, 1024]> x_347_cast_fp16 = layer_norm(axes = x_347_axes_0, beta = module_layers_15_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_conv_weight_to_fp16, x = input_819_cast_fp16)[name = string("x_347_cast_fp16")];
            tensor<int32, [3]> input_821_perm_0 = const()[name = string("input_821_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_823_pad_type_0 = const()[name = string("input_823_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_823_strides_0 = const()[name = string("input_823_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_823_pad_0 = const()[name = string("input_823_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_823_dilations_0 = const()[name = string("input_823_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_823_groups_0 = const()[name = string("input_823_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217385152))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218433792))))[name = string("module_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_15_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_15_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218564928)))];
            tensor<fp16, [1, 1024, 188]> input_821_cast_fp16 = transpose(perm = input_821_perm_0, x = x_347_cast_fp16)[name = string("transpose_306")];
            tensor<fp16, [1, 2048, 188]> input_823_cast_fp16 = conv(bias = module_layers_15_conv_pointwise_conv1_bias_to_fp16, dilations = input_823_dilations_0, groups = input_823_groups_0, pad = input_823_pad_0, pad_type = input_823_pad_type_0, strides = input_823_strides_0, weight = module_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_821_cast_fp16)[name = string("input_823_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, 188]> x_349_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_823_cast_fp16)[name = string("x_349_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_825_cast_fp16 = select(a = var_11_to_fp16, b = x_349_cast_fp16, cond = var_483)[name = string("input_825_cast_fp16")];
            tensor<int32, [6]> input_827_pad_0 = const()[name = string("input_827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_827_mode_0 = const()[name = string("input_827_mode_0"), val = string("constant")];
            fp16 const_223_to_fp16 = const()[name = string("const_223_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_827_cast_fp16 = pad(constant_val = const_223_to_fp16, mode = input_827_mode_0, pad = input_827_pad_0, x = input_825_cast_fp16)[name = string("input_827_cast_fp16")];
            string input_829_pad_type_0 = const()[name = string("input_829_pad_type_0"), val = string("valid")];
            int32 input_829_groups_0 = const()[name = string("input_829_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_829_strides_0 = const()[name = string("input_829_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_829_pad_0 = const()[name = string("input_829_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_829_dilations_0 = const()[name = string("input_829_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_414_to_fp16 = const()[name = string("const_414_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218569088)))];
            tensor<fp16, [1024]> const_415_to_fp16 = const()[name = string("const_415_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218587584)))];
            tensor<fp16, [1, 1024, 188]> input_831_cast_fp16 = conv(bias = const_415_to_fp16, dilations = input_829_dilations_0, groups = input_829_groups_0, pad = input_829_pad_0, pad_type = input_829_pad_type_0, strides = input_829_strides_0, weight = const_414_to_fp16, x = input_827_cast_fp16)[name = string("input_831_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_833_cast_fp16 = silu(x = input_831_cast_fp16)[name = string("input_833_cast_fp16")];
            string x_351_pad_type_0 = const()[name = string("x_351_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_351_strides_0 = const()[name = string("x_351_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_351_pad_0 = const()[name = string("x_351_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_351_dilations_0 = const()[name = string("x_351_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_351_groups_0 = const()[name = string("x_351_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218589696))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219114048))))[name = string("module_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_15_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_15_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219179648)))];
            tensor<fp16, [1, 1024, 188]> x_351_cast_fp16 = conv(bias = module_layers_15_conv_pointwise_conv2_bias_to_fp16, dilations = x_351_dilations_0, groups = x_351_groups_0, pad = x_351_pad_0, pad_type = x_351_pad_type_0, strides = x_351_strides_0, weight = module_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_833_cast_fp16)[name = string("x_351_cast_fp16")];
            tensor<int32, [3]> input_835_perm_0 = const()[name = string("input_835_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_835_cast_fp16 = transpose(perm = input_835_perm_0, x = x_351_cast_fp16)[name = string("transpose_305")];
            tensor<fp16, [1, 188, 1024]> input_837_cast_fp16 = add(x = input_819_cast_fp16, y = input_835_cast_fp16)[name = string("input_837_cast_fp16")];
            tensor<int32, [1]> input_839_axes_0 = const()[name = string("input_839_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219181760)))];
            tensor<fp16, [1024]> module_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219183872)))];
            tensor<fp16, [1, 188, 1024]> input_839_cast_fp16 = layer_norm(axes = input_839_axes_0, beta = module_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_feed_forward2_weight_to_fp16, x = input_837_cast_fp16)[name = string("input_839_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_15_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219185984))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221283200))))[name = string("module_layers_15_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_15_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_15_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221545408)))];
            tensor<fp16, [1, 188, 4096]> linear_143_cast_fp16 = linear(bias = module_layers_15_feed_forward2_linear1_bias_to_fp16, weight = module_layers_15_feed_forward2_linear1_weight_to_fp16_quantized, x = input_839_cast_fp16)[name = string("linear_143_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_843_cast_fp16 = silu(x = linear_143_cast_fp16)[name = string("input_843_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_15_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221553664))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223650880))))[name = string("module_layers_15_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_15_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_15_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223913088)))];
            tensor<fp16, [1, 188, 1024]> linear_144_cast_fp16 = linear(bias = module_layers_15_feed_forward2_linear2_bias_to_fp16, weight = module_layers_15_feed_forward2_linear2_weight_to_fp16_quantized, x = input_843_cast_fp16)[name = string("linear_144_cast_fp16")];
            fp16 var_3135_to_fp16 = const()[name = string("op_3135_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3136_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_3135_to_fp16)[name = string("op_3136_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_849_cast_fp16 = add(x = input_837_cast_fp16, y = var_3136_cast_fp16)[name = string("input_849_cast_fp16")];
            tensor<int32, [1]> input_851_axes_0 = const()[name = string("input_851_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_15_norm_out_weight_to_fp16 = const()[name = string("module_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223915200)))];
            tensor<fp16, [1024]> module_layers_15_norm_out_bias_to_fp16 = const()[name = string("module_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223917312)))];
            tensor<fp16, [1, 188, 1024]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = module_layers_15_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_15_norm_out_weight_to_fp16, x = input_849_cast_fp16)[name = string("input_851_cast_fp16")];
            tensor<int32, [1]> input_853_axes_0 = const()[name = string("input_853_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223919424)))];
            tensor<fp16, [1024]> module_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223921536)))];
            tensor<fp16, [1, 188, 1024]> input_853_cast_fp16 = layer_norm(axes = input_853_axes_0, beta = module_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward1_weight_to_fp16, x = input_851_cast_fp16)[name = string("input_853_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_16_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(223923648))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226020864))))[name = string("module_layers_16_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_16_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_16_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226283072)))];
            tensor<fp16, [1, 188, 4096]> linear_145_cast_fp16 = linear(bias = module_layers_16_feed_forward1_linear1_bias_to_fp16, weight = module_layers_16_feed_forward1_linear1_weight_to_fp16_quantized, x = input_853_cast_fp16)[name = string("linear_145_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_857_cast_fp16 = silu(x = linear_145_cast_fp16)[name = string("input_857_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_16_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(226291328))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228388544))))[name = string("module_layers_16_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_16_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_16_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228650752)))];
            tensor<fp16, [1, 188, 1024]> linear_146_cast_fp16 = linear(bias = module_layers_16_feed_forward1_linear2_bias_to_fp16, weight = module_layers_16_feed_forward1_linear2_weight_to_fp16_quantized, x = input_857_cast_fp16)[name = string("linear_146_cast_fp16")];
            fp16 var_3166_to_fp16 = const()[name = string("op_3166_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3167_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_3166_to_fp16)[name = string("op_3167_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_863_cast_fp16 = add(x = input_851_cast_fp16, y = var_3167_cast_fp16)[name = string("input_863_cast_fp16")];
            tensor<int32, [1]> query_33_axes_0 = const()[name = string("query_33_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_16_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228652864)))];
            tensor<fp16, [1024]> module_layers_16_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228654976)))];
            tensor<fp16, [1, 188, 1024]> query_33_cast_fp16 = layer_norm(axes = query_33_axes_0, beta = module_layers_16_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_self_att_weight_to_fp16, x = input_863_cast_fp16)[name = string("query_33_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228657088))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229181440))))[name = string("module_layers_16_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_16_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_16_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229247040)))];
            tensor<fp16, [1, 188, 1024]> linear_147_cast_fp16 = linear(bias = module_layers_16_self_attn_linear_q_bias_to_fp16, weight = module_layers_16_self_attn_linear_q_weight_to_fp16_quantized, x = query_33_cast_fp16)[name = string("linear_147_cast_fp16")];
            tensor<int32, [4]> var_3184 = const()[name = string("op_3184"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_97_cast_fp16 = reshape(shape = var_3184, x = linear_147_cast_fp16)[name = string("q_97_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229249152))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229773504))))[name = string("module_layers_16_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_16_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_16_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229839104)))];
            tensor<fp16, [1, 188, 1024]> linear_148_cast_fp16 = linear(bias = module_layers_16_self_attn_linear_k_bias_to_fp16, weight = module_layers_16_self_attn_linear_k_weight_to_fp16_quantized, x = query_33_cast_fp16)[name = string("linear_148_cast_fp16")];
            tensor<int32, [4]> var_3189 = const()[name = string("op_3189"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_65_cast_fp16 = reshape(shape = var_3189, x = linear_148_cast_fp16)[name = string("k_65_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(229841216))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230365568))))[name = string("module_layers_16_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_16_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_16_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230431168)))];
            tensor<fp16, [1, 188, 1024]> linear_149_cast_fp16 = linear(bias = module_layers_16_self_attn_linear_v_bias_to_fp16, weight = module_layers_16_self_attn_linear_v_weight_to_fp16_quantized, x = query_33_cast_fp16)[name = string("linear_149_cast_fp16")];
            tensor<int32, [4]> var_3194 = const()[name = string("op_3194"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_33_cast_fp16 = reshape(shape = var_3194, x = linear_149_cast_fp16)[name = string("v_33_cast_fp16")];
            tensor<int32, [4]> value_35_perm_0 = const()[name = string("value_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230433280))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230433856))))[name = string("module_layers_16_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3206_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_3206_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_16_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230433984))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230434560))))[name = string("module_layers_16_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3208_cast_fp16 = add(x = q_97_cast_fp16, y = module_layers_16_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_3208_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_33_perm_0 = const()[name = string("q_with_bias_v_33_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_359_transpose_x_0 = const()[name = string("x_359_transpose_x_0"), val = bool(false)];
            bool x_359_transpose_y_0 = const()[name = string("x_359_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_3210_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230434688))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230626752))))[name = string("op_3210_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_33_cast_fp16 = transpose(perm = q_with_bias_v_33_perm_0, x = var_3208_cast_fp16)[name = string("transpose_304")];
            tensor<fp16, [1, 8, 188, 375]> x_359_cast_fp16 = matmul(transpose_x = x_359_transpose_x_0, transpose_y = x_359_transpose_y_0, x = q_with_bias_v_33_cast_fp16, y = op_3210_to_fp16_quantized)[name = string("x_359_cast_fp16")];
            tensor<int32, [8]> x_361_pad_0 = const()[name = string("x_361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_361_mode_0 = const()[name = string("x_361_mode_0"), val = string("constant")];
            fp16 const_230_to_fp16 = const()[name = string("const_230_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_361_cast_fp16 = pad(constant_val = const_230_to_fp16, mode = x_361_mode_0, pad = x_361_pad_0, x = x_359_cast_fp16)[name = string("x_361_cast_fp16")];
            tensor<int32, [4]> var_3218 = const()[name = string("op_3218"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_363_cast_fp16 = reshape(shape = var_3218, x = x_361_cast_fp16)[name = string("x_363_cast_fp16")];
            tensor<int32, [4]> var_3222_begin_0 = const()[name = string("op_3222_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3222_end_0 = const()[name = string("op_3222_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3222_end_mask_0 = const()[name = string("op_3222_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3222_cast_fp16 = slice_by_index(begin = var_3222_begin_0, end = var_3222_end_0, end_mask = var_3222_end_mask_0, x = x_363_cast_fp16)[name = string("op_3222_cast_fp16")];
            tensor<int32, [4]> var_3223 = const()[name = string("op_3223"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_65_cast_fp16 = reshape(shape = var_3223, x = var_3222_cast_fp16)[name = string("matrix_bd_65_cast_fp16")];
            bool matrix_ac_33_transpose_x_0 = const()[name = string("matrix_ac_33_transpose_x_0"), val = bool(false)];
            bool matrix_ac_33_transpose_y_0 = const()[name = string("matrix_ac_33_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_160_perm_0 = const()[name = string("transpose_160_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_161_perm_0 = const()[name = string("transpose_161_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_161 = transpose(perm = transpose_161_perm_0, x = k_65_cast_fp16)[name = string("transpose_302")];
            tensor<fp16, [1, 8, 188, 128]> transpose_160 = transpose(perm = transpose_160_perm_0, x = var_3206_cast_fp16)[name = string("transpose_303")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_33_cast_fp16 = matmul(transpose_x = matrix_ac_33_transpose_x_0, transpose_y = matrix_ac_33_transpose_y_0, x = transpose_160, y = transpose_161)[name = string("matrix_ac_33_cast_fp16")];
            tensor<int32, [4]> matrix_bd_67_begin_0 = const()[name = string("matrix_bd_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_67_end_0 = const()[name = string("matrix_bd_67_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_67_end_mask_0 = const()[name = string("matrix_bd_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_67_cast_fp16 = slice_by_index(begin = matrix_bd_67_begin_0, end = matrix_bd_67_end_0, end_mask = matrix_bd_67_end_mask_0, x = matrix_bd_65_cast_fp16)[name = string("matrix_bd_67_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3232_cast_fp16 = add(x = matrix_ac_33_cast_fp16, y = matrix_bd_67_cast_fp16)[name = string("op_3232_cast_fp16")];
            fp16 _inversed_scores_65_y_0_to_fp16 = const()[name = string("_inversed_scores_65_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_65_cast_fp16 = mul(x = var_3232_cast_fp16, y = _inversed_scores_65_y_0_to_fp16)[name = string("_inversed_scores_65_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_67_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_65_cast_fp16, cond = mask_11)[name = string("scores_67_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3238_cast_fp16 = softmax(axis = var_23, x = scores_67_cast_fp16)[name = string("op_3238_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_865_cast_fp16 = select(a = var_11_to_fp16, b = var_3238_cast_fp16, cond = mask_11)[name = string("input_865_cast_fp16")];
            bool x_365_transpose_x_0 = const()[name = string("x_365_transpose_x_0"), val = bool(false)];
            bool x_365_transpose_y_0 = const()[name = string("x_365_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_35_cast_fp16 = transpose(perm = value_35_perm_0, x = v_33_cast_fp16)[name = string("transpose_301")];
            tensor<fp16, [1, 8, 188, 128]> x_365_cast_fp16 = matmul(transpose_x = x_365_transpose_x_0, transpose_y = x_365_transpose_y_0, x = input_865_cast_fp16, y = value_35_cast_fp16)[name = string("x_365_cast_fp16")];
            tensor<int32, [4]> var_3242_perm_0 = const()[name = string("op_3242_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3243 = const()[name = string("op_3243"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3242_cast_fp16 = transpose(perm = var_3242_perm_0, x = x_365_cast_fp16)[name = string("transpose_300")];
            tensor<fp16, [1, 188, 1024]> input_867_cast_fp16 = reshape(shape = var_3243, x = var_3242_cast_fp16)[name = string("input_867_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_16_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(230629824))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231154176))))[name = string("module_layers_16_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_16_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_16_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231219776)))];
            tensor<fp16, [1, 188, 1024]> linear_151_cast_fp16 = linear(bias = module_layers_16_self_attn_linear_out_bias_to_fp16, weight = module_layers_16_self_attn_linear_out_weight_to_fp16_quantized, x = input_867_cast_fp16)[name = string("linear_151_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_871_cast_fp16 = add(x = input_863_cast_fp16, y = linear_151_cast_fp16)[name = string("input_871_cast_fp16")];
            tensor<int32, [1]> x_369_axes_0 = const()[name = string("x_369_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_16_norm_conv_weight_to_fp16 = const()[name = string("module_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231221888)))];
            tensor<fp16, [1024]> module_layers_16_norm_conv_bias_to_fp16 = const()[name = string("module_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231224000)))];
            tensor<fp16, [1, 188, 1024]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = module_layers_16_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_conv_weight_to_fp16, x = input_871_cast_fp16)[name = string("x_369_cast_fp16")];
            tensor<int32, [3]> input_873_perm_0 = const()[name = string("input_873_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_875_pad_type_0 = const()[name = string("input_875_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_875_strides_0 = const()[name = string("input_875_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_875_pad_0 = const()[name = string("input_875_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_875_dilations_0 = const()[name = string("input_875_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_875_groups_0 = const()[name = string("input_875_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_16_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(231226112))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232274752))))[name = string("module_layers_16_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_16_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_16_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232405888)))];
            tensor<fp16, [1, 1024, 188]> input_873_cast_fp16 = transpose(perm = input_873_perm_0, x = x_369_cast_fp16)[name = string("transpose_299")];
            tensor<fp16, [1, 2048, 188]> input_875_cast_fp16 = conv(bias = module_layers_16_conv_pointwise_conv1_bias_to_fp16, dilations = input_875_dilations_0, groups = input_875_groups_0, pad = input_875_pad_0, pad_type = input_875_pad_type_0, strides = input_875_strides_0, weight = module_layers_16_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_873_cast_fp16)[name = string("input_875_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, 188]> x_371_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_875_cast_fp16)[name = string("x_371_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_877_cast_fp16 = select(a = var_11_to_fp16, b = x_371_cast_fp16, cond = var_483)[name = string("input_877_cast_fp16")];
            tensor<int32, [6]> input_879_pad_0 = const()[name = string("input_879_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_879_mode_0 = const()[name = string("input_879_mode_0"), val = string("constant")];
            fp16 const_233_to_fp16 = const()[name = string("const_233_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_879_cast_fp16 = pad(constant_val = const_233_to_fp16, mode = input_879_mode_0, pad = input_879_pad_0, x = input_877_cast_fp16)[name = string("input_879_cast_fp16")];
            string input_881_pad_type_0 = const()[name = string("input_881_pad_type_0"), val = string("valid")];
            int32 input_881_groups_0 = const()[name = string("input_881_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_881_strides_0 = const()[name = string("input_881_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_881_pad_0 = const()[name = string("input_881_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_881_dilations_0 = const()[name = string("input_881_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_416_to_fp16 = const()[name = string("const_416_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232410048)))];
            tensor<fp16, [1024]> const_417_to_fp16 = const()[name = string("const_417_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232428544)))];
            tensor<fp16, [1, 1024, 188]> input_883_cast_fp16 = conv(bias = const_417_to_fp16, dilations = input_881_dilations_0, groups = input_881_groups_0, pad = input_881_pad_0, pad_type = input_881_pad_type_0, strides = input_881_strides_0, weight = const_416_to_fp16, x = input_879_cast_fp16)[name = string("input_883_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_885_cast_fp16 = silu(x = input_883_cast_fp16)[name = string("input_885_cast_fp16")];
            string x_373_pad_type_0 = const()[name = string("x_373_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_373_strides_0 = const()[name = string("x_373_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_373_pad_0 = const()[name = string("x_373_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_373_dilations_0 = const()[name = string("x_373_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_373_groups_0 = const()[name = string("x_373_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_16_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232430656))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(232955008))))[name = string("module_layers_16_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_16_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_16_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233020608)))];
            tensor<fp16, [1, 1024, 188]> x_373_cast_fp16 = conv(bias = module_layers_16_conv_pointwise_conv2_bias_to_fp16, dilations = x_373_dilations_0, groups = x_373_groups_0, pad = x_373_pad_0, pad_type = x_373_pad_type_0, strides = x_373_strides_0, weight = module_layers_16_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_885_cast_fp16)[name = string("x_373_cast_fp16")];
            tensor<int32, [3]> input_887_perm_0 = const()[name = string("input_887_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_887_cast_fp16 = transpose(perm = input_887_perm_0, x = x_373_cast_fp16)[name = string("transpose_298")];
            tensor<fp16, [1, 188, 1024]> input_889_cast_fp16 = add(x = input_871_cast_fp16, y = input_887_cast_fp16)[name = string("input_889_cast_fp16")];
            tensor<int32, [1]> input_891_axes_0 = const()[name = string("input_891_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233022720)))];
            tensor<fp16, [1024]> module_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233024832)))];
            tensor<fp16, [1, 188, 1024]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = module_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = string("input_891_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_16_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(233026944))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235124160))))[name = string("module_layers_16_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_16_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_16_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235386368)))];
            tensor<fp16, [1, 188, 4096]> linear_152_cast_fp16 = linear(bias = module_layers_16_feed_forward2_linear1_bias_to_fp16, weight = module_layers_16_feed_forward2_linear1_weight_to_fp16_quantized, x = input_891_cast_fp16)[name = string("linear_152_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_895_cast_fp16 = silu(x = linear_152_cast_fp16)[name = string("input_895_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_16_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235394624))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237491840))))[name = string("module_layers_16_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_16_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_16_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237754048)))];
            tensor<fp16, [1, 188, 1024]> linear_153_cast_fp16 = linear(bias = module_layers_16_feed_forward2_linear2_bias_to_fp16, weight = module_layers_16_feed_forward2_linear2_weight_to_fp16_quantized, x = input_895_cast_fp16)[name = string("linear_153_cast_fp16")];
            fp16 var_3309_to_fp16 = const()[name = string("op_3309_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3310_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_3309_to_fp16)[name = string("op_3310_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_901_cast_fp16 = add(x = input_889_cast_fp16, y = var_3310_cast_fp16)[name = string("input_901_cast_fp16")];
            tensor<int32, [1]> input_903_axes_0 = const()[name = string("input_903_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_16_norm_out_weight_to_fp16 = const()[name = string("module_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237756160)))];
            tensor<fp16, [1024]> module_layers_16_norm_out_bias_to_fp16 = const()[name = string("module_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237758272)))];
            tensor<fp16, [1, 188, 1024]> input_903_cast_fp16 = layer_norm(axes = input_903_axes_0, beta = module_layers_16_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_16_norm_out_weight_to_fp16, x = input_901_cast_fp16)[name = string("input_903_cast_fp16")];
            tensor<int32, [1]> input_905_axes_0 = const()[name = string("input_905_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237760384)))];
            tensor<fp16, [1024]> module_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237762496)))];
            tensor<fp16, [1, 188, 1024]> input_905_cast_fp16 = layer_norm(axes = input_905_axes_0, beta = module_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward1_weight_to_fp16, x = input_903_cast_fp16)[name = string("input_905_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_17_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237764608))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239861824))))[name = string("module_layers_17_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_17_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_17_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240124032)))];
            tensor<fp16, [1, 188, 4096]> linear_154_cast_fp16 = linear(bias = module_layers_17_feed_forward1_linear1_bias_to_fp16, weight = module_layers_17_feed_forward1_linear1_weight_to_fp16_quantized, x = input_905_cast_fp16)[name = string("linear_154_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_909_cast_fp16 = silu(x = linear_154_cast_fp16)[name = string("input_909_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_17_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240132288))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242229504))))[name = string("module_layers_17_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_17_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_17_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242491712)))];
            tensor<fp16, [1, 188, 1024]> linear_155_cast_fp16 = linear(bias = module_layers_17_feed_forward1_linear2_bias_to_fp16, weight = module_layers_17_feed_forward1_linear2_weight_to_fp16_quantized, x = input_909_cast_fp16)[name = string("linear_155_cast_fp16")];
            fp16 var_3340_to_fp16 = const()[name = string("op_3340_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3341_cast_fp16 = mul(x = linear_155_cast_fp16, y = var_3340_to_fp16)[name = string("op_3341_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_915_cast_fp16 = add(x = input_903_cast_fp16, y = var_3341_cast_fp16)[name = string("input_915_cast_fp16")];
            tensor<int32, [1]> query_35_axes_0 = const()[name = string("query_35_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_17_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242493824)))];
            tensor<fp16, [1024]> module_layers_17_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242495936)))];
            tensor<fp16, [1, 188, 1024]> query_35_cast_fp16 = layer_norm(axes = query_35_axes_0, beta = module_layers_17_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_self_att_weight_to_fp16, x = input_915_cast_fp16)[name = string("query_35_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242498048))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243022400))))[name = string("module_layers_17_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_17_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_17_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243088000)))];
            tensor<fp16, [1, 188, 1024]> linear_156_cast_fp16 = linear(bias = module_layers_17_self_attn_linear_q_bias_to_fp16, weight = module_layers_17_self_attn_linear_q_weight_to_fp16_quantized, x = query_35_cast_fp16)[name = string("linear_156_cast_fp16")];
            tensor<int32, [4]> var_3358 = const()[name = string("op_3358"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_103_cast_fp16 = reshape(shape = var_3358, x = linear_156_cast_fp16)[name = string("q_103_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243090112))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243614464))))[name = string("module_layers_17_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_17_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_17_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243680064)))];
            tensor<fp16, [1, 188, 1024]> linear_157_cast_fp16 = linear(bias = module_layers_17_self_attn_linear_k_bias_to_fp16, weight = module_layers_17_self_attn_linear_k_weight_to_fp16_quantized, x = query_35_cast_fp16)[name = string("linear_157_cast_fp16")];
            tensor<int32, [4]> var_3363 = const()[name = string("op_3363"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_69_cast_fp16 = reshape(shape = var_3363, x = linear_157_cast_fp16)[name = string("k_69_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243682176))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244206528))))[name = string("module_layers_17_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_17_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_17_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244272128)))];
            tensor<fp16, [1, 188, 1024]> linear_158_cast_fp16 = linear(bias = module_layers_17_self_attn_linear_v_bias_to_fp16, weight = module_layers_17_self_attn_linear_v_weight_to_fp16_quantized, x = query_35_cast_fp16)[name = string("linear_158_cast_fp16")];
            tensor<int32, [4]> var_3368 = const()[name = string("op_3368"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_35_cast_fp16 = reshape(shape = var_3368, x = linear_158_cast_fp16)[name = string("v_35_cast_fp16")];
            tensor<int32, [4]> value_37_perm_0 = const()[name = string("value_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244274240))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244274816))))[name = string("module_layers_17_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3380_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_3380_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_17_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244274944))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244275520))))[name = string("module_layers_17_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3382_cast_fp16 = add(x = q_103_cast_fp16, y = module_layers_17_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_3382_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_35_perm_0 = const()[name = string("q_with_bias_v_35_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_381_transpose_x_0 = const()[name = string("x_381_transpose_x_0"), val = bool(false)];
            bool x_381_transpose_y_0 = const()[name = string("x_381_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_3384_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244275648))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244467712))))[name = string("op_3384_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_35_cast_fp16 = transpose(perm = q_with_bias_v_35_perm_0, x = var_3382_cast_fp16)[name = string("transpose_297")];
            tensor<fp16, [1, 8, 188, 375]> x_381_cast_fp16 = matmul(transpose_x = x_381_transpose_x_0, transpose_y = x_381_transpose_y_0, x = q_with_bias_v_35_cast_fp16, y = op_3384_to_fp16_quantized)[name = string("x_381_cast_fp16")];
            tensor<int32, [8]> x_383_pad_0 = const()[name = string("x_383_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_383_mode_0 = const()[name = string("x_383_mode_0"), val = string("constant")];
            fp16 const_240_to_fp16 = const()[name = string("const_240_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_383_cast_fp16 = pad(constant_val = const_240_to_fp16, mode = x_383_mode_0, pad = x_383_pad_0, x = x_381_cast_fp16)[name = string("x_383_cast_fp16")];
            tensor<int32, [4]> var_3392 = const()[name = string("op_3392"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_385_cast_fp16 = reshape(shape = var_3392, x = x_383_cast_fp16)[name = string("x_385_cast_fp16")];
            tensor<int32, [4]> var_3396_begin_0 = const()[name = string("op_3396_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3396_end_0 = const()[name = string("op_3396_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3396_end_mask_0 = const()[name = string("op_3396_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3396_cast_fp16 = slice_by_index(begin = var_3396_begin_0, end = var_3396_end_0, end_mask = var_3396_end_mask_0, x = x_385_cast_fp16)[name = string("op_3396_cast_fp16")];
            tensor<int32, [4]> var_3397 = const()[name = string("op_3397"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_69_cast_fp16 = reshape(shape = var_3397, x = var_3396_cast_fp16)[name = string("matrix_bd_69_cast_fp16")];
            bool matrix_ac_35_transpose_x_0 = const()[name = string("matrix_ac_35_transpose_x_0"), val = bool(false)];
            bool matrix_ac_35_transpose_y_0 = const()[name = string("matrix_ac_35_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_162_perm_0 = const()[name = string("transpose_162_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_163_perm_0 = const()[name = string("transpose_163_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_163 = transpose(perm = transpose_163_perm_0, x = k_69_cast_fp16)[name = string("transpose_295")];
            tensor<fp16, [1, 8, 188, 128]> transpose_162 = transpose(perm = transpose_162_perm_0, x = var_3380_cast_fp16)[name = string("transpose_296")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_35_cast_fp16 = matmul(transpose_x = matrix_ac_35_transpose_x_0, transpose_y = matrix_ac_35_transpose_y_0, x = transpose_162, y = transpose_163)[name = string("matrix_ac_35_cast_fp16")];
            tensor<int32, [4]> matrix_bd_71_begin_0 = const()[name = string("matrix_bd_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_71_end_0 = const()[name = string("matrix_bd_71_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_71_end_mask_0 = const()[name = string("matrix_bd_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_71_cast_fp16 = slice_by_index(begin = matrix_bd_71_begin_0, end = matrix_bd_71_end_0, end_mask = matrix_bd_71_end_mask_0, x = matrix_bd_69_cast_fp16)[name = string("matrix_bd_71_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3406_cast_fp16 = add(x = matrix_ac_35_cast_fp16, y = matrix_bd_71_cast_fp16)[name = string("op_3406_cast_fp16")];
            fp16 _inversed_scores_69_y_0_to_fp16 = const()[name = string("_inversed_scores_69_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_69_cast_fp16 = mul(x = var_3406_cast_fp16, y = _inversed_scores_69_y_0_to_fp16)[name = string("_inversed_scores_69_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_71_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_69_cast_fp16, cond = mask_11)[name = string("scores_71_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3412_cast_fp16 = softmax(axis = var_23, x = scores_71_cast_fp16)[name = string("op_3412_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_917_cast_fp16 = select(a = var_11_to_fp16, b = var_3412_cast_fp16, cond = mask_11)[name = string("input_917_cast_fp16")];
            bool x_387_transpose_x_0 = const()[name = string("x_387_transpose_x_0"), val = bool(false)];
            bool x_387_transpose_y_0 = const()[name = string("x_387_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_37_cast_fp16 = transpose(perm = value_37_perm_0, x = v_35_cast_fp16)[name = string("transpose_294")];
            tensor<fp16, [1, 8, 188, 128]> x_387_cast_fp16 = matmul(transpose_x = x_387_transpose_x_0, transpose_y = x_387_transpose_y_0, x = input_917_cast_fp16, y = value_37_cast_fp16)[name = string("x_387_cast_fp16")];
            tensor<int32, [4]> var_3416_perm_0 = const()[name = string("op_3416_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3417 = const()[name = string("op_3417"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3416_cast_fp16 = transpose(perm = var_3416_perm_0, x = x_387_cast_fp16)[name = string("transpose_293")];
            tensor<fp16, [1, 188, 1024]> input_919_cast_fp16 = reshape(shape = var_3417, x = var_3416_cast_fp16)[name = string("input_919_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_17_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244470784))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244995136))))[name = string("module_layers_17_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_17_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_17_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245060736)))];
            tensor<fp16, [1, 188, 1024]> linear_160_cast_fp16 = linear(bias = module_layers_17_self_attn_linear_out_bias_to_fp16, weight = module_layers_17_self_attn_linear_out_weight_to_fp16_quantized, x = input_919_cast_fp16)[name = string("linear_160_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_923_cast_fp16 = add(x = input_915_cast_fp16, y = linear_160_cast_fp16)[name = string("input_923_cast_fp16")];
            tensor<int32, [1]> x_391_axes_0 = const()[name = string("x_391_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_17_norm_conv_weight_to_fp16 = const()[name = string("module_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245062848)))];
            tensor<fp16, [1024]> module_layers_17_norm_conv_bias_to_fp16 = const()[name = string("module_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245064960)))];
            tensor<fp16, [1, 188, 1024]> x_391_cast_fp16 = layer_norm(axes = x_391_axes_0, beta = module_layers_17_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_conv_weight_to_fp16, x = input_923_cast_fp16)[name = string("x_391_cast_fp16")];
            tensor<int32, [3]> input_925_perm_0 = const()[name = string("input_925_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_927_pad_type_0 = const()[name = string("input_927_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_927_strides_0 = const()[name = string("input_927_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_927_pad_0 = const()[name = string("input_927_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_927_dilations_0 = const()[name = string("input_927_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_927_groups_0 = const()[name = string("input_927_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_17_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245067072))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246115712))))[name = string("module_layers_17_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_17_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_17_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246246848)))];
            tensor<fp16, [1, 1024, 188]> input_925_cast_fp16 = transpose(perm = input_925_perm_0, x = x_391_cast_fp16)[name = string("transpose_292")];
            tensor<fp16, [1, 2048, 188]> input_927_cast_fp16 = conv(bias = module_layers_17_conv_pointwise_conv1_bias_to_fp16, dilations = input_927_dilations_0, groups = input_927_groups_0, pad = input_927_pad_0, pad_type = input_927_pad_type_0, strides = input_927_strides_0, weight = module_layers_17_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_925_cast_fp16)[name = string("input_927_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, 188]> x_393_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_393_split_cast_fp16_1 = split(axis = x_393_split_axis_0, num_splits = x_393_split_num_splits_0, x = input_927_cast_fp16)[name = string("x_393_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_929_cast_fp16 = select(a = var_11_to_fp16, b = x_393_cast_fp16, cond = var_483)[name = string("input_929_cast_fp16")];
            tensor<int32, [6]> input_931_pad_0 = const()[name = string("input_931_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_931_mode_0 = const()[name = string("input_931_mode_0"), val = string("constant")];
            fp16 const_243_to_fp16 = const()[name = string("const_243_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_931_cast_fp16 = pad(constant_val = const_243_to_fp16, mode = input_931_mode_0, pad = input_931_pad_0, x = input_929_cast_fp16)[name = string("input_931_cast_fp16")];
            string input_933_pad_type_0 = const()[name = string("input_933_pad_type_0"), val = string("valid")];
            int32 input_933_groups_0 = const()[name = string("input_933_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_933_strides_0 = const()[name = string("input_933_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_933_pad_0 = const()[name = string("input_933_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_933_dilations_0 = const()[name = string("input_933_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_418_to_fp16 = const()[name = string("const_418_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246251008)))];
            tensor<fp16, [1024]> const_419_to_fp16 = const()[name = string("const_419_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246269504)))];
            tensor<fp16, [1, 1024, 188]> input_935_cast_fp16 = conv(bias = const_419_to_fp16, dilations = input_933_dilations_0, groups = input_933_groups_0, pad = input_933_pad_0, pad_type = input_933_pad_type_0, strides = input_933_strides_0, weight = const_418_to_fp16, x = input_931_cast_fp16)[name = string("input_935_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_937_cast_fp16 = silu(x = input_935_cast_fp16)[name = string("input_937_cast_fp16")];
            string x_395_pad_type_0 = const()[name = string("x_395_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_395_strides_0 = const()[name = string("x_395_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_395_pad_0 = const()[name = string("x_395_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_395_dilations_0 = const()[name = string("x_395_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_395_groups_0 = const()[name = string("x_395_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_17_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246271616))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246795968))))[name = string("module_layers_17_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_17_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_17_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246861568)))];
            tensor<fp16, [1, 1024, 188]> x_395_cast_fp16 = conv(bias = module_layers_17_conv_pointwise_conv2_bias_to_fp16, dilations = x_395_dilations_0, groups = x_395_groups_0, pad = x_395_pad_0, pad_type = x_395_pad_type_0, strides = x_395_strides_0, weight = module_layers_17_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_937_cast_fp16)[name = string("x_395_cast_fp16")];
            tensor<int32, [3]> input_939_perm_0 = const()[name = string("input_939_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_939_cast_fp16 = transpose(perm = input_939_perm_0, x = x_395_cast_fp16)[name = string("transpose_291")];
            tensor<fp16, [1, 188, 1024]> input_941_cast_fp16 = add(x = input_923_cast_fp16, y = input_939_cast_fp16)[name = string("input_941_cast_fp16")];
            tensor<int32, [1]> input_943_axes_0 = const()[name = string("input_943_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246863680)))];
            tensor<fp16, [1024]> module_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246865792)))];
            tensor<fp16, [1, 188, 1024]> input_943_cast_fp16 = layer_norm(axes = input_943_axes_0, beta = module_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_feed_forward2_weight_to_fp16, x = input_941_cast_fp16)[name = string("input_943_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_17_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246867904))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(248965120))))[name = string("module_layers_17_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_17_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_17_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249227328)))];
            tensor<fp16, [1, 188, 4096]> linear_161_cast_fp16 = linear(bias = module_layers_17_feed_forward2_linear1_bias_to_fp16, weight = module_layers_17_feed_forward2_linear1_weight_to_fp16_quantized, x = input_943_cast_fp16)[name = string("linear_161_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_947_cast_fp16 = silu(x = linear_161_cast_fp16)[name = string("input_947_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_17_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(249235584))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251332800))))[name = string("module_layers_17_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_17_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_17_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251595008)))];
            tensor<fp16, [1, 188, 1024]> linear_162_cast_fp16 = linear(bias = module_layers_17_feed_forward2_linear2_bias_to_fp16, weight = module_layers_17_feed_forward2_linear2_weight_to_fp16_quantized, x = input_947_cast_fp16)[name = string("linear_162_cast_fp16")];
            fp16 var_3483_to_fp16 = const()[name = string("op_3483_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3484_cast_fp16 = mul(x = linear_162_cast_fp16, y = var_3483_to_fp16)[name = string("op_3484_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_953_cast_fp16 = add(x = input_941_cast_fp16, y = var_3484_cast_fp16)[name = string("input_953_cast_fp16")];
            tensor<int32, [1]> input_955_axes_0 = const()[name = string("input_955_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_17_norm_out_weight_to_fp16 = const()[name = string("module_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251597120)))];
            tensor<fp16, [1024]> module_layers_17_norm_out_bias_to_fp16 = const()[name = string("module_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251599232)))];
            tensor<fp16, [1, 188, 1024]> input_955_cast_fp16 = layer_norm(axes = input_955_axes_0, beta = module_layers_17_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_17_norm_out_weight_to_fp16, x = input_953_cast_fp16)[name = string("input_955_cast_fp16")];
            tensor<int32, [1]> input_957_axes_0 = const()[name = string("input_957_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251601344)))];
            tensor<fp16, [1024]> module_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251603456)))];
            tensor<fp16, [1, 188, 1024]> input_957_cast_fp16 = layer_norm(axes = input_957_axes_0, beta = module_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward1_weight_to_fp16, x = input_955_cast_fp16)[name = string("input_957_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_18_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(251605568))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253702784))))[name = string("module_layers_18_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_18_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_18_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253964992)))];
            tensor<fp16, [1, 188, 4096]> linear_163_cast_fp16 = linear(bias = module_layers_18_feed_forward1_linear1_bias_to_fp16, weight = module_layers_18_feed_forward1_linear1_weight_to_fp16_quantized, x = input_957_cast_fp16)[name = string("linear_163_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_961_cast_fp16 = silu(x = linear_163_cast_fp16)[name = string("input_961_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_18_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253973248))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256070464))))[name = string("module_layers_18_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_18_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_18_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256332672)))];
            tensor<fp16, [1, 188, 1024]> linear_164_cast_fp16 = linear(bias = module_layers_18_feed_forward1_linear2_bias_to_fp16, weight = module_layers_18_feed_forward1_linear2_weight_to_fp16_quantized, x = input_961_cast_fp16)[name = string("linear_164_cast_fp16")];
            fp16 var_3514_to_fp16 = const()[name = string("op_3514_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3515_cast_fp16 = mul(x = linear_164_cast_fp16, y = var_3514_to_fp16)[name = string("op_3515_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_967_cast_fp16 = add(x = input_955_cast_fp16, y = var_3515_cast_fp16)[name = string("input_967_cast_fp16")];
            tensor<int32, [1]> query_37_axes_0 = const()[name = string("query_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_18_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256334784)))];
            tensor<fp16, [1024]> module_layers_18_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256336896)))];
            tensor<fp16, [1, 188, 1024]> query_37_cast_fp16 = layer_norm(axes = query_37_axes_0, beta = module_layers_18_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_self_att_weight_to_fp16, x = input_967_cast_fp16)[name = string("query_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256339008))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256863360))))[name = string("module_layers_18_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_18_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_18_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256928960)))];
            tensor<fp16, [1, 188, 1024]> linear_165_cast_fp16 = linear(bias = module_layers_18_self_attn_linear_q_bias_to_fp16, weight = module_layers_18_self_attn_linear_q_weight_to_fp16_quantized, x = query_37_cast_fp16)[name = string("linear_165_cast_fp16")];
            tensor<int32, [4]> var_3532 = const()[name = string("op_3532"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_109_cast_fp16 = reshape(shape = var_3532, x = linear_165_cast_fp16)[name = string("q_109_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256931072))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257455424))))[name = string("module_layers_18_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_18_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_18_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257521024)))];
            tensor<fp16, [1, 188, 1024]> linear_166_cast_fp16 = linear(bias = module_layers_18_self_attn_linear_k_bias_to_fp16, weight = module_layers_18_self_attn_linear_k_weight_to_fp16_quantized, x = query_37_cast_fp16)[name = string("linear_166_cast_fp16")];
            tensor<int32, [4]> var_3537 = const()[name = string("op_3537"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_73_cast_fp16 = reshape(shape = var_3537, x = linear_166_cast_fp16)[name = string("k_73_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(257523136))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258047488))))[name = string("module_layers_18_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_18_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_18_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258113088)))];
            tensor<fp16, [1, 188, 1024]> linear_167_cast_fp16 = linear(bias = module_layers_18_self_attn_linear_v_bias_to_fp16, weight = module_layers_18_self_attn_linear_v_weight_to_fp16_quantized, x = query_37_cast_fp16)[name = string("linear_167_cast_fp16")];
            tensor<int32, [4]> var_3542 = const()[name = string("op_3542"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_37_cast_fp16 = reshape(shape = var_3542, x = linear_167_cast_fp16)[name = string("v_37_cast_fp16")];
            tensor<int32, [4]> value_39_perm_0 = const()[name = string("value_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258115200))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258115776))))[name = string("module_layers_18_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3554_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_3554_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_18_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258115904))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258116480))))[name = string("module_layers_18_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3556_cast_fp16 = add(x = q_109_cast_fp16, y = module_layers_18_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_3556_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_37_perm_0 = const()[name = string("q_with_bias_v_37_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_403_transpose_x_0 = const()[name = string("x_403_transpose_x_0"), val = bool(false)];
            bool x_403_transpose_y_0 = const()[name = string("x_403_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_3558_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258116608))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258308672))))[name = string("op_3558_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_37_cast_fp16 = transpose(perm = q_with_bias_v_37_perm_0, x = var_3556_cast_fp16)[name = string("transpose_290")];
            tensor<fp16, [1, 8, 188, 375]> x_403_cast_fp16 = matmul(transpose_x = x_403_transpose_x_0, transpose_y = x_403_transpose_y_0, x = q_with_bias_v_37_cast_fp16, y = op_3558_to_fp16_quantized)[name = string("x_403_cast_fp16")];
            tensor<int32, [8]> x_405_pad_0 = const()[name = string("x_405_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_405_mode_0 = const()[name = string("x_405_mode_0"), val = string("constant")];
            fp16 const_250_to_fp16 = const()[name = string("const_250_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_405_cast_fp16 = pad(constant_val = const_250_to_fp16, mode = x_405_mode_0, pad = x_405_pad_0, x = x_403_cast_fp16)[name = string("x_405_cast_fp16")];
            tensor<int32, [4]> var_3566 = const()[name = string("op_3566"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_407_cast_fp16 = reshape(shape = var_3566, x = x_405_cast_fp16)[name = string("x_407_cast_fp16")];
            tensor<int32, [4]> var_3570_begin_0 = const()[name = string("op_3570_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3570_end_0 = const()[name = string("op_3570_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3570_end_mask_0 = const()[name = string("op_3570_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3570_cast_fp16 = slice_by_index(begin = var_3570_begin_0, end = var_3570_end_0, end_mask = var_3570_end_mask_0, x = x_407_cast_fp16)[name = string("op_3570_cast_fp16")];
            tensor<int32, [4]> var_3571 = const()[name = string("op_3571"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_73_cast_fp16 = reshape(shape = var_3571, x = var_3570_cast_fp16)[name = string("matrix_bd_73_cast_fp16")];
            bool matrix_ac_37_transpose_x_0 = const()[name = string("matrix_ac_37_transpose_x_0"), val = bool(false)];
            bool matrix_ac_37_transpose_y_0 = const()[name = string("matrix_ac_37_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_164_perm_0 = const()[name = string("transpose_164_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_165_perm_0 = const()[name = string("transpose_165_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_165 = transpose(perm = transpose_165_perm_0, x = k_73_cast_fp16)[name = string("transpose_288")];
            tensor<fp16, [1, 8, 188, 128]> transpose_164 = transpose(perm = transpose_164_perm_0, x = var_3554_cast_fp16)[name = string("transpose_289")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_37_cast_fp16 = matmul(transpose_x = matrix_ac_37_transpose_x_0, transpose_y = matrix_ac_37_transpose_y_0, x = transpose_164, y = transpose_165)[name = string("matrix_ac_37_cast_fp16")];
            tensor<int32, [4]> matrix_bd_75_begin_0 = const()[name = string("matrix_bd_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_75_end_0 = const()[name = string("matrix_bd_75_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_75_end_mask_0 = const()[name = string("matrix_bd_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_75_cast_fp16 = slice_by_index(begin = matrix_bd_75_begin_0, end = matrix_bd_75_end_0, end_mask = matrix_bd_75_end_mask_0, x = matrix_bd_73_cast_fp16)[name = string("matrix_bd_75_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3580_cast_fp16 = add(x = matrix_ac_37_cast_fp16, y = matrix_bd_75_cast_fp16)[name = string("op_3580_cast_fp16")];
            fp16 _inversed_scores_73_y_0_to_fp16 = const()[name = string("_inversed_scores_73_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_73_cast_fp16 = mul(x = var_3580_cast_fp16, y = _inversed_scores_73_y_0_to_fp16)[name = string("_inversed_scores_73_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_75_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_73_cast_fp16, cond = mask_11)[name = string("scores_75_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3586_cast_fp16 = softmax(axis = var_23, x = scores_75_cast_fp16)[name = string("op_3586_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_969_cast_fp16 = select(a = var_11_to_fp16, b = var_3586_cast_fp16, cond = mask_11)[name = string("input_969_cast_fp16")];
            bool x_409_transpose_x_0 = const()[name = string("x_409_transpose_x_0"), val = bool(false)];
            bool x_409_transpose_y_0 = const()[name = string("x_409_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_39_cast_fp16 = transpose(perm = value_39_perm_0, x = v_37_cast_fp16)[name = string("transpose_287")];
            tensor<fp16, [1, 8, 188, 128]> x_409_cast_fp16 = matmul(transpose_x = x_409_transpose_x_0, transpose_y = x_409_transpose_y_0, x = input_969_cast_fp16, y = value_39_cast_fp16)[name = string("x_409_cast_fp16")];
            tensor<int32, [4]> var_3590_perm_0 = const()[name = string("op_3590_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3591 = const()[name = string("op_3591"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3590_cast_fp16 = transpose(perm = var_3590_perm_0, x = x_409_cast_fp16)[name = string("transpose_286")];
            tensor<fp16, [1, 188, 1024]> input_971_cast_fp16 = reshape(shape = var_3591, x = var_3590_cast_fp16)[name = string("input_971_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_18_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258311744))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258836096))))[name = string("module_layers_18_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_18_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_18_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258901696)))];
            tensor<fp16, [1, 188, 1024]> linear_169_cast_fp16 = linear(bias = module_layers_18_self_attn_linear_out_bias_to_fp16, weight = module_layers_18_self_attn_linear_out_weight_to_fp16_quantized, x = input_971_cast_fp16)[name = string("linear_169_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_975_cast_fp16 = add(x = input_967_cast_fp16, y = linear_169_cast_fp16)[name = string("input_975_cast_fp16")];
            tensor<int32, [1]> x_413_axes_0 = const()[name = string("x_413_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_18_norm_conv_weight_to_fp16 = const()[name = string("module_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258903808)))];
            tensor<fp16, [1024]> module_layers_18_norm_conv_bias_to_fp16 = const()[name = string("module_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258905920)))];
            tensor<fp16, [1, 188, 1024]> x_413_cast_fp16 = layer_norm(axes = x_413_axes_0, beta = module_layers_18_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_conv_weight_to_fp16, x = input_975_cast_fp16)[name = string("x_413_cast_fp16")];
            tensor<int32, [3]> input_977_perm_0 = const()[name = string("input_977_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_979_pad_type_0 = const()[name = string("input_979_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_979_strides_0 = const()[name = string("input_979_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_979_pad_0 = const()[name = string("input_979_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_979_dilations_0 = const()[name = string("input_979_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_979_groups_0 = const()[name = string("input_979_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_18_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258908032))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(259956672))))[name = string("module_layers_18_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_18_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_18_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260087808)))];
            tensor<fp16, [1, 1024, 188]> input_977_cast_fp16 = transpose(perm = input_977_perm_0, x = x_413_cast_fp16)[name = string("transpose_285")];
            tensor<fp16, [1, 2048, 188]> input_979_cast_fp16 = conv(bias = module_layers_18_conv_pointwise_conv1_bias_to_fp16, dilations = input_979_dilations_0, groups = input_979_groups_0, pad = input_979_pad_0, pad_type = input_979_pad_type_0, strides = input_979_strides_0, weight = module_layers_18_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_977_cast_fp16)[name = string("input_979_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, 188]> x_415_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_415_split_cast_fp16_1 = split(axis = x_415_split_axis_0, num_splits = x_415_split_num_splits_0, x = input_979_cast_fp16)[name = string("x_415_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_981_cast_fp16 = select(a = var_11_to_fp16, b = x_415_cast_fp16, cond = var_483)[name = string("input_981_cast_fp16")];
            tensor<int32, [6]> input_983_pad_0 = const()[name = string("input_983_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_983_mode_0 = const()[name = string("input_983_mode_0"), val = string("constant")];
            fp16 const_253_to_fp16 = const()[name = string("const_253_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_983_cast_fp16 = pad(constant_val = const_253_to_fp16, mode = input_983_mode_0, pad = input_983_pad_0, x = input_981_cast_fp16)[name = string("input_983_cast_fp16")];
            string input_985_pad_type_0 = const()[name = string("input_985_pad_type_0"), val = string("valid")];
            int32 input_985_groups_0 = const()[name = string("input_985_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_985_strides_0 = const()[name = string("input_985_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_985_pad_0 = const()[name = string("input_985_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_985_dilations_0 = const()[name = string("input_985_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_420_to_fp16 = const()[name = string("const_420_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260091968)))];
            tensor<fp16, [1024]> const_421_to_fp16 = const()[name = string("const_421_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260110464)))];
            tensor<fp16, [1, 1024, 188]> input_987_cast_fp16 = conv(bias = const_421_to_fp16, dilations = input_985_dilations_0, groups = input_985_groups_0, pad = input_985_pad_0, pad_type = input_985_pad_type_0, strides = input_985_strides_0, weight = const_420_to_fp16, x = input_983_cast_fp16)[name = string("input_987_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_989_cast_fp16 = silu(x = input_987_cast_fp16)[name = string("input_989_cast_fp16")];
            string x_417_pad_type_0 = const()[name = string("x_417_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_417_strides_0 = const()[name = string("x_417_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_417_pad_0 = const()[name = string("x_417_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_417_dilations_0 = const()[name = string("x_417_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_417_groups_0 = const()[name = string("x_417_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_18_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260112576))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260636928))))[name = string("module_layers_18_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_18_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_18_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260702528)))];
            tensor<fp16, [1, 1024, 188]> x_417_cast_fp16 = conv(bias = module_layers_18_conv_pointwise_conv2_bias_to_fp16, dilations = x_417_dilations_0, groups = x_417_groups_0, pad = x_417_pad_0, pad_type = x_417_pad_type_0, strides = x_417_strides_0, weight = module_layers_18_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_989_cast_fp16)[name = string("x_417_cast_fp16")];
            tensor<int32, [3]> input_991_perm_0 = const()[name = string("input_991_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_991_cast_fp16 = transpose(perm = input_991_perm_0, x = x_417_cast_fp16)[name = string("transpose_284")];
            tensor<fp16, [1, 188, 1024]> input_993_cast_fp16 = add(x = input_975_cast_fp16, y = input_991_cast_fp16)[name = string("input_993_cast_fp16")];
            tensor<int32, [1]> input_995_axes_0 = const()[name = string("input_995_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260704640)))];
            tensor<fp16, [1024]> module_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260706752)))];
            tensor<fp16, [1, 188, 1024]> input_995_cast_fp16 = layer_norm(axes = input_995_axes_0, beta = module_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_feed_forward2_weight_to_fp16, x = input_993_cast_fp16)[name = string("input_995_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_18_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260708864))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262806080))))[name = string("module_layers_18_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_18_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_18_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263068288)))];
            tensor<fp16, [1, 188, 4096]> linear_170_cast_fp16 = linear(bias = module_layers_18_feed_forward2_linear1_bias_to_fp16, weight = module_layers_18_feed_forward2_linear1_weight_to_fp16_quantized, x = input_995_cast_fp16)[name = string("linear_170_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_999_cast_fp16 = silu(x = linear_170_cast_fp16)[name = string("input_999_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_18_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263076544))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265173760))))[name = string("module_layers_18_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_18_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_18_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265435968)))];
            tensor<fp16, [1, 188, 1024]> linear_171_cast_fp16 = linear(bias = module_layers_18_feed_forward2_linear2_bias_to_fp16, weight = module_layers_18_feed_forward2_linear2_weight_to_fp16_quantized, x = input_999_cast_fp16)[name = string("linear_171_cast_fp16")];
            fp16 var_3657_to_fp16 = const()[name = string("op_3657_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3658_cast_fp16 = mul(x = linear_171_cast_fp16, y = var_3657_to_fp16)[name = string("op_3658_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1005_cast_fp16 = add(x = input_993_cast_fp16, y = var_3658_cast_fp16)[name = string("input_1005_cast_fp16")];
            tensor<int32, [1]> input_1007_axes_0 = const()[name = string("input_1007_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_18_norm_out_weight_to_fp16 = const()[name = string("module_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265438080)))];
            tensor<fp16, [1024]> module_layers_18_norm_out_bias_to_fp16 = const()[name = string("module_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265440192)))];
            tensor<fp16, [1, 188, 1024]> input_1007_cast_fp16 = layer_norm(axes = input_1007_axes_0, beta = module_layers_18_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_18_norm_out_weight_to_fp16, x = input_1005_cast_fp16)[name = string("input_1007_cast_fp16")];
            tensor<int32, [1]> input_1009_axes_0 = const()[name = string("input_1009_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265442304)))];
            tensor<fp16, [1024]> module_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265444416)))];
            tensor<fp16, [1, 188, 1024]> input_1009_cast_fp16 = layer_norm(axes = input_1009_axes_0, beta = module_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward1_weight_to_fp16, x = input_1007_cast_fp16)[name = string("input_1009_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_19_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265446528))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267543744))))[name = string("module_layers_19_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_19_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_19_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267805952)))];
            tensor<fp16, [1, 188, 4096]> linear_172_cast_fp16 = linear(bias = module_layers_19_feed_forward1_linear1_bias_to_fp16, weight = module_layers_19_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1009_cast_fp16)[name = string("linear_172_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1013_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("input_1013_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_19_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(267814208))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(269911424))))[name = string("module_layers_19_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_19_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_19_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270173632)))];
            tensor<fp16, [1, 188, 1024]> linear_173_cast_fp16 = linear(bias = module_layers_19_feed_forward1_linear2_bias_to_fp16, weight = module_layers_19_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1013_cast_fp16)[name = string("linear_173_cast_fp16")];
            fp16 var_3688_to_fp16 = const()[name = string("op_3688_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3689_cast_fp16 = mul(x = linear_173_cast_fp16, y = var_3688_to_fp16)[name = string("op_3689_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1019_cast_fp16 = add(x = input_1007_cast_fp16, y = var_3689_cast_fp16)[name = string("input_1019_cast_fp16")];
            tensor<int32, [1]> query_39_axes_0 = const()[name = string("query_39_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_19_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270175744)))];
            tensor<fp16, [1024]> module_layers_19_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270177856)))];
            tensor<fp16, [1, 188, 1024]> query_39_cast_fp16 = layer_norm(axes = query_39_axes_0, beta = module_layers_19_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_self_att_weight_to_fp16, x = input_1019_cast_fp16)[name = string("query_39_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270179968))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270704320))))[name = string("module_layers_19_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_19_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_19_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270769920)))];
            tensor<fp16, [1, 188, 1024]> linear_174_cast_fp16 = linear(bias = module_layers_19_self_attn_linear_q_bias_to_fp16, weight = module_layers_19_self_attn_linear_q_weight_to_fp16_quantized, x = query_39_cast_fp16)[name = string("linear_174_cast_fp16")];
            tensor<int32, [4]> var_3706 = const()[name = string("op_3706"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_115_cast_fp16 = reshape(shape = var_3706, x = linear_174_cast_fp16)[name = string("q_115_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270772032))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271296384))))[name = string("module_layers_19_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_19_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_19_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271361984)))];
            tensor<fp16, [1, 188, 1024]> linear_175_cast_fp16 = linear(bias = module_layers_19_self_attn_linear_k_bias_to_fp16, weight = module_layers_19_self_attn_linear_k_weight_to_fp16_quantized, x = query_39_cast_fp16)[name = string("linear_175_cast_fp16")];
            tensor<int32, [4]> var_3711 = const()[name = string("op_3711"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_77_cast_fp16 = reshape(shape = var_3711, x = linear_175_cast_fp16)[name = string("k_77_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271364096))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271888448))))[name = string("module_layers_19_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_19_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_19_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271954048)))];
            tensor<fp16, [1, 188, 1024]> linear_176_cast_fp16 = linear(bias = module_layers_19_self_attn_linear_v_bias_to_fp16, weight = module_layers_19_self_attn_linear_v_weight_to_fp16_quantized, x = query_39_cast_fp16)[name = string("linear_176_cast_fp16")];
            tensor<int32, [4]> var_3716 = const()[name = string("op_3716"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_39_cast_fp16 = reshape(shape = var_3716, x = linear_176_cast_fp16)[name = string("v_39_cast_fp16")];
            tensor<int32, [4]> value_41_perm_0 = const()[name = string("value_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271956160))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271956736))))[name = string("module_layers_19_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3728_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_3728_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_19_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271956864))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271957440))))[name = string("module_layers_19_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3730_cast_fp16 = add(x = q_115_cast_fp16, y = module_layers_19_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_3730_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_39_perm_0 = const()[name = string("q_with_bias_v_39_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_425_transpose_x_0 = const()[name = string("x_425_transpose_x_0"), val = bool(false)];
            bool x_425_transpose_y_0 = const()[name = string("x_425_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_3732_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(271957568))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272149632))))[name = string("op_3732_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_39_cast_fp16 = transpose(perm = q_with_bias_v_39_perm_0, x = var_3730_cast_fp16)[name = string("transpose_283")];
            tensor<fp16, [1, 8, 188, 375]> x_425_cast_fp16 = matmul(transpose_x = x_425_transpose_x_0, transpose_y = x_425_transpose_y_0, x = q_with_bias_v_39_cast_fp16, y = op_3732_to_fp16_quantized)[name = string("x_425_cast_fp16")];
            tensor<int32, [8]> x_427_pad_0 = const()[name = string("x_427_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_427_mode_0 = const()[name = string("x_427_mode_0"), val = string("constant")];
            fp16 const_260_to_fp16 = const()[name = string("const_260_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_427_cast_fp16 = pad(constant_val = const_260_to_fp16, mode = x_427_mode_0, pad = x_427_pad_0, x = x_425_cast_fp16)[name = string("x_427_cast_fp16")];
            tensor<int32, [4]> var_3740 = const()[name = string("op_3740"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_429_cast_fp16 = reshape(shape = var_3740, x = x_427_cast_fp16)[name = string("x_429_cast_fp16")];
            tensor<int32, [4]> var_3744_begin_0 = const()[name = string("op_3744_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3744_end_0 = const()[name = string("op_3744_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3744_end_mask_0 = const()[name = string("op_3744_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3744_cast_fp16 = slice_by_index(begin = var_3744_begin_0, end = var_3744_end_0, end_mask = var_3744_end_mask_0, x = x_429_cast_fp16)[name = string("op_3744_cast_fp16")];
            tensor<int32, [4]> var_3745 = const()[name = string("op_3745"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_77_cast_fp16 = reshape(shape = var_3745, x = var_3744_cast_fp16)[name = string("matrix_bd_77_cast_fp16")];
            bool matrix_ac_39_transpose_x_0 = const()[name = string("matrix_ac_39_transpose_x_0"), val = bool(false)];
            bool matrix_ac_39_transpose_y_0 = const()[name = string("matrix_ac_39_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_166_perm_0 = const()[name = string("transpose_166_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_167_perm_0 = const()[name = string("transpose_167_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_167 = transpose(perm = transpose_167_perm_0, x = k_77_cast_fp16)[name = string("transpose_281")];
            tensor<fp16, [1, 8, 188, 128]> transpose_166 = transpose(perm = transpose_166_perm_0, x = var_3728_cast_fp16)[name = string("transpose_282")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_39_cast_fp16 = matmul(transpose_x = matrix_ac_39_transpose_x_0, transpose_y = matrix_ac_39_transpose_y_0, x = transpose_166, y = transpose_167)[name = string("matrix_ac_39_cast_fp16")];
            tensor<int32, [4]> matrix_bd_79_begin_0 = const()[name = string("matrix_bd_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_79_end_0 = const()[name = string("matrix_bd_79_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_79_end_mask_0 = const()[name = string("matrix_bd_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_79_cast_fp16 = slice_by_index(begin = matrix_bd_79_begin_0, end = matrix_bd_79_end_0, end_mask = matrix_bd_79_end_mask_0, x = matrix_bd_77_cast_fp16)[name = string("matrix_bd_79_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3754_cast_fp16 = add(x = matrix_ac_39_cast_fp16, y = matrix_bd_79_cast_fp16)[name = string("op_3754_cast_fp16")];
            fp16 _inversed_scores_77_y_0_to_fp16 = const()[name = string("_inversed_scores_77_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_77_cast_fp16 = mul(x = var_3754_cast_fp16, y = _inversed_scores_77_y_0_to_fp16)[name = string("_inversed_scores_77_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_79_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_77_cast_fp16, cond = mask_11)[name = string("scores_79_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3760_cast_fp16 = softmax(axis = var_23, x = scores_79_cast_fp16)[name = string("op_3760_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1021_cast_fp16 = select(a = var_11_to_fp16, b = var_3760_cast_fp16, cond = mask_11)[name = string("input_1021_cast_fp16")];
            bool x_431_transpose_x_0 = const()[name = string("x_431_transpose_x_0"), val = bool(false)];
            bool x_431_transpose_y_0 = const()[name = string("x_431_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_41_cast_fp16 = transpose(perm = value_41_perm_0, x = v_39_cast_fp16)[name = string("transpose_280")];
            tensor<fp16, [1, 8, 188, 128]> x_431_cast_fp16 = matmul(transpose_x = x_431_transpose_x_0, transpose_y = x_431_transpose_y_0, x = input_1021_cast_fp16, y = value_41_cast_fp16)[name = string("x_431_cast_fp16")];
            tensor<int32, [4]> var_3764_perm_0 = const()[name = string("op_3764_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3765 = const()[name = string("op_3765"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3764_cast_fp16 = transpose(perm = var_3764_perm_0, x = x_431_cast_fp16)[name = string("transpose_279")];
            tensor<fp16, [1, 188, 1024]> input_1023_cast_fp16 = reshape(shape = var_3765, x = var_3764_cast_fp16)[name = string("input_1023_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_19_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272152704))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272677056))))[name = string("module_layers_19_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_19_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_19_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272742656)))];
            tensor<fp16, [1, 188, 1024]> linear_178_cast_fp16 = linear(bias = module_layers_19_self_attn_linear_out_bias_to_fp16, weight = module_layers_19_self_attn_linear_out_weight_to_fp16_quantized, x = input_1023_cast_fp16)[name = string("linear_178_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1027_cast_fp16 = add(x = input_1019_cast_fp16, y = linear_178_cast_fp16)[name = string("input_1027_cast_fp16")];
            tensor<int32, [1]> x_435_axes_0 = const()[name = string("x_435_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_19_norm_conv_weight_to_fp16 = const()[name = string("module_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272744768)))];
            tensor<fp16, [1024]> module_layers_19_norm_conv_bias_to_fp16 = const()[name = string("module_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272746880)))];
            tensor<fp16, [1, 188, 1024]> x_435_cast_fp16 = layer_norm(axes = x_435_axes_0, beta = module_layers_19_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_conv_weight_to_fp16, x = input_1027_cast_fp16)[name = string("x_435_cast_fp16")];
            tensor<int32, [3]> input_1029_perm_0 = const()[name = string("input_1029_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1031_pad_type_0 = const()[name = string("input_1031_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1031_strides_0 = const()[name = string("input_1031_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1031_pad_0 = const()[name = string("input_1031_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1031_dilations_0 = const()[name = string("input_1031_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1031_groups_0 = const()[name = string("input_1031_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_19_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272748992))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273797632))))[name = string("module_layers_19_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_19_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_19_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273928768)))];
            tensor<fp16, [1, 1024, 188]> input_1029_cast_fp16 = transpose(perm = input_1029_perm_0, x = x_435_cast_fp16)[name = string("transpose_278")];
            tensor<fp16, [1, 2048, 188]> input_1031_cast_fp16 = conv(bias = module_layers_19_conv_pointwise_conv1_bias_to_fp16, dilations = input_1031_dilations_0, groups = input_1031_groups_0, pad = input_1031_pad_0, pad_type = input_1031_pad_type_0, strides = input_1031_strides_0, weight = module_layers_19_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1029_cast_fp16)[name = string("input_1031_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, 188]> x_437_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_437_split_cast_fp16_1 = split(axis = x_437_split_axis_0, num_splits = x_437_split_num_splits_0, x = input_1031_cast_fp16)[name = string("x_437_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_1033_cast_fp16 = select(a = var_11_to_fp16, b = x_437_cast_fp16, cond = var_483)[name = string("input_1033_cast_fp16")];
            tensor<int32, [6]> input_1035_pad_0 = const()[name = string("input_1035_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1035_mode_0 = const()[name = string("input_1035_mode_0"), val = string("constant")];
            fp16 const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1035_cast_fp16 = pad(constant_val = const_263_to_fp16, mode = input_1035_mode_0, pad = input_1035_pad_0, x = input_1033_cast_fp16)[name = string("input_1035_cast_fp16")];
            string input_1037_pad_type_0 = const()[name = string("input_1037_pad_type_0"), val = string("valid")];
            int32 input_1037_groups_0 = const()[name = string("input_1037_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1037_strides_0 = const()[name = string("input_1037_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1037_pad_0 = const()[name = string("input_1037_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1037_dilations_0 = const()[name = string("input_1037_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_422_to_fp16 = const()[name = string("const_422_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273932928)))];
            tensor<fp16, [1024]> const_423_to_fp16 = const()[name = string("const_423_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273951424)))];
            tensor<fp16, [1, 1024, 188]> input_1039_cast_fp16 = conv(bias = const_423_to_fp16, dilations = input_1037_dilations_0, groups = input_1037_groups_0, pad = input_1037_pad_0, pad_type = input_1037_pad_type_0, strides = input_1037_strides_0, weight = const_422_to_fp16, x = input_1035_cast_fp16)[name = string("input_1039_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1041_cast_fp16 = silu(x = input_1039_cast_fp16)[name = string("input_1041_cast_fp16")];
            string x_439_pad_type_0 = const()[name = string("x_439_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_439_strides_0 = const()[name = string("x_439_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_439_pad_0 = const()[name = string("x_439_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_439_dilations_0 = const()[name = string("x_439_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_439_groups_0 = const()[name = string("x_439_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_19_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(273953536))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274477888))))[name = string("module_layers_19_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_19_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_19_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274543488)))];
            tensor<fp16, [1, 1024, 188]> x_439_cast_fp16 = conv(bias = module_layers_19_conv_pointwise_conv2_bias_to_fp16, dilations = x_439_dilations_0, groups = x_439_groups_0, pad = x_439_pad_0, pad_type = x_439_pad_type_0, strides = x_439_strides_0, weight = module_layers_19_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1041_cast_fp16)[name = string("x_439_cast_fp16")];
            tensor<int32, [3]> input_1043_perm_0 = const()[name = string("input_1043_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1043_cast_fp16 = transpose(perm = input_1043_perm_0, x = x_439_cast_fp16)[name = string("transpose_277")];
            tensor<fp16, [1, 188, 1024]> input_1045_cast_fp16 = add(x = input_1027_cast_fp16, y = input_1043_cast_fp16)[name = string("input_1045_cast_fp16")];
            tensor<int32, [1]> input_1047_axes_0 = const()[name = string("input_1047_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274545600)))];
            tensor<fp16, [1024]> module_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274547712)))];
            tensor<fp16, [1, 188, 1024]> input_1047_cast_fp16 = layer_norm(axes = input_1047_axes_0, beta = module_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_feed_forward2_weight_to_fp16, x = input_1045_cast_fp16)[name = string("input_1047_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_19_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274549824))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276647040))))[name = string("module_layers_19_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_19_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_19_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276909248)))];
            tensor<fp16, [1, 188, 4096]> linear_179_cast_fp16 = linear(bias = module_layers_19_feed_forward2_linear1_bias_to_fp16, weight = module_layers_19_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1047_cast_fp16)[name = string("linear_179_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1051_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("input_1051_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_19_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276917504))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279014720))))[name = string("module_layers_19_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_19_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_19_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279276928)))];
            tensor<fp16, [1, 188, 1024]> linear_180_cast_fp16 = linear(bias = module_layers_19_feed_forward2_linear2_bias_to_fp16, weight = module_layers_19_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1051_cast_fp16)[name = string("linear_180_cast_fp16")];
            fp16 var_3831_to_fp16 = const()[name = string("op_3831_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3832_cast_fp16 = mul(x = linear_180_cast_fp16, y = var_3831_to_fp16)[name = string("op_3832_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1057_cast_fp16 = add(x = input_1045_cast_fp16, y = var_3832_cast_fp16)[name = string("input_1057_cast_fp16")];
            tensor<int32, [1]> input_1059_axes_0 = const()[name = string("input_1059_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_19_norm_out_weight_to_fp16 = const()[name = string("module_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279279040)))];
            tensor<fp16, [1024]> module_layers_19_norm_out_bias_to_fp16 = const()[name = string("module_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279281152)))];
            tensor<fp16, [1, 188, 1024]> input_1059_cast_fp16 = layer_norm(axes = input_1059_axes_0, beta = module_layers_19_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_19_norm_out_weight_to_fp16, x = input_1057_cast_fp16)[name = string("input_1059_cast_fp16")];
            tensor<int32, [1]> input_1061_axes_0 = const()[name = string("input_1061_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279283264)))];
            tensor<fp16, [1024]> module_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279285376)))];
            tensor<fp16, [1, 188, 1024]> input_1061_cast_fp16 = layer_norm(axes = input_1061_axes_0, beta = module_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward1_weight_to_fp16, x = input_1059_cast_fp16)[name = string("input_1061_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_20_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279287488))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281384704))))[name = string("module_layers_20_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_20_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_20_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281646912)))];
            tensor<fp16, [1, 188, 4096]> linear_181_cast_fp16 = linear(bias = module_layers_20_feed_forward1_linear1_bias_to_fp16, weight = module_layers_20_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1061_cast_fp16)[name = string("linear_181_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1065_cast_fp16 = silu(x = linear_181_cast_fp16)[name = string("input_1065_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_20_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(281655168))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(283752384))))[name = string("module_layers_20_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_20_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_20_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284014592)))];
            tensor<fp16, [1, 188, 1024]> linear_182_cast_fp16 = linear(bias = module_layers_20_feed_forward1_linear2_bias_to_fp16, weight = module_layers_20_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1065_cast_fp16)[name = string("linear_182_cast_fp16")];
            fp16 var_3862_to_fp16 = const()[name = string("op_3862_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_3863_cast_fp16 = mul(x = linear_182_cast_fp16, y = var_3862_to_fp16)[name = string("op_3863_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1071_cast_fp16 = add(x = input_1059_cast_fp16, y = var_3863_cast_fp16)[name = string("input_1071_cast_fp16")];
            tensor<int32, [1]> query_41_axes_0 = const()[name = string("query_41_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_20_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284016704)))];
            tensor<fp16, [1024]> module_layers_20_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284018816)))];
            tensor<fp16, [1, 188, 1024]> query_41_cast_fp16 = layer_norm(axes = query_41_axes_0, beta = module_layers_20_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_self_att_weight_to_fp16, x = input_1071_cast_fp16)[name = string("query_41_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284020928))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284545280))))[name = string("module_layers_20_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_20_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_20_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284610880)))];
            tensor<fp16, [1, 188, 1024]> linear_183_cast_fp16 = linear(bias = module_layers_20_self_attn_linear_q_bias_to_fp16, weight = module_layers_20_self_attn_linear_q_weight_to_fp16_quantized, x = query_41_cast_fp16)[name = string("linear_183_cast_fp16")];
            tensor<int32, [4]> var_3880 = const()[name = string("op_3880"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_121_cast_fp16 = reshape(shape = var_3880, x = linear_183_cast_fp16)[name = string("q_121_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(284612992))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285137344))))[name = string("module_layers_20_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_20_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_20_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285202944)))];
            tensor<fp16, [1, 188, 1024]> linear_184_cast_fp16 = linear(bias = module_layers_20_self_attn_linear_k_bias_to_fp16, weight = module_layers_20_self_attn_linear_k_weight_to_fp16_quantized, x = query_41_cast_fp16)[name = string("linear_184_cast_fp16")];
            tensor<int32, [4]> var_3885 = const()[name = string("op_3885"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_81_cast_fp16 = reshape(shape = var_3885, x = linear_184_cast_fp16)[name = string("k_81_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285205056))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285729408))))[name = string("module_layers_20_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_20_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_20_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285795008)))];
            tensor<fp16, [1, 188, 1024]> linear_185_cast_fp16 = linear(bias = module_layers_20_self_attn_linear_v_bias_to_fp16, weight = module_layers_20_self_attn_linear_v_weight_to_fp16_quantized, x = query_41_cast_fp16)[name = string("linear_185_cast_fp16")];
            tensor<int32, [4]> var_3890 = const()[name = string("op_3890"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_41_cast_fp16 = reshape(shape = var_3890, x = linear_185_cast_fp16)[name = string("v_41_cast_fp16")];
            tensor<int32, [4]> value_43_perm_0 = const()[name = string("value_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285797120))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285797696))))[name = string("module_layers_20_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3902_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_3902_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_20_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285797824))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285798400))))[name = string("module_layers_20_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_3904_cast_fp16 = add(x = q_121_cast_fp16, y = module_layers_20_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_3904_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_41_perm_0 = const()[name = string("q_with_bias_v_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_447_transpose_x_0 = const()[name = string("x_447_transpose_x_0"), val = bool(false)];
            bool x_447_transpose_y_0 = const()[name = string("x_447_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_3906_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285798528))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285990592))))[name = string("op_3906_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_41_cast_fp16 = transpose(perm = q_with_bias_v_41_perm_0, x = var_3904_cast_fp16)[name = string("transpose_276")];
            tensor<fp16, [1, 8, 188, 375]> x_447_cast_fp16 = matmul(transpose_x = x_447_transpose_x_0, transpose_y = x_447_transpose_y_0, x = q_with_bias_v_41_cast_fp16, y = op_3906_to_fp16_quantized)[name = string("x_447_cast_fp16")];
            tensor<int32, [8]> x_449_pad_0 = const()[name = string("x_449_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_449_mode_0 = const()[name = string("x_449_mode_0"), val = string("constant")];
            fp16 const_270_to_fp16 = const()[name = string("const_270_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_449_cast_fp16 = pad(constant_val = const_270_to_fp16, mode = x_449_mode_0, pad = x_449_pad_0, x = x_447_cast_fp16)[name = string("x_449_cast_fp16")];
            tensor<int32, [4]> var_3914 = const()[name = string("op_3914"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_451_cast_fp16 = reshape(shape = var_3914, x = x_449_cast_fp16)[name = string("x_451_cast_fp16")];
            tensor<int32, [4]> var_3918_begin_0 = const()[name = string("op_3918_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3918_end_0 = const()[name = string("op_3918_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_3918_end_mask_0 = const()[name = string("op_3918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_3918_cast_fp16 = slice_by_index(begin = var_3918_begin_0, end = var_3918_end_0, end_mask = var_3918_end_mask_0, x = x_451_cast_fp16)[name = string("op_3918_cast_fp16")];
            tensor<int32, [4]> var_3919 = const()[name = string("op_3919"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_81_cast_fp16 = reshape(shape = var_3919, x = var_3918_cast_fp16)[name = string("matrix_bd_81_cast_fp16")];
            bool matrix_ac_41_transpose_x_0 = const()[name = string("matrix_ac_41_transpose_x_0"), val = bool(false)];
            bool matrix_ac_41_transpose_y_0 = const()[name = string("matrix_ac_41_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_168_perm_0 = const()[name = string("transpose_168_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_169_perm_0 = const()[name = string("transpose_169_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_169 = transpose(perm = transpose_169_perm_0, x = k_81_cast_fp16)[name = string("transpose_274")];
            tensor<fp16, [1, 8, 188, 128]> transpose_168 = transpose(perm = transpose_168_perm_0, x = var_3902_cast_fp16)[name = string("transpose_275")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_41_cast_fp16 = matmul(transpose_x = matrix_ac_41_transpose_x_0, transpose_y = matrix_ac_41_transpose_y_0, x = transpose_168, y = transpose_169)[name = string("matrix_ac_41_cast_fp16")];
            tensor<int32, [4]> matrix_bd_83_begin_0 = const()[name = string("matrix_bd_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_83_end_0 = const()[name = string("matrix_bd_83_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_83_end_mask_0 = const()[name = string("matrix_bd_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_83_cast_fp16 = slice_by_index(begin = matrix_bd_83_begin_0, end = matrix_bd_83_end_0, end_mask = matrix_bd_83_end_mask_0, x = matrix_bd_81_cast_fp16)[name = string("matrix_bd_83_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3928_cast_fp16 = add(x = matrix_ac_41_cast_fp16, y = matrix_bd_83_cast_fp16)[name = string("op_3928_cast_fp16")];
            fp16 _inversed_scores_81_y_0_to_fp16 = const()[name = string("_inversed_scores_81_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_81_cast_fp16 = mul(x = var_3928_cast_fp16, y = _inversed_scores_81_y_0_to_fp16)[name = string("_inversed_scores_81_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_83_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_81_cast_fp16, cond = mask_11)[name = string("scores_83_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_3934_cast_fp16 = softmax(axis = var_23, x = scores_83_cast_fp16)[name = string("op_3934_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1073_cast_fp16 = select(a = var_11_to_fp16, b = var_3934_cast_fp16, cond = mask_11)[name = string("input_1073_cast_fp16")];
            bool x_453_transpose_x_0 = const()[name = string("x_453_transpose_x_0"), val = bool(false)];
            bool x_453_transpose_y_0 = const()[name = string("x_453_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_43_cast_fp16 = transpose(perm = value_43_perm_0, x = v_41_cast_fp16)[name = string("transpose_273")];
            tensor<fp16, [1, 8, 188, 128]> x_453_cast_fp16 = matmul(transpose_x = x_453_transpose_x_0, transpose_y = x_453_transpose_y_0, x = input_1073_cast_fp16, y = value_43_cast_fp16)[name = string("x_453_cast_fp16")];
            tensor<int32, [4]> var_3938_perm_0 = const()[name = string("op_3938_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3939 = const()[name = string("op_3939"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_3938_cast_fp16 = transpose(perm = var_3938_perm_0, x = x_453_cast_fp16)[name = string("transpose_272")];
            tensor<fp16, [1, 188, 1024]> input_1075_cast_fp16 = reshape(shape = var_3939, x = var_3938_cast_fp16)[name = string("input_1075_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_20_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285993664))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286518016))))[name = string("module_layers_20_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_20_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_20_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286583616)))];
            tensor<fp16, [1, 188, 1024]> linear_187_cast_fp16 = linear(bias = module_layers_20_self_attn_linear_out_bias_to_fp16, weight = module_layers_20_self_attn_linear_out_weight_to_fp16_quantized, x = input_1075_cast_fp16)[name = string("linear_187_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1079_cast_fp16 = add(x = input_1071_cast_fp16, y = linear_187_cast_fp16)[name = string("input_1079_cast_fp16")];
            tensor<int32, [1]> x_457_axes_0 = const()[name = string("x_457_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_20_norm_conv_weight_to_fp16 = const()[name = string("module_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286585728)))];
            tensor<fp16, [1024]> module_layers_20_norm_conv_bias_to_fp16 = const()[name = string("module_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286587840)))];
            tensor<fp16, [1, 188, 1024]> x_457_cast_fp16 = layer_norm(axes = x_457_axes_0, beta = module_layers_20_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_conv_weight_to_fp16, x = input_1079_cast_fp16)[name = string("x_457_cast_fp16")];
            tensor<int32, [3]> input_1081_perm_0 = const()[name = string("input_1081_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1083_pad_type_0 = const()[name = string("input_1083_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1083_strides_0 = const()[name = string("input_1083_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1083_pad_0 = const()[name = string("input_1083_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1083_dilations_0 = const()[name = string("input_1083_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1083_groups_0 = const()[name = string("input_1083_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_20_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286589952))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287638592))))[name = string("module_layers_20_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_20_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_20_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287769728)))];
            tensor<fp16, [1, 1024, 188]> input_1081_cast_fp16 = transpose(perm = input_1081_perm_0, x = x_457_cast_fp16)[name = string("transpose_271")];
            tensor<fp16, [1, 2048, 188]> input_1083_cast_fp16 = conv(bias = module_layers_20_conv_pointwise_conv1_bias_to_fp16, dilations = input_1083_dilations_0, groups = input_1083_groups_0, pad = input_1083_pad_0, pad_type = input_1083_pad_type_0, strides = input_1083_strides_0, weight = module_layers_20_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1081_cast_fp16)[name = string("input_1083_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, 188]> x_459_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_459_split_cast_fp16_1 = split(axis = x_459_split_axis_0, num_splits = x_459_split_num_splits_0, x = input_1083_cast_fp16)[name = string("x_459_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_1085_cast_fp16 = select(a = var_11_to_fp16, b = x_459_cast_fp16, cond = var_483)[name = string("input_1085_cast_fp16")];
            tensor<int32, [6]> input_1087_pad_0 = const()[name = string("input_1087_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1087_mode_0 = const()[name = string("input_1087_mode_0"), val = string("constant")];
            fp16 const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1087_cast_fp16 = pad(constant_val = const_273_to_fp16, mode = input_1087_mode_0, pad = input_1087_pad_0, x = input_1085_cast_fp16)[name = string("input_1087_cast_fp16")];
            string input_1089_pad_type_0 = const()[name = string("input_1089_pad_type_0"), val = string("valid")];
            int32 input_1089_groups_0 = const()[name = string("input_1089_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1089_strides_0 = const()[name = string("input_1089_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1089_pad_0 = const()[name = string("input_1089_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1089_dilations_0 = const()[name = string("input_1089_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_424_to_fp16 = const()[name = string("const_424_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287773888)))];
            tensor<fp16, [1024]> const_425_to_fp16 = const()[name = string("const_425_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287792384)))];
            tensor<fp16, [1, 1024, 188]> input_1091_cast_fp16 = conv(bias = const_425_to_fp16, dilations = input_1089_dilations_0, groups = input_1089_groups_0, pad = input_1089_pad_0, pad_type = input_1089_pad_type_0, strides = input_1089_strides_0, weight = const_424_to_fp16, x = input_1087_cast_fp16)[name = string("input_1091_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1093_cast_fp16 = silu(x = input_1091_cast_fp16)[name = string("input_1093_cast_fp16")];
            string x_461_pad_type_0 = const()[name = string("x_461_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_461_strides_0 = const()[name = string("x_461_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_461_pad_0 = const()[name = string("x_461_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_461_dilations_0 = const()[name = string("x_461_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_461_groups_0 = const()[name = string("x_461_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_20_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287794496))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288318848))))[name = string("module_layers_20_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_20_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_20_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288384448)))];
            tensor<fp16, [1, 1024, 188]> x_461_cast_fp16 = conv(bias = module_layers_20_conv_pointwise_conv2_bias_to_fp16, dilations = x_461_dilations_0, groups = x_461_groups_0, pad = x_461_pad_0, pad_type = x_461_pad_type_0, strides = x_461_strides_0, weight = module_layers_20_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1093_cast_fp16)[name = string("x_461_cast_fp16")];
            tensor<int32, [3]> input_1095_perm_0 = const()[name = string("input_1095_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1095_cast_fp16 = transpose(perm = input_1095_perm_0, x = x_461_cast_fp16)[name = string("transpose_270")];
            tensor<fp16, [1, 188, 1024]> input_1097_cast_fp16 = add(x = input_1079_cast_fp16, y = input_1095_cast_fp16)[name = string("input_1097_cast_fp16")];
            tensor<int32, [1]> input_1099_axes_0 = const()[name = string("input_1099_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288386560)))];
            tensor<fp16, [1024]> module_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288388672)))];
            tensor<fp16, [1, 188, 1024]> input_1099_cast_fp16 = layer_norm(axes = input_1099_axes_0, beta = module_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_feed_forward2_weight_to_fp16, x = input_1097_cast_fp16)[name = string("input_1099_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_20_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(288390784))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290488000))))[name = string("module_layers_20_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_20_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_20_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290750208)))];
            tensor<fp16, [1, 188, 4096]> linear_188_cast_fp16 = linear(bias = module_layers_20_feed_forward2_linear1_bias_to_fp16, weight = module_layers_20_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1099_cast_fp16)[name = string("linear_188_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1103_cast_fp16 = silu(x = linear_188_cast_fp16)[name = string("input_1103_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_20_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(290758464))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292855680))))[name = string("module_layers_20_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_20_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_20_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293117888)))];
            tensor<fp16, [1, 188, 1024]> linear_189_cast_fp16 = linear(bias = module_layers_20_feed_forward2_linear2_bias_to_fp16, weight = module_layers_20_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1103_cast_fp16)[name = string("linear_189_cast_fp16")];
            fp16 var_4005_to_fp16 = const()[name = string("op_4005_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4006_cast_fp16 = mul(x = linear_189_cast_fp16, y = var_4005_to_fp16)[name = string("op_4006_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1109_cast_fp16 = add(x = input_1097_cast_fp16, y = var_4006_cast_fp16)[name = string("input_1109_cast_fp16")];
            tensor<int32, [1]> input_1111_axes_0 = const()[name = string("input_1111_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_20_norm_out_weight_to_fp16 = const()[name = string("module_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293120000)))];
            tensor<fp16, [1024]> module_layers_20_norm_out_bias_to_fp16 = const()[name = string("module_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293122112)))];
            tensor<fp16, [1, 188, 1024]> input_1111_cast_fp16 = layer_norm(axes = input_1111_axes_0, beta = module_layers_20_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_20_norm_out_weight_to_fp16, x = input_1109_cast_fp16)[name = string("input_1111_cast_fp16")];
            tensor<int32, [1]> input_1113_axes_0 = const()[name = string("input_1113_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293124224)))];
            tensor<fp16, [1024]> module_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293126336)))];
            tensor<fp16, [1, 188, 1024]> input_1113_cast_fp16 = layer_norm(axes = input_1113_axes_0, beta = module_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward1_weight_to_fp16, x = input_1111_cast_fp16)[name = string("input_1113_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_21_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293128448))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295225664))))[name = string("module_layers_21_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_21_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_21_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295487872)))];
            tensor<fp16, [1, 188, 4096]> linear_190_cast_fp16 = linear(bias = module_layers_21_feed_forward1_linear1_bias_to_fp16, weight = module_layers_21_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1113_cast_fp16)[name = string("linear_190_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1117_cast_fp16 = silu(x = linear_190_cast_fp16)[name = string("input_1117_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_21_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295496128))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297593344))))[name = string("module_layers_21_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_21_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_21_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297855552)))];
            tensor<fp16, [1, 188, 1024]> linear_191_cast_fp16 = linear(bias = module_layers_21_feed_forward1_linear2_bias_to_fp16, weight = module_layers_21_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1117_cast_fp16)[name = string("linear_191_cast_fp16")];
            fp16 var_4036_to_fp16 = const()[name = string("op_4036_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4037_cast_fp16 = mul(x = linear_191_cast_fp16, y = var_4036_to_fp16)[name = string("op_4037_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1123_cast_fp16 = add(x = input_1111_cast_fp16, y = var_4037_cast_fp16)[name = string("input_1123_cast_fp16")];
            tensor<int32, [1]> query_43_axes_0 = const()[name = string("query_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_21_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297857664)))];
            tensor<fp16, [1024]> module_layers_21_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297859776)))];
            tensor<fp16, [1, 188, 1024]> query_43_cast_fp16 = layer_norm(axes = query_43_axes_0, beta = module_layers_21_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_self_att_weight_to_fp16, x = input_1123_cast_fp16)[name = string("query_43_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297861888))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298386240))))[name = string("module_layers_21_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_21_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_21_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298451840)))];
            tensor<fp16, [1, 188, 1024]> linear_192_cast_fp16 = linear(bias = module_layers_21_self_attn_linear_q_bias_to_fp16, weight = module_layers_21_self_attn_linear_q_weight_to_fp16_quantized, x = query_43_cast_fp16)[name = string("linear_192_cast_fp16")];
            tensor<int32, [4]> var_4054 = const()[name = string("op_4054"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_127_cast_fp16 = reshape(shape = var_4054, x = linear_192_cast_fp16)[name = string("q_127_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298453952))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298978304))))[name = string("module_layers_21_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_21_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_21_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299043904)))];
            tensor<fp16, [1, 188, 1024]> linear_193_cast_fp16 = linear(bias = module_layers_21_self_attn_linear_k_bias_to_fp16, weight = module_layers_21_self_attn_linear_k_weight_to_fp16_quantized, x = query_43_cast_fp16)[name = string("linear_193_cast_fp16")];
            tensor<int32, [4]> var_4059 = const()[name = string("op_4059"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_85_cast_fp16 = reshape(shape = var_4059, x = linear_193_cast_fp16)[name = string("k_85_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299046016))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299570368))))[name = string("module_layers_21_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_21_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_21_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299635968)))];
            tensor<fp16, [1, 188, 1024]> linear_194_cast_fp16 = linear(bias = module_layers_21_self_attn_linear_v_bias_to_fp16, weight = module_layers_21_self_attn_linear_v_weight_to_fp16_quantized, x = query_43_cast_fp16)[name = string("linear_194_cast_fp16")];
            tensor<int32, [4]> var_4064 = const()[name = string("op_4064"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_43_cast_fp16 = reshape(shape = var_4064, x = linear_194_cast_fp16)[name = string("v_43_cast_fp16")];
            tensor<int32, [4]> value_45_perm_0 = const()[name = string("value_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299638080))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299638656))))[name = string("module_layers_21_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4076_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_4076_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_21_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299638784))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299639360))))[name = string("module_layers_21_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4078_cast_fp16 = add(x = q_127_cast_fp16, y = module_layers_21_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_4078_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_43_perm_0 = const()[name = string("q_with_bias_v_43_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_469_transpose_x_0 = const()[name = string("x_469_transpose_x_0"), val = bool(false)];
            bool x_469_transpose_y_0 = const()[name = string("x_469_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_4080_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299639488))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299831552))))[name = string("op_4080_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_43_cast_fp16 = transpose(perm = q_with_bias_v_43_perm_0, x = var_4078_cast_fp16)[name = string("transpose_269")];
            tensor<fp16, [1, 8, 188, 375]> x_469_cast_fp16 = matmul(transpose_x = x_469_transpose_x_0, transpose_y = x_469_transpose_y_0, x = q_with_bias_v_43_cast_fp16, y = op_4080_to_fp16_quantized)[name = string("x_469_cast_fp16")];
            tensor<int32, [8]> x_471_pad_0 = const()[name = string("x_471_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_471_mode_0 = const()[name = string("x_471_mode_0"), val = string("constant")];
            fp16 const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_471_cast_fp16 = pad(constant_val = const_280_to_fp16, mode = x_471_mode_0, pad = x_471_pad_0, x = x_469_cast_fp16)[name = string("x_471_cast_fp16")];
            tensor<int32, [4]> var_4088 = const()[name = string("op_4088"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_473_cast_fp16 = reshape(shape = var_4088, x = x_471_cast_fp16)[name = string("x_473_cast_fp16")];
            tensor<int32, [4]> var_4092_begin_0 = const()[name = string("op_4092_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4092_end_0 = const()[name = string("op_4092_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4092_end_mask_0 = const()[name = string("op_4092_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4092_cast_fp16 = slice_by_index(begin = var_4092_begin_0, end = var_4092_end_0, end_mask = var_4092_end_mask_0, x = x_473_cast_fp16)[name = string("op_4092_cast_fp16")];
            tensor<int32, [4]> var_4093 = const()[name = string("op_4093"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_85_cast_fp16 = reshape(shape = var_4093, x = var_4092_cast_fp16)[name = string("matrix_bd_85_cast_fp16")];
            bool matrix_ac_43_transpose_x_0 = const()[name = string("matrix_ac_43_transpose_x_0"), val = bool(false)];
            bool matrix_ac_43_transpose_y_0 = const()[name = string("matrix_ac_43_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_170_perm_0 = const()[name = string("transpose_170_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_171_perm_0 = const()[name = string("transpose_171_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_171 = transpose(perm = transpose_171_perm_0, x = k_85_cast_fp16)[name = string("transpose_267")];
            tensor<fp16, [1, 8, 188, 128]> transpose_170 = transpose(perm = transpose_170_perm_0, x = var_4076_cast_fp16)[name = string("transpose_268")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_43_cast_fp16 = matmul(transpose_x = matrix_ac_43_transpose_x_0, transpose_y = matrix_ac_43_transpose_y_0, x = transpose_170, y = transpose_171)[name = string("matrix_ac_43_cast_fp16")];
            tensor<int32, [4]> matrix_bd_87_begin_0 = const()[name = string("matrix_bd_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_87_end_0 = const()[name = string("matrix_bd_87_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_87_end_mask_0 = const()[name = string("matrix_bd_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_87_cast_fp16 = slice_by_index(begin = matrix_bd_87_begin_0, end = matrix_bd_87_end_0, end_mask = matrix_bd_87_end_mask_0, x = matrix_bd_85_cast_fp16)[name = string("matrix_bd_87_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4102_cast_fp16 = add(x = matrix_ac_43_cast_fp16, y = matrix_bd_87_cast_fp16)[name = string("op_4102_cast_fp16")];
            fp16 _inversed_scores_85_y_0_to_fp16 = const()[name = string("_inversed_scores_85_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_85_cast_fp16 = mul(x = var_4102_cast_fp16, y = _inversed_scores_85_y_0_to_fp16)[name = string("_inversed_scores_85_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_87_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_85_cast_fp16, cond = mask_11)[name = string("scores_87_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4108_cast_fp16 = softmax(axis = var_23, x = scores_87_cast_fp16)[name = string("op_4108_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1125_cast_fp16 = select(a = var_11_to_fp16, b = var_4108_cast_fp16, cond = mask_11)[name = string("input_1125_cast_fp16")];
            bool x_475_transpose_x_0 = const()[name = string("x_475_transpose_x_0"), val = bool(false)];
            bool x_475_transpose_y_0 = const()[name = string("x_475_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_45_cast_fp16 = transpose(perm = value_45_perm_0, x = v_43_cast_fp16)[name = string("transpose_266")];
            tensor<fp16, [1, 8, 188, 128]> x_475_cast_fp16 = matmul(transpose_x = x_475_transpose_x_0, transpose_y = x_475_transpose_y_0, x = input_1125_cast_fp16, y = value_45_cast_fp16)[name = string("x_475_cast_fp16")];
            tensor<int32, [4]> var_4112_perm_0 = const()[name = string("op_4112_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4113 = const()[name = string("op_4113"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4112_cast_fp16 = transpose(perm = var_4112_perm_0, x = x_475_cast_fp16)[name = string("transpose_265")];
            tensor<fp16, [1, 188, 1024]> input_1127_cast_fp16 = reshape(shape = var_4113, x = var_4112_cast_fp16)[name = string("input_1127_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_21_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(299834624))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300358976))))[name = string("module_layers_21_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_21_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_21_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300424576)))];
            tensor<fp16, [1, 188, 1024]> linear_196_cast_fp16 = linear(bias = module_layers_21_self_attn_linear_out_bias_to_fp16, weight = module_layers_21_self_attn_linear_out_weight_to_fp16_quantized, x = input_1127_cast_fp16)[name = string("linear_196_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1131_cast_fp16 = add(x = input_1123_cast_fp16, y = linear_196_cast_fp16)[name = string("input_1131_cast_fp16")];
            tensor<int32, [1]> x_479_axes_0 = const()[name = string("x_479_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_21_norm_conv_weight_to_fp16 = const()[name = string("module_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300426688)))];
            tensor<fp16, [1024]> module_layers_21_norm_conv_bias_to_fp16 = const()[name = string("module_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300428800)))];
            tensor<fp16, [1, 188, 1024]> x_479_cast_fp16 = layer_norm(axes = x_479_axes_0, beta = module_layers_21_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_conv_weight_to_fp16, x = input_1131_cast_fp16)[name = string("x_479_cast_fp16")];
            tensor<int32, [3]> input_1133_perm_0 = const()[name = string("input_1133_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1135_pad_type_0 = const()[name = string("input_1135_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1135_strides_0 = const()[name = string("input_1135_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1135_pad_0 = const()[name = string("input_1135_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1135_dilations_0 = const()[name = string("input_1135_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1135_groups_0 = const()[name = string("input_1135_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_21_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(300430912))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301479552))))[name = string("module_layers_21_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_21_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_21_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301610688)))];
            tensor<fp16, [1, 1024, 188]> input_1133_cast_fp16 = transpose(perm = input_1133_perm_0, x = x_479_cast_fp16)[name = string("transpose_264")];
            tensor<fp16, [1, 2048, 188]> input_1135_cast_fp16 = conv(bias = module_layers_21_conv_pointwise_conv1_bias_to_fp16, dilations = input_1135_dilations_0, groups = input_1135_groups_0, pad = input_1135_pad_0, pad_type = input_1135_pad_type_0, strides = input_1135_strides_0, weight = module_layers_21_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1133_cast_fp16)[name = string("input_1135_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, 188]> x_481_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_481_split_cast_fp16_1 = split(axis = x_481_split_axis_0, num_splits = x_481_split_num_splits_0, x = input_1135_cast_fp16)[name = string("x_481_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_1137_cast_fp16 = select(a = var_11_to_fp16, b = x_481_cast_fp16, cond = var_483)[name = string("input_1137_cast_fp16")];
            tensor<int32, [6]> input_1139_pad_0 = const()[name = string("input_1139_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1139_mode_0 = const()[name = string("input_1139_mode_0"), val = string("constant")];
            fp16 const_283_to_fp16 = const()[name = string("const_283_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1139_cast_fp16 = pad(constant_val = const_283_to_fp16, mode = input_1139_mode_0, pad = input_1139_pad_0, x = input_1137_cast_fp16)[name = string("input_1139_cast_fp16")];
            string input_1141_pad_type_0 = const()[name = string("input_1141_pad_type_0"), val = string("valid")];
            int32 input_1141_groups_0 = const()[name = string("input_1141_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1141_strides_0 = const()[name = string("input_1141_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1141_pad_0 = const()[name = string("input_1141_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1141_dilations_0 = const()[name = string("input_1141_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_426_to_fp16 = const()[name = string("const_426_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301614848)))];
            tensor<fp16, [1024]> const_427_to_fp16 = const()[name = string("const_427_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301633344)))];
            tensor<fp16, [1, 1024, 188]> input_1143_cast_fp16 = conv(bias = const_427_to_fp16, dilations = input_1141_dilations_0, groups = input_1141_groups_0, pad = input_1141_pad_0, pad_type = input_1141_pad_type_0, strides = input_1141_strides_0, weight = const_426_to_fp16, x = input_1139_cast_fp16)[name = string("input_1143_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1145_cast_fp16 = silu(x = input_1143_cast_fp16)[name = string("input_1145_cast_fp16")];
            string x_483_pad_type_0 = const()[name = string("x_483_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_483_strides_0 = const()[name = string("x_483_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_483_pad_0 = const()[name = string("x_483_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_483_dilations_0 = const()[name = string("x_483_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_483_groups_0 = const()[name = string("x_483_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_21_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(301635456))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302159808))))[name = string("module_layers_21_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_21_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_21_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302225408)))];
            tensor<fp16, [1, 1024, 188]> x_483_cast_fp16 = conv(bias = module_layers_21_conv_pointwise_conv2_bias_to_fp16, dilations = x_483_dilations_0, groups = x_483_groups_0, pad = x_483_pad_0, pad_type = x_483_pad_type_0, strides = x_483_strides_0, weight = module_layers_21_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1145_cast_fp16)[name = string("x_483_cast_fp16")];
            tensor<int32, [3]> input_1147_perm_0 = const()[name = string("input_1147_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1147_cast_fp16 = transpose(perm = input_1147_perm_0, x = x_483_cast_fp16)[name = string("transpose_263")];
            tensor<fp16, [1, 188, 1024]> input_1149_cast_fp16 = add(x = input_1131_cast_fp16, y = input_1147_cast_fp16)[name = string("input_1149_cast_fp16")];
            tensor<int32, [1]> input_1151_axes_0 = const()[name = string("input_1151_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302227520)))];
            tensor<fp16, [1024]> module_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302229632)))];
            tensor<fp16, [1, 188, 1024]> input_1151_cast_fp16 = layer_norm(axes = input_1151_axes_0, beta = module_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_feed_forward2_weight_to_fp16, x = input_1149_cast_fp16)[name = string("input_1151_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_21_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(302231744))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304328960))))[name = string("module_layers_21_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_21_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_21_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304591168)))];
            tensor<fp16, [1, 188, 4096]> linear_197_cast_fp16 = linear(bias = module_layers_21_feed_forward2_linear1_bias_to_fp16, weight = module_layers_21_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1151_cast_fp16)[name = string("linear_197_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1155_cast_fp16 = silu(x = linear_197_cast_fp16)[name = string("input_1155_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_21_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304599424))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306696640))))[name = string("module_layers_21_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_21_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_21_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306958848)))];
            tensor<fp16, [1, 188, 1024]> linear_198_cast_fp16 = linear(bias = module_layers_21_feed_forward2_linear2_bias_to_fp16, weight = module_layers_21_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1155_cast_fp16)[name = string("linear_198_cast_fp16")];
            fp16 var_4179_to_fp16 = const()[name = string("op_4179_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4180_cast_fp16 = mul(x = linear_198_cast_fp16, y = var_4179_to_fp16)[name = string("op_4180_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1161_cast_fp16 = add(x = input_1149_cast_fp16, y = var_4180_cast_fp16)[name = string("input_1161_cast_fp16")];
            tensor<int32, [1]> input_1163_axes_0 = const()[name = string("input_1163_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_21_norm_out_weight_to_fp16 = const()[name = string("module_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306960960)))];
            tensor<fp16, [1024]> module_layers_21_norm_out_bias_to_fp16 = const()[name = string("module_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306963072)))];
            tensor<fp16, [1, 188, 1024]> input_1163_cast_fp16 = layer_norm(axes = input_1163_axes_0, beta = module_layers_21_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_21_norm_out_weight_to_fp16, x = input_1161_cast_fp16)[name = string("input_1163_cast_fp16")];
            tensor<int32, [1]> input_1165_axes_0 = const()[name = string("input_1165_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306965184)))];
            tensor<fp16, [1024]> module_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306967296)))];
            tensor<fp16, [1, 188, 1024]> input_1165_cast_fp16 = layer_norm(axes = input_1165_axes_0, beta = module_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward1_weight_to_fp16, x = input_1163_cast_fp16)[name = string("input_1165_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_22_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306969408))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309066624))))[name = string("module_layers_22_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_22_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_22_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309328832)))];
            tensor<fp16, [1, 188, 4096]> linear_199_cast_fp16 = linear(bias = module_layers_22_feed_forward1_linear1_bias_to_fp16, weight = module_layers_22_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1165_cast_fp16)[name = string("linear_199_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1169_cast_fp16 = silu(x = linear_199_cast_fp16)[name = string("input_1169_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_22_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(309337088))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311434304))))[name = string("module_layers_22_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_22_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_22_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311696512)))];
            tensor<fp16, [1, 188, 1024]> linear_200_cast_fp16 = linear(bias = module_layers_22_feed_forward1_linear2_bias_to_fp16, weight = module_layers_22_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1169_cast_fp16)[name = string("linear_200_cast_fp16")];
            fp16 var_4210_to_fp16 = const()[name = string("op_4210_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4211_cast_fp16 = mul(x = linear_200_cast_fp16, y = var_4210_to_fp16)[name = string("op_4211_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1175_cast_fp16 = add(x = input_1163_cast_fp16, y = var_4211_cast_fp16)[name = string("input_1175_cast_fp16")];
            tensor<int32, [1]> query_45_axes_0 = const()[name = string("query_45_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_22_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311698624)))];
            tensor<fp16, [1024]> module_layers_22_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311700736)))];
            tensor<fp16, [1, 188, 1024]> query_45_cast_fp16 = layer_norm(axes = query_45_axes_0, beta = module_layers_22_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_self_att_weight_to_fp16, x = input_1175_cast_fp16)[name = string("query_45_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311702848))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312227200))))[name = string("module_layers_22_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_22_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_22_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312292800)))];
            tensor<fp16, [1, 188, 1024]> linear_201_cast_fp16 = linear(bias = module_layers_22_self_attn_linear_q_bias_to_fp16, weight = module_layers_22_self_attn_linear_q_weight_to_fp16_quantized, x = query_45_cast_fp16)[name = string("linear_201_cast_fp16")];
            tensor<int32, [4]> var_4228 = const()[name = string("op_4228"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_133_cast_fp16 = reshape(shape = var_4228, x = linear_201_cast_fp16)[name = string("q_133_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312294912))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312819264))))[name = string("module_layers_22_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_22_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_22_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312884864)))];
            tensor<fp16, [1, 188, 1024]> linear_202_cast_fp16 = linear(bias = module_layers_22_self_attn_linear_k_bias_to_fp16, weight = module_layers_22_self_attn_linear_k_weight_to_fp16_quantized, x = query_45_cast_fp16)[name = string("linear_202_cast_fp16")];
            tensor<int32, [4]> var_4233 = const()[name = string("op_4233"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_89_cast_fp16 = reshape(shape = var_4233, x = linear_202_cast_fp16)[name = string("k_89_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312886976))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313411328))))[name = string("module_layers_22_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_22_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_22_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313476928)))];
            tensor<fp16, [1, 188, 1024]> linear_203_cast_fp16 = linear(bias = module_layers_22_self_attn_linear_v_bias_to_fp16, weight = module_layers_22_self_attn_linear_v_weight_to_fp16_quantized, x = query_45_cast_fp16)[name = string("linear_203_cast_fp16")];
            tensor<int32, [4]> var_4238 = const()[name = string("op_4238"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_45_cast_fp16 = reshape(shape = var_4238, x = linear_203_cast_fp16)[name = string("v_45_cast_fp16")];
            tensor<int32, [4]> value_47_perm_0 = const()[name = string("value_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313479040))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313479616))))[name = string("module_layers_22_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4250_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_4250_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_22_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313479744))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313480320))))[name = string("module_layers_22_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4252_cast_fp16 = add(x = q_133_cast_fp16, y = module_layers_22_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_4252_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_45_perm_0 = const()[name = string("q_with_bias_v_45_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_491_transpose_x_0 = const()[name = string("x_491_transpose_x_0"), val = bool(false)];
            bool x_491_transpose_y_0 = const()[name = string("x_491_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_4254_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313480448))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313672512))))[name = string("op_4254_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_45_cast_fp16 = transpose(perm = q_with_bias_v_45_perm_0, x = var_4252_cast_fp16)[name = string("transpose_262")];
            tensor<fp16, [1, 8, 188, 375]> x_491_cast_fp16 = matmul(transpose_x = x_491_transpose_x_0, transpose_y = x_491_transpose_y_0, x = q_with_bias_v_45_cast_fp16, y = op_4254_to_fp16_quantized)[name = string("x_491_cast_fp16")];
            tensor<int32, [8]> x_493_pad_0 = const()[name = string("x_493_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_493_mode_0 = const()[name = string("x_493_mode_0"), val = string("constant")];
            fp16 const_290_to_fp16 = const()[name = string("const_290_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_493_cast_fp16 = pad(constant_val = const_290_to_fp16, mode = x_493_mode_0, pad = x_493_pad_0, x = x_491_cast_fp16)[name = string("x_493_cast_fp16")];
            tensor<int32, [4]> var_4262 = const()[name = string("op_4262"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_495_cast_fp16 = reshape(shape = var_4262, x = x_493_cast_fp16)[name = string("x_495_cast_fp16")];
            tensor<int32, [4]> var_4266_begin_0 = const()[name = string("op_4266_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4266_end_0 = const()[name = string("op_4266_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4266_end_mask_0 = const()[name = string("op_4266_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4266_cast_fp16 = slice_by_index(begin = var_4266_begin_0, end = var_4266_end_0, end_mask = var_4266_end_mask_0, x = x_495_cast_fp16)[name = string("op_4266_cast_fp16")];
            tensor<int32, [4]> var_4267 = const()[name = string("op_4267"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_89_cast_fp16 = reshape(shape = var_4267, x = var_4266_cast_fp16)[name = string("matrix_bd_89_cast_fp16")];
            bool matrix_ac_45_transpose_x_0 = const()[name = string("matrix_ac_45_transpose_x_0"), val = bool(false)];
            bool matrix_ac_45_transpose_y_0 = const()[name = string("matrix_ac_45_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_172_perm_0 = const()[name = string("transpose_172_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_173_perm_0 = const()[name = string("transpose_173_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_173 = transpose(perm = transpose_173_perm_0, x = k_89_cast_fp16)[name = string("transpose_260")];
            tensor<fp16, [1, 8, 188, 128]> transpose_172 = transpose(perm = transpose_172_perm_0, x = var_4250_cast_fp16)[name = string("transpose_261")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_45_cast_fp16 = matmul(transpose_x = matrix_ac_45_transpose_x_0, transpose_y = matrix_ac_45_transpose_y_0, x = transpose_172, y = transpose_173)[name = string("matrix_ac_45_cast_fp16")];
            tensor<int32, [4]> matrix_bd_91_begin_0 = const()[name = string("matrix_bd_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_91_end_0 = const()[name = string("matrix_bd_91_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_91_end_mask_0 = const()[name = string("matrix_bd_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_91_cast_fp16 = slice_by_index(begin = matrix_bd_91_begin_0, end = matrix_bd_91_end_0, end_mask = matrix_bd_91_end_mask_0, x = matrix_bd_89_cast_fp16)[name = string("matrix_bd_91_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4276_cast_fp16 = add(x = matrix_ac_45_cast_fp16, y = matrix_bd_91_cast_fp16)[name = string("op_4276_cast_fp16")];
            fp16 _inversed_scores_89_y_0_to_fp16 = const()[name = string("_inversed_scores_89_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_89_cast_fp16 = mul(x = var_4276_cast_fp16, y = _inversed_scores_89_y_0_to_fp16)[name = string("_inversed_scores_89_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_91_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_89_cast_fp16, cond = mask_11)[name = string("scores_91_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4282_cast_fp16 = softmax(axis = var_23, x = scores_91_cast_fp16)[name = string("op_4282_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1177_cast_fp16 = select(a = var_11_to_fp16, b = var_4282_cast_fp16, cond = mask_11)[name = string("input_1177_cast_fp16")];
            bool x_497_transpose_x_0 = const()[name = string("x_497_transpose_x_0"), val = bool(false)];
            bool x_497_transpose_y_0 = const()[name = string("x_497_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_47_cast_fp16 = transpose(perm = value_47_perm_0, x = v_45_cast_fp16)[name = string("transpose_259")];
            tensor<fp16, [1, 8, 188, 128]> x_497_cast_fp16 = matmul(transpose_x = x_497_transpose_x_0, transpose_y = x_497_transpose_y_0, x = input_1177_cast_fp16, y = value_47_cast_fp16)[name = string("x_497_cast_fp16")];
            tensor<int32, [4]> var_4286_perm_0 = const()[name = string("op_4286_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4287 = const()[name = string("op_4287"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4286_cast_fp16 = transpose(perm = var_4286_perm_0, x = x_497_cast_fp16)[name = string("transpose_258")];
            tensor<fp16, [1, 188, 1024]> input_1179_cast_fp16 = reshape(shape = var_4287, x = var_4286_cast_fp16)[name = string("input_1179_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_22_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313675584))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314199936))))[name = string("module_layers_22_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_22_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_22_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314265536)))];
            tensor<fp16, [1, 188, 1024]> linear_205_cast_fp16 = linear(bias = module_layers_22_self_attn_linear_out_bias_to_fp16, weight = module_layers_22_self_attn_linear_out_weight_to_fp16_quantized, x = input_1179_cast_fp16)[name = string("linear_205_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1183_cast_fp16 = add(x = input_1175_cast_fp16, y = linear_205_cast_fp16)[name = string("input_1183_cast_fp16")];
            tensor<int32, [1]> x_501_axes_0 = const()[name = string("x_501_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_22_norm_conv_weight_to_fp16 = const()[name = string("module_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314267648)))];
            tensor<fp16, [1024]> module_layers_22_norm_conv_bias_to_fp16 = const()[name = string("module_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314269760)))];
            tensor<fp16, [1, 188, 1024]> x_501_cast_fp16 = layer_norm(axes = x_501_axes_0, beta = module_layers_22_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_conv_weight_to_fp16, x = input_1183_cast_fp16)[name = string("x_501_cast_fp16")];
            tensor<int32, [3]> input_1185_perm_0 = const()[name = string("input_1185_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1187_pad_type_0 = const()[name = string("input_1187_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1187_strides_0 = const()[name = string("input_1187_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1187_pad_0 = const()[name = string("input_1187_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1187_dilations_0 = const()[name = string("input_1187_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1187_groups_0 = const()[name = string("input_1187_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_22_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314271872))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315320512))))[name = string("module_layers_22_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_22_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_22_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315451648)))];
            tensor<fp16, [1, 1024, 188]> input_1185_cast_fp16 = transpose(perm = input_1185_perm_0, x = x_501_cast_fp16)[name = string("transpose_257")];
            tensor<fp16, [1, 2048, 188]> input_1187_cast_fp16 = conv(bias = module_layers_22_conv_pointwise_conv1_bias_to_fp16, dilations = input_1187_dilations_0, groups = input_1187_groups_0, pad = input_1187_pad_0, pad_type = input_1187_pad_type_0, strides = input_1187_strides_0, weight = module_layers_22_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1185_cast_fp16)[name = string("input_1187_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, 188]> x_503_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_503_split_cast_fp16_1 = split(axis = x_503_split_axis_0, num_splits = x_503_split_num_splits_0, x = input_1187_cast_fp16)[name = string("x_503_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_1189_cast_fp16 = select(a = var_11_to_fp16, b = x_503_cast_fp16, cond = var_483)[name = string("input_1189_cast_fp16")];
            tensor<int32, [6]> input_1191_pad_0 = const()[name = string("input_1191_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1191_mode_0 = const()[name = string("input_1191_mode_0"), val = string("constant")];
            fp16 const_293_to_fp16 = const()[name = string("const_293_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1191_cast_fp16 = pad(constant_val = const_293_to_fp16, mode = input_1191_mode_0, pad = input_1191_pad_0, x = input_1189_cast_fp16)[name = string("input_1191_cast_fp16")];
            string input_1193_pad_type_0 = const()[name = string("input_1193_pad_type_0"), val = string("valid")];
            int32 input_1193_groups_0 = const()[name = string("input_1193_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1193_strides_0 = const()[name = string("input_1193_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1193_pad_0 = const()[name = string("input_1193_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1193_dilations_0 = const()[name = string("input_1193_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_428_to_fp16 = const()[name = string("const_428_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315455808)))];
            tensor<fp16, [1024]> const_429_to_fp16 = const()[name = string("const_429_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315474304)))];
            tensor<fp16, [1, 1024, 188]> input_1195_cast_fp16 = conv(bias = const_429_to_fp16, dilations = input_1193_dilations_0, groups = input_1193_groups_0, pad = input_1193_pad_0, pad_type = input_1193_pad_type_0, strides = input_1193_strides_0, weight = const_428_to_fp16, x = input_1191_cast_fp16)[name = string("input_1195_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1197_cast_fp16 = silu(x = input_1195_cast_fp16)[name = string("input_1197_cast_fp16")];
            string x_505_pad_type_0 = const()[name = string("x_505_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_505_strides_0 = const()[name = string("x_505_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_505_pad_0 = const()[name = string("x_505_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_505_dilations_0 = const()[name = string("x_505_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_505_groups_0 = const()[name = string("x_505_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_22_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(315476416))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316000768))))[name = string("module_layers_22_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_22_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_22_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316066368)))];
            tensor<fp16, [1, 1024, 188]> x_505_cast_fp16 = conv(bias = module_layers_22_conv_pointwise_conv2_bias_to_fp16, dilations = x_505_dilations_0, groups = x_505_groups_0, pad = x_505_pad_0, pad_type = x_505_pad_type_0, strides = x_505_strides_0, weight = module_layers_22_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1197_cast_fp16)[name = string("x_505_cast_fp16")];
            tensor<int32, [3]> input_1199_perm_0 = const()[name = string("input_1199_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1199_cast_fp16 = transpose(perm = input_1199_perm_0, x = x_505_cast_fp16)[name = string("transpose_256")];
            tensor<fp16, [1, 188, 1024]> input_1201_cast_fp16 = add(x = input_1183_cast_fp16, y = input_1199_cast_fp16)[name = string("input_1201_cast_fp16")];
            tensor<int32, [1]> input_1203_axes_0 = const()[name = string("input_1203_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316068480)))];
            tensor<fp16, [1024]> module_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316070592)))];
            tensor<fp16, [1, 188, 1024]> input_1203_cast_fp16 = layer_norm(axes = input_1203_axes_0, beta = module_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_feed_forward2_weight_to_fp16, x = input_1201_cast_fp16)[name = string("input_1203_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_22_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316072704))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318169920))))[name = string("module_layers_22_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_22_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_22_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318432128)))];
            tensor<fp16, [1, 188, 4096]> linear_206_cast_fp16 = linear(bias = module_layers_22_feed_forward2_linear1_bias_to_fp16, weight = module_layers_22_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1203_cast_fp16)[name = string("linear_206_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1207_cast_fp16 = silu(x = linear_206_cast_fp16)[name = string("input_1207_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_22_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318440384))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320537600))))[name = string("module_layers_22_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_22_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_22_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320799808)))];
            tensor<fp16, [1, 188, 1024]> linear_207_cast_fp16 = linear(bias = module_layers_22_feed_forward2_linear2_bias_to_fp16, weight = module_layers_22_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1207_cast_fp16)[name = string("linear_207_cast_fp16")];
            fp16 var_4353_to_fp16 = const()[name = string("op_4353_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4354_cast_fp16 = mul(x = linear_207_cast_fp16, y = var_4353_to_fp16)[name = string("op_4354_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1213_cast_fp16 = add(x = input_1201_cast_fp16, y = var_4354_cast_fp16)[name = string("input_1213_cast_fp16")];
            tensor<int32, [1]> input_1215_axes_0 = const()[name = string("input_1215_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_22_norm_out_weight_to_fp16 = const()[name = string("module_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320801920)))];
            tensor<fp16, [1024]> module_layers_22_norm_out_bias_to_fp16 = const()[name = string("module_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320804032)))];
            tensor<fp16, [1, 188, 1024]> input_1215_cast_fp16 = layer_norm(axes = input_1215_axes_0, beta = module_layers_22_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_22_norm_out_weight_to_fp16, x = input_1213_cast_fp16)[name = string("input_1215_cast_fp16")];
            tensor<int32, [1]> input_1217_axes_0 = const()[name = string("input_1217_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320806144)))];
            tensor<fp16, [1024]> module_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320808256)))];
            tensor<fp16, [1, 188, 1024]> input_1217_cast_fp16 = layer_norm(axes = input_1217_axes_0, beta = module_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward1_weight_to_fp16, x = input_1215_cast_fp16)[name = string("input_1217_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_23_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320810368))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(322907584))))[name = string("module_layers_23_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_23_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_23_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323169792)))];
            tensor<fp16, [1, 188, 4096]> linear_208_cast_fp16 = linear(bias = module_layers_23_feed_forward1_linear1_bias_to_fp16, weight = module_layers_23_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1217_cast_fp16)[name = string("linear_208_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1221_cast_fp16 = silu(x = linear_208_cast_fp16)[name = string("input_1221_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_23_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(323178048))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325275264))))[name = string("module_layers_23_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_23_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_23_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325537472)))];
            tensor<fp16, [1, 188, 1024]> linear_209_cast_fp16 = linear(bias = module_layers_23_feed_forward1_linear2_bias_to_fp16, weight = module_layers_23_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1221_cast_fp16)[name = string("linear_209_cast_fp16")];
            fp16 var_4384_to_fp16 = const()[name = string("op_4384_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4385_cast_fp16 = mul(x = linear_209_cast_fp16, y = var_4384_to_fp16)[name = string("op_4385_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1227_cast_fp16 = add(x = input_1215_cast_fp16, y = var_4385_cast_fp16)[name = string("input_1227_cast_fp16")];
            tensor<int32, [1]> query_47_axes_0 = const()[name = string("query_47_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_23_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325539584)))];
            tensor<fp16, [1024]> module_layers_23_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325541696)))];
            tensor<fp16, [1, 188, 1024]> query_47_cast_fp16 = layer_norm(axes = query_47_axes_0, beta = module_layers_23_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_self_att_weight_to_fp16, x = input_1227_cast_fp16)[name = string("query_47_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(325543808))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326068160))))[name = string("module_layers_23_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_23_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_23_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326133760)))];
            tensor<fp16, [1, 188, 1024]> linear_210_cast_fp16 = linear(bias = module_layers_23_self_attn_linear_q_bias_to_fp16, weight = module_layers_23_self_attn_linear_q_weight_to_fp16_quantized, x = query_47_cast_fp16)[name = string("linear_210_cast_fp16")];
            tensor<int32, [4]> var_4402 = const()[name = string("op_4402"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_139_cast_fp16 = reshape(shape = var_4402, x = linear_210_cast_fp16)[name = string("q_139_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326135872))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326660224))))[name = string("module_layers_23_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_23_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_23_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326725824)))];
            tensor<fp16, [1, 188, 1024]> linear_211_cast_fp16 = linear(bias = module_layers_23_self_attn_linear_k_bias_to_fp16, weight = module_layers_23_self_attn_linear_k_weight_to_fp16_quantized, x = query_47_cast_fp16)[name = string("linear_211_cast_fp16")];
            tensor<int32, [4]> var_4407 = const()[name = string("op_4407"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_93_cast_fp16 = reshape(shape = var_4407, x = linear_211_cast_fp16)[name = string("k_93_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(326727936))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327252288))))[name = string("module_layers_23_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_23_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_23_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327317888)))];
            tensor<fp16, [1, 188, 1024]> linear_212_cast_fp16 = linear(bias = module_layers_23_self_attn_linear_v_bias_to_fp16, weight = module_layers_23_self_attn_linear_v_weight_to_fp16_quantized, x = query_47_cast_fp16)[name = string("linear_212_cast_fp16")];
            tensor<int32, [4]> var_4412 = const()[name = string("op_4412"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_47_cast_fp16 = reshape(shape = var_4412, x = linear_212_cast_fp16)[name = string("v_47_cast_fp16")];
            tensor<int32, [4]> value_49_perm_0 = const()[name = string("value_49_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327320000))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327320576))))[name = string("module_layers_23_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4424_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_4424_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_23_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327320704))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327321280))))[name = string("module_layers_23_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4426_cast_fp16 = add(x = q_139_cast_fp16, y = module_layers_23_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_4426_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_47_perm_0 = const()[name = string("q_with_bias_v_47_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_513_transpose_x_0 = const()[name = string("x_513_transpose_x_0"), val = bool(false)];
            bool x_513_transpose_y_0 = const()[name = string("x_513_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_4428_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327321408))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327513472))))[name = string("op_4428_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_47_cast_fp16 = transpose(perm = q_with_bias_v_47_perm_0, x = var_4426_cast_fp16)[name = string("transpose_255")];
            tensor<fp16, [1, 8, 188, 375]> x_513_cast_fp16 = matmul(transpose_x = x_513_transpose_x_0, transpose_y = x_513_transpose_y_0, x = q_with_bias_v_47_cast_fp16, y = op_4428_to_fp16_quantized)[name = string("x_513_cast_fp16")];
            tensor<int32, [8]> x_515_pad_0 = const()[name = string("x_515_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_515_mode_0 = const()[name = string("x_515_mode_0"), val = string("constant")];
            fp16 const_300_to_fp16 = const()[name = string("const_300_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_515_cast_fp16 = pad(constant_val = const_300_to_fp16, mode = x_515_mode_0, pad = x_515_pad_0, x = x_513_cast_fp16)[name = string("x_515_cast_fp16")];
            tensor<int32, [4]> var_4436 = const()[name = string("op_4436"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_517_cast_fp16 = reshape(shape = var_4436, x = x_515_cast_fp16)[name = string("x_517_cast_fp16")];
            tensor<int32, [4]> var_4440_begin_0 = const()[name = string("op_4440_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4440_end_0 = const()[name = string("op_4440_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4440_end_mask_0 = const()[name = string("op_4440_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4440_cast_fp16 = slice_by_index(begin = var_4440_begin_0, end = var_4440_end_0, end_mask = var_4440_end_mask_0, x = x_517_cast_fp16)[name = string("op_4440_cast_fp16")];
            tensor<int32, [4]> var_4441 = const()[name = string("op_4441"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_93_cast_fp16 = reshape(shape = var_4441, x = var_4440_cast_fp16)[name = string("matrix_bd_93_cast_fp16")];
            bool matrix_ac_47_transpose_x_0 = const()[name = string("matrix_ac_47_transpose_x_0"), val = bool(false)];
            bool matrix_ac_47_transpose_y_0 = const()[name = string("matrix_ac_47_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_174_perm_0 = const()[name = string("transpose_174_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_175_perm_0 = const()[name = string("transpose_175_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_175 = transpose(perm = transpose_175_perm_0, x = k_93_cast_fp16)[name = string("transpose_253")];
            tensor<fp16, [1, 8, 188, 128]> transpose_174 = transpose(perm = transpose_174_perm_0, x = var_4424_cast_fp16)[name = string("transpose_254")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_47_cast_fp16 = matmul(transpose_x = matrix_ac_47_transpose_x_0, transpose_y = matrix_ac_47_transpose_y_0, x = transpose_174, y = transpose_175)[name = string("matrix_ac_47_cast_fp16")];
            tensor<int32, [4]> matrix_bd_95_begin_0 = const()[name = string("matrix_bd_95_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_95_end_0 = const()[name = string("matrix_bd_95_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_95_end_mask_0 = const()[name = string("matrix_bd_95_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_95_cast_fp16 = slice_by_index(begin = matrix_bd_95_begin_0, end = matrix_bd_95_end_0, end_mask = matrix_bd_95_end_mask_0, x = matrix_bd_93_cast_fp16)[name = string("matrix_bd_95_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4450_cast_fp16 = add(x = matrix_ac_47_cast_fp16, y = matrix_bd_95_cast_fp16)[name = string("op_4450_cast_fp16")];
            fp16 _inversed_scores_93_y_0_to_fp16 = const()[name = string("_inversed_scores_93_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_93_cast_fp16 = mul(x = var_4450_cast_fp16, y = _inversed_scores_93_y_0_to_fp16)[name = string("_inversed_scores_93_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_95_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_93_cast_fp16, cond = mask_11)[name = string("scores_95_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4456_cast_fp16 = softmax(axis = var_23, x = scores_95_cast_fp16)[name = string("op_4456_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1229_cast_fp16 = select(a = var_11_to_fp16, b = var_4456_cast_fp16, cond = mask_11)[name = string("input_1229_cast_fp16")];
            bool x_519_transpose_x_0 = const()[name = string("x_519_transpose_x_0"), val = bool(false)];
            bool x_519_transpose_y_0 = const()[name = string("x_519_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_49_cast_fp16 = transpose(perm = value_49_perm_0, x = v_47_cast_fp16)[name = string("transpose_252")];
            tensor<fp16, [1, 8, 188, 128]> x_519_cast_fp16 = matmul(transpose_x = x_519_transpose_x_0, transpose_y = x_519_transpose_y_0, x = input_1229_cast_fp16, y = value_49_cast_fp16)[name = string("x_519_cast_fp16")];
            tensor<int32, [4]> var_4460_perm_0 = const()[name = string("op_4460_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4461 = const()[name = string("op_4461"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4460_cast_fp16 = transpose(perm = var_4460_perm_0, x = x_519_cast_fp16)[name = string("transpose_251")];
            tensor<fp16, [1, 188, 1024]> input_1231_cast_fp16 = reshape(shape = var_4461, x = var_4460_cast_fp16)[name = string("input_1231_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_23_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(327516544))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328040896))))[name = string("module_layers_23_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_23_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_23_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328106496)))];
            tensor<fp16, [1, 188, 1024]> linear_214_cast_fp16 = linear(bias = module_layers_23_self_attn_linear_out_bias_to_fp16, weight = module_layers_23_self_attn_linear_out_weight_to_fp16_quantized, x = input_1231_cast_fp16)[name = string("linear_214_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1235_cast_fp16 = add(x = input_1227_cast_fp16, y = linear_214_cast_fp16)[name = string("input_1235_cast_fp16")];
            tensor<int32, [1]> x_523_axes_0 = const()[name = string("x_523_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_23_norm_conv_weight_to_fp16 = const()[name = string("module_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328108608)))];
            tensor<fp16, [1024]> module_layers_23_norm_conv_bias_to_fp16 = const()[name = string("module_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328110720)))];
            tensor<fp16, [1, 188, 1024]> x_523_cast_fp16 = layer_norm(axes = x_523_axes_0, beta = module_layers_23_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_conv_weight_to_fp16, x = input_1235_cast_fp16)[name = string("x_523_cast_fp16")];
            tensor<int32, [3]> input_1237_perm_0 = const()[name = string("input_1237_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1239_pad_type_0 = const()[name = string("input_1239_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1239_strides_0 = const()[name = string("input_1239_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1239_pad_0 = const()[name = string("input_1239_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1239_dilations_0 = const()[name = string("input_1239_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1239_groups_0 = const()[name = string("input_1239_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_23_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328112832))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329161472))))[name = string("module_layers_23_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_23_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_23_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329292608)))];
            tensor<fp16, [1, 1024, 188]> input_1237_cast_fp16 = transpose(perm = input_1237_perm_0, x = x_523_cast_fp16)[name = string("transpose_250")];
            tensor<fp16, [1, 2048, 188]> input_1239_cast_fp16 = conv(bias = module_layers_23_conv_pointwise_conv1_bias_to_fp16, dilations = input_1239_dilations_0, groups = input_1239_groups_0, pad = input_1239_pad_0, pad_type = input_1239_pad_type_0, strides = input_1239_strides_0, weight = module_layers_23_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1237_cast_fp16)[name = string("input_1239_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, 188]> x_525_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_525_split_cast_fp16_1 = split(axis = x_525_split_axis_0, num_splits = x_525_split_num_splits_0, x = input_1239_cast_fp16)[name = string("x_525_split_cast_fp16")];
            tensor<fp16, [1, 1024, 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, 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, 188]> input_1241_cast_fp16 = select(a = var_11_to_fp16, b = x_525_cast_fp16, cond = var_483)[name = string("input_1241_cast_fp16")];
            tensor<int32, [6]> input_1243_pad_0 = const()[name = string("input_1243_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1243_mode_0 = const()[name = string("input_1243_mode_0"), val = string("constant")];
            fp16 const_303_to_fp16 = const()[name = string("const_303_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1243_cast_fp16 = pad(constant_val = const_303_to_fp16, mode = input_1243_mode_0, pad = input_1243_pad_0, x = input_1241_cast_fp16)[name = string("input_1243_cast_fp16")];
            string input_1245_pad_type_0 = const()[name = string("input_1245_pad_type_0"), val = string("valid")];
            int32 input_1245_groups_0 = const()[name = string("input_1245_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1245_strides_0 = const()[name = string("input_1245_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1245_pad_0 = const()[name = string("input_1245_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1245_dilations_0 = const()[name = string("input_1245_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_430_to_fp16 = const()[name = string("const_430_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329296768)))];
            tensor<fp16, [1024]> const_431_to_fp16 = const()[name = string("const_431_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329315264)))];
            tensor<fp16, [1, 1024, 188]> input_1247_cast_fp16 = conv(bias = const_431_to_fp16, dilations = input_1245_dilations_0, groups = input_1245_groups_0, pad = input_1245_pad_0, pad_type = input_1245_pad_type_0, strides = input_1245_strides_0, weight = const_430_to_fp16, x = input_1243_cast_fp16)[name = string("input_1247_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1249_cast_fp16 = silu(x = input_1247_cast_fp16)[name = string("input_1249_cast_fp16")];
            string x_527_pad_type_0 = const()[name = string("x_527_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_527_strides_0 = const()[name = string("x_527_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_527_pad_0 = const()[name = string("x_527_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_527_dilations_0 = const()[name = string("x_527_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_527_groups_0 = const()[name = string("x_527_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_23_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329317376))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329841728))))[name = string("module_layers_23_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_23_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_23_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329907328)))];
            tensor<fp16, [1, 1024, 188]> x_527_cast_fp16 = conv(bias = module_layers_23_conv_pointwise_conv2_bias_to_fp16, dilations = x_527_dilations_0, groups = x_527_groups_0, pad = x_527_pad_0, pad_type = x_527_pad_type_0, strides = x_527_strides_0, weight = module_layers_23_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1249_cast_fp16)[name = string("x_527_cast_fp16")];
            tensor<int32, [3]> input_1251_perm_0 = const()[name = string("input_1251_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1251_cast_fp16 = transpose(perm = input_1251_perm_0, x = x_527_cast_fp16)[name = string("transpose_249")];
            tensor<fp16, [1, 188, 1024]> input_1253_cast_fp16 = add(x = input_1235_cast_fp16, y = input_1251_cast_fp16)[name = string("input_1253_cast_fp16")];
            tensor<int32, [1]> input_1255_axes_0 = const()[name = string("input_1255_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329909440)))];
            tensor<fp16, [1024]> module_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329911552)))];
            tensor<fp16, [1, 188, 1024]> input_1255_cast_fp16 = layer_norm(axes = input_1255_axes_0, beta = module_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_feed_forward2_weight_to_fp16, x = input_1253_cast_fp16)[name = string("input_1255_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_23_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329913664))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332010880))))[name = string("module_layers_23_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_23_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_23_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332273088)))];
            tensor<fp16, [1, 188, 4096]> linear_215_cast_fp16 = linear(bias = module_layers_23_feed_forward2_linear1_bias_to_fp16, weight = module_layers_23_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1255_cast_fp16)[name = string("linear_215_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1259_cast_fp16 = silu(x = linear_215_cast_fp16)[name = string("input_1259_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_23_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332281344))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334378560))))[name = string("module_layers_23_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_23_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_23_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334640768)))];
            tensor<fp16, [1, 188, 1024]> linear_216_cast_fp16 = linear(bias = module_layers_23_feed_forward2_linear2_bias_to_fp16, weight = module_layers_23_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1259_cast_fp16)[name = string("linear_216_cast_fp16")];
            fp16 var_4527_to_fp16 = const()[name = string("op_4527_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4528_cast_fp16 = mul(x = linear_216_cast_fp16, y = var_4527_to_fp16)[name = string("op_4528_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1265_cast_fp16 = add(x = input_1253_cast_fp16, y = var_4528_cast_fp16)[name = string("input_1265_cast_fp16")];
            tensor<int32, [1]> input_1267_axes_0 = const()[name = string("input_1267_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_23_norm_out_weight_to_fp16 = const()[name = string("module_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334642880)))];
            tensor<fp16, [1024]> module_layers_23_norm_out_bias_to_fp16 = const()[name = string("module_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334644992)))];
            tensor<fp16, [1, 188, 1024]> input_1267_cast_fp16 = layer_norm(axes = input_1267_axes_0, beta = module_layers_23_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_23_norm_out_weight_to_fp16, x = input_1265_cast_fp16)[name = string("input_1267_cast_fp16")];
            tensor<int32, [1]> input_1269_axes_0 = const()[name = string("input_1269_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_24_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_24_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334647104)))];
            tensor<fp16, [1024]> module_layers_24_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_24_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334649216)))];
            tensor<fp16, [1, 188, 1024]> input_1269_cast_fp16 = layer_norm(axes = input_1269_axes_0, beta = module_layers_24_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_24_norm_feed_forward1_weight_to_fp16, x = input_1267_cast_fp16)[name = string("input_1269_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_24_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(334651328))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(336748544))))[name = string("module_layers_24_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_24_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_24_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337010752)))];
            tensor<fp16, [1, 188, 4096]> linear_217_cast_fp16 = linear(bias = module_layers_24_feed_forward1_linear1_bias_to_fp16, weight = module_layers_24_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1269_cast_fp16)[name = string("linear_217_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1273_cast_fp16 = silu(x = linear_217_cast_fp16)[name = string("input_1273_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_24_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337019008))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339116224))))[name = string("module_layers_24_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_24_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_24_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339378432)))];
            tensor<fp16, [1, 188, 1024]> linear_218_cast_fp16 = linear(bias = module_layers_24_feed_forward1_linear2_bias_to_fp16, weight = module_layers_24_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1273_cast_fp16)[name = string("linear_218_cast_fp16")];
            fp16 var_4558_to_fp16 = const()[name = string("op_4558_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4559_cast_fp16 = mul(x = linear_218_cast_fp16, y = var_4558_to_fp16)[name = string("op_4559_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1279_cast_fp16 = add(x = input_1267_cast_fp16, y = var_4559_cast_fp16)[name = string("input_1279_cast_fp16")];
            tensor<int32, [1]> query_49_axes_0 = const()[name = string("query_49_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_24_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_24_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339380544)))];
            tensor<fp16, [1024]> module_layers_24_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_24_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339382656)))];
            tensor<fp16, [1, 188, 1024]> query_49_cast_fp16 = layer_norm(axes = query_49_axes_0, beta = module_layers_24_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_24_norm_self_att_weight_to_fp16, x = input_1279_cast_fp16)[name = string("query_49_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_24_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339384768))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339909120))))[name = string("module_layers_24_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_24_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_24_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339974720)))];
            tensor<fp16, [1, 188, 1024]> linear_219_cast_fp16 = linear(bias = module_layers_24_self_attn_linear_q_bias_to_fp16, weight = module_layers_24_self_attn_linear_q_weight_to_fp16_quantized, x = query_49_cast_fp16)[name = string("linear_219_cast_fp16")];
            tensor<int32, [4]> var_4576 = const()[name = string("op_4576"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_145_cast_fp16 = reshape(shape = var_4576, x = linear_219_cast_fp16)[name = string("q_145_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_24_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339976832))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(340501184))))[name = string("module_layers_24_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_24_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_24_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(340566784)))];
            tensor<fp16, [1, 188, 1024]> linear_220_cast_fp16 = linear(bias = module_layers_24_self_attn_linear_k_bias_to_fp16, weight = module_layers_24_self_attn_linear_k_weight_to_fp16_quantized, x = query_49_cast_fp16)[name = string("linear_220_cast_fp16")];
            tensor<int32, [4]> var_4581 = const()[name = string("op_4581"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_97_cast_fp16 = reshape(shape = var_4581, x = linear_220_cast_fp16)[name = string("k_97_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_24_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(340568896))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341093248))))[name = string("module_layers_24_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_24_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_24_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341158848)))];
            tensor<fp16, [1, 188, 1024]> linear_221_cast_fp16 = linear(bias = module_layers_24_self_attn_linear_v_bias_to_fp16, weight = module_layers_24_self_attn_linear_v_weight_to_fp16_quantized, x = query_49_cast_fp16)[name = string("linear_221_cast_fp16")];
            tensor<int32, [4]> var_4586 = const()[name = string("op_4586"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_49_cast_fp16 = reshape(shape = var_4586, x = linear_221_cast_fp16)[name = string("v_49_cast_fp16")];
            tensor<int32, [4]> value_51_perm_0 = const()[name = string("value_51_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_24_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341160960))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341161536))))[name = string("module_layers_24_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4598_cast_fp16 = add(x = q_145_cast_fp16, y = module_layers_24_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_4598_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_24_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341161664))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341162240))))[name = string("module_layers_24_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4600_cast_fp16 = add(x = q_145_cast_fp16, y = module_layers_24_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_4600_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_49_perm_0 = const()[name = string("q_with_bias_v_49_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_535_transpose_x_0 = const()[name = string("x_535_transpose_x_0"), val = bool(false)];
            bool x_535_transpose_y_0 = const()[name = string("x_535_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_4602_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341162368))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341354432))))[name = string("op_4602_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_49_cast_fp16 = transpose(perm = q_with_bias_v_49_perm_0, x = var_4600_cast_fp16)[name = string("transpose_248")];
            tensor<fp16, [1, 8, 188, 375]> x_535_cast_fp16 = matmul(transpose_x = x_535_transpose_x_0, transpose_y = x_535_transpose_y_0, x = q_with_bias_v_49_cast_fp16, y = op_4602_to_fp16_quantized)[name = string("x_535_cast_fp16")];
            tensor<int32, [8]> x_537_pad_0 = const()[name = string("x_537_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_537_mode_0 = const()[name = string("x_537_mode_0"), val = string("constant")];
            fp16 const_310_to_fp16 = const()[name = string("const_310_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_537_cast_fp16 = pad(constant_val = const_310_to_fp16, mode = x_537_mode_0, pad = x_537_pad_0, x = x_535_cast_fp16)[name = string("x_537_cast_fp16")];
            tensor<int32, [4]> var_4610 = const()[name = string("op_4610"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_539_cast_fp16 = reshape(shape = var_4610, x = x_537_cast_fp16)[name = string("x_539_cast_fp16")];
            tensor<int32, [4]> var_4614_begin_0 = const()[name = string("op_4614_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4614_end_0 = const()[name = string("op_4614_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4614_end_mask_0 = const()[name = string("op_4614_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4614_cast_fp16 = slice_by_index(begin = var_4614_begin_0, end = var_4614_end_0, end_mask = var_4614_end_mask_0, x = x_539_cast_fp16)[name = string("op_4614_cast_fp16")];
            tensor<int32, [4]> var_4615 = const()[name = string("op_4615"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_97_cast_fp16 = reshape(shape = var_4615, x = var_4614_cast_fp16)[name = string("matrix_bd_97_cast_fp16")];
            bool matrix_ac_49_transpose_x_0 = const()[name = string("matrix_ac_49_transpose_x_0"), val = bool(false)];
            bool matrix_ac_49_transpose_y_0 = const()[name = string("matrix_ac_49_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_176_perm_0 = const()[name = string("transpose_176_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_177_perm_0 = const()[name = string("transpose_177_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_177 = transpose(perm = transpose_177_perm_0, x = k_97_cast_fp16)[name = string("transpose_246")];
            tensor<fp16, [1, 8, 188, 128]> transpose_176 = transpose(perm = transpose_176_perm_0, x = var_4598_cast_fp16)[name = string("transpose_247")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_49_cast_fp16 = matmul(transpose_x = matrix_ac_49_transpose_x_0, transpose_y = matrix_ac_49_transpose_y_0, x = transpose_176, y = transpose_177)[name = string("matrix_ac_49_cast_fp16")];
            tensor<int32, [4]> matrix_bd_99_begin_0 = const()[name = string("matrix_bd_99_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_99_end_0 = const()[name = string("matrix_bd_99_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_99_end_mask_0 = const()[name = string("matrix_bd_99_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_99_cast_fp16 = slice_by_index(begin = matrix_bd_99_begin_0, end = matrix_bd_99_end_0, end_mask = matrix_bd_99_end_mask_0, x = matrix_bd_97_cast_fp16)[name = string("matrix_bd_99_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4624_cast_fp16 = add(x = matrix_ac_49_cast_fp16, y = matrix_bd_99_cast_fp16)[name = string("op_4624_cast_fp16")];
            fp16 _inversed_scores_97_y_0_to_fp16 = const()[name = string("_inversed_scores_97_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_97_cast_fp16 = mul(x = var_4624_cast_fp16, y = _inversed_scores_97_y_0_to_fp16)[name = string("_inversed_scores_97_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_99_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_97_cast_fp16, cond = mask_11)[name = string("scores_99_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4630_cast_fp16 = softmax(axis = var_23, x = scores_99_cast_fp16)[name = string("op_4630_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1281_cast_fp16 = select(a = var_11_to_fp16, b = var_4630_cast_fp16, cond = mask_11)[name = string("input_1281_cast_fp16")];
            bool x_541_transpose_x_0 = const()[name = string("x_541_transpose_x_0"), val = bool(false)];
            bool x_541_transpose_y_0 = const()[name = string("x_541_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_51_cast_fp16 = transpose(perm = value_51_perm_0, x = v_49_cast_fp16)[name = string("transpose_245")];
            tensor<fp16, [1, 8, 188, 128]> x_541_cast_fp16 = matmul(transpose_x = x_541_transpose_x_0, transpose_y = x_541_transpose_y_0, x = input_1281_cast_fp16, y = value_51_cast_fp16)[name = string("x_541_cast_fp16")];
            tensor<int32, [4]> var_4634_perm_0 = const()[name = string("op_4634_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4635 = const()[name = string("op_4635"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4634_cast_fp16 = transpose(perm = var_4634_perm_0, x = x_541_cast_fp16)[name = string("transpose_244")];
            tensor<fp16, [1, 188, 1024]> input_1283_cast_fp16 = reshape(shape = var_4635, x = var_4634_cast_fp16)[name = string("input_1283_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_24_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341357504))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341881856))))[name = string("module_layers_24_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_24_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_24_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341947456)))];
            tensor<fp16, [1, 188, 1024]> linear_223_cast_fp16 = linear(bias = module_layers_24_self_attn_linear_out_bias_to_fp16, weight = module_layers_24_self_attn_linear_out_weight_to_fp16_quantized, x = input_1283_cast_fp16)[name = string("linear_223_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1287_cast_fp16 = add(x = input_1279_cast_fp16, y = linear_223_cast_fp16)[name = string("input_1287_cast_fp16")];
            tensor<int32, [1]> x_545_axes_0 = const()[name = string("x_545_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_24_norm_conv_weight_to_fp16 = const()[name = string("module_layers_24_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341949568)))];
            tensor<fp16, [1024]> module_layers_24_norm_conv_bias_to_fp16 = const()[name = string("module_layers_24_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341951680)))];
            tensor<fp16, [1, 188, 1024]> x_545_cast_fp16 = layer_norm(axes = x_545_axes_0, beta = module_layers_24_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_24_norm_conv_weight_to_fp16, x = input_1287_cast_fp16)[name = string("x_545_cast_fp16")];
            tensor<int32, [3]> input_1289_perm_0 = const()[name = string("input_1289_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1291_pad_type_0 = const()[name = string("input_1291_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1291_strides_0 = const()[name = string("input_1291_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1291_pad_0 = const()[name = string("input_1291_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1291_dilations_0 = const()[name = string("input_1291_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1291_groups_0 = const()[name = string("input_1291_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_24_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341953792))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343002432))))[name = string("module_layers_24_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_24_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_24_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343133568)))];
            tensor<fp16, [1, 1024, 188]> input_1289_cast_fp16 = transpose(perm = input_1289_perm_0, x = x_545_cast_fp16)[name = string("transpose_243")];
            tensor<fp16, [1, 2048, 188]> input_1291_cast_fp16 = conv(bias = module_layers_24_conv_pointwise_conv1_bias_to_fp16, dilations = input_1291_dilations_0, groups = input_1291_groups_0, pad = input_1291_pad_0, pad_type = input_1291_pad_type_0, strides = input_1291_strides_0, weight = module_layers_24_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1289_cast_fp16)[name = string("input_1291_cast_fp16")];
            int32 x_547_split_num_splits_0 = const()[name = string("x_547_split_num_splits_0"), val = int32(2)];
            int32 x_547_split_axis_0 = const()[name = string("x_547_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> x_547_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_547_split_cast_fp16_1 = split(axis = x_547_split_axis_0, num_splits = x_547_split_num_splits_0, x = input_1291_cast_fp16)[name = string("x_547_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_547_split_1_sigmoid_cast_fp16 = sigmoid(x = x_547_split_cast_fp16_1)[name = string("x_547_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_547_cast_fp16 = mul(x = x_547_split_cast_fp16_0, y = x_547_split_1_sigmoid_cast_fp16)[name = string("x_547_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1293_cast_fp16 = select(a = var_11_to_fp16, b = x_547_cast_fp16, cond = var_483)[name = string("input_1293_cast_fp16")];
            tensor<int32, [6]> input_1295_pad_0 = const()[name = string("input_1295_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1295_mode_0 = const()[name = string("input_1295_mode_0"), val = string("constant")];
            fp16 const_313_to_fp16 = const()[name = string("const_313_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1295_cast_fp16 = pad(constant_val = const_313_to_fp16, mode = input_1295_mode_0, pad = input_1295_pad_0, x = input_1293_cast_fp16)[name = string("input_1295_cast_fp16")];
            string input_1297_pad_type_0 = const()[name = string("input_1297_pad_type_0"), val = string("valid")];
            int32 input_1297_groups_0 = const()[name = string("input_1297_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1297_strides_0 = const()[name = string("input_1297_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1297_pad_0 = const()[name = string("input_1297_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1297_dilations_0 = const()[name = string("input_1297_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_432_to_fp16 = const()[name = string("const_432_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343137728)))];
            tensor<fp16, [1024]> const_433_to_fp16 = const()[name = string("const_433_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343156224)))];
            tensor<fp16, [1, 1024, 188]> input_1299_cast_fp16 = conv(bias = const_433_to_fp16, dilations = input_1297_dilations_0, groups = input_1297_groups_0, pad = input_1297_pad_0, pad_type = input_1297_pad_type_0, strides = input_1297_strides_0, weight = const_432_to_fp16, x = input_1295_cast_fp16)[name = string("input_1299_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1301_cast_fp16 = silu(x = input_1299_cast_fp16)[name = string("input_1301_cast_fp16")];
            string x_549_pad_type_0 = const()[name = string("x_549_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_549_strides_0 = const()[name = string("x_549_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_549_pad_0 = const()[name = string("x_549_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_549_dilations_0 = const()[name = string("x_549_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_549_groups_0 = const()[name = string("x_549_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_24_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343158336))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343682688))))[name = string("module_layers_24_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_24_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_24_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343748288)))];
            tensor<fp16, [1, 1024, 188]> x_549_cast_fp16 = conv(bias = module_layers_24_conv_pointwise_conv2_bias_to_fp16, dilations = x_549_dilations_0, groups = x_549_groups_0, pad = x_549_pad_0, pad_type = x_549_pad_type_0, strides = x_549_strides_0, weight = module_layers_24_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1301_cast_fp16)[name = string("x_549_cast_fp16")];
            tensor<int32, [3]> input_1303_perm_0 = const()[name = string("input_1303_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1303_cast_fp16 = transpose(perm = input_1303_perm_0, x = x_549_cast_fp16)[name = string("transpose_242")];
            tensor<fp16, [1, 188, 1024]> input_1305_cast_fp16 = add(x = input_1287_cast_fp16, y = input_1303_cast_fp16)[name = string("input_1305_cast_fp16")];
            tensor<int32, [1]> input_1307_axes_0 = const()[name = string("input_1307_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_24_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_24_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343750400)))];
            tensor<fp16, [1024]> module_layers_24_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_24_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343752512)))];
            tensor<fp16, [1, 188, 1024]> input_1307_cast_fp16 = layer_norm(axes = input_1307_axes_0, beta = module_layers_24_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_24_norm_feed_forward2_weight_to_fp16, x = input_1305_cast_fp16)[name = string("input_1307_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_24_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343754624))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345851840))))[name = string("module_layers_24_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_24_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_24_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(346114048)))];
            tensor<fp16, [1, 188, 4096]> linear_224_cast_fp16 = linear(bias = module_layers_24_feed_forward2_linear1_bias_to_fp16, weight = module_layers_24_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1307_cast_fp16)[name = string("linear_224_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1311_cast_fp16 = silu(x = linear_224_cast_fp16)[name = string("input_1311_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_24_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(346122304))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348219520))))[name = string("module_layers_24_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_24_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_24_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348481728)))];
            tensor<fp16, [1, 188, 1024]> linear_225_cast_fp16 = linear(bias = module_layers_24_feed_forward2_linear2_bias_to_fp16, weight = module_layers_24_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1311_cast_fp16)[name = string("linear_225_cast_fp16")];
            fp16 var_4701_to_fp16 = const()[name = string("op_4701_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4702_cast_fp16 = mul(x = linear_225_cast_fp16, y = var_4701_to_fp16)[name = string("op_4702_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1317_cast_fp16 = add(x = input_1305_cast_fp16, y = var_4702_cast_fp16)[name = string("input_1317_cast_fp16")];
            tensor<int32, [1]> input_1319_axes_0 = const()[name = string("input_1319_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_24_norm_out_weight_to_fp16 = const()[name = string("module_layers_24_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348483840)))];
            tensor<fp16, [1024]> module_layers_24_norm_out_bias_to_fp16 = const()[name = string("module_layers_24_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348485952)))];
            tensor<fp16, [1, 188, 1024]> input_1319_cast_fp16 = layer_norm(axes = input_1319_axes_0, beta = module_layers_24_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_24_norm_out_weight_to_fp16, x = input_1317_cast_fp16)[name = string("input_1319_cast_fp16")];
            tensor<int32, [1]> input_1321_axes_0 = const()[name = string("input_1321_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_25_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_25_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348488064)))];
            tensor<fp16, [1024]> module_layers_25_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_25_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348490176)))];
            tensor<fp16, [1, 188, 1024]> input_1321_cast_fp16 = layer_norm(axes = input_1321_axes_0, beta = module_layers_25_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_25_norm_feed_forward1_weight_to_fp16, x = input_1319_cast_fp16)[name = string("input_1321_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_25_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(348492288))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350589504))))[name = string("module_layers_25_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_25_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_25_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350851712)))];
            tensor<fp16, [1, 188, 4096]> linear_226_cast_fp16 = linear(bias = module_layers_25_feed_forward1_linear1_bias_to_fp16, weight = module_layers_25_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1321_cast_fp16)[name = string("linear_226_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1325_cast_fp16 = silu(x = linear_226_cast_fp16)[name = string("input_1325_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_25_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350859968))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(352957184))))[name = string("module_layers_25_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_25_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_25_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353219392)))];
            tensor<fp16, [1, 188, 1024]> linear_227_cast_fp16 = linear(bias = module_layers_25_feed_forward1_linear2_bias_to_fp16, weight = module_layers_25_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1325_cast_fp16)[name = string("linear_227_cast_fp16")];
            fp16 var_4732_to_fp16 = const()[name = string("op_4732_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4733_cast_fp16 = mul(x = linear_227_cast_fp16, y = var_4732_to_fp16)[name = string("op_4733_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1331_cast_fp16 = add(x = input_1319_cast_fp16, y = var_4733_cast_fp16)[name = string("input_1331_cast_fp16")];
            tensor<int32, [1]> query_51_axes_0 = const()[name = string("query_51_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_25_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_25_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353221504)))];
            tensor<fp16, [1024]> module_layers_25_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_25_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353223616)))];
            tensor<fp16, [1, 188, 1024]> query_51_cast_fp16 = layer_norm(axes = query_51_axes_0, beta = module_layers_25_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_25_norm_self_att_weight_to_fp16, x = input_1331_cast_fp16)[name = string("query_51_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_25_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353225728))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353750080))))[name = string("module_layers_25_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_25_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_25_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353815680)))];
            tensor<fp16, [1, 188, 1024]> linear_228_cast_fp16 = linear(bias = module_layers_25_self_attn_linear_q_bias_to_fp16, weight = module_layers_25_self_attn_linear_q_weight_to_fp16_quantized, x = query_51_cast_fp16)[name = string("linear_228_cast_fp16")];
            tensor<int32, [4]> var_4750 = const()[name = string("op_4750"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_151_cast_fp16 = reshape(shape = var_4750, x = linear_228_cast_fp16)[name = string("q_151_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_25_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353817792))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354342144))))[name = string("module_layers_25_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_25_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_25_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354407744)))];
            tensor<fp16, [1, 188, 1024]> linear_229_cast_fp16 = linear(bias = module_layers_25_self_attn_linear_k_bias_to_fp16, weight = module_layers_25_self_attn_linear_k_weight_to_fp16_quantized, x = query_51_cast_fp16)[name = string("linear_229_cast_fp16")];
            tensor<int32, [4]> var_4755 = const()[name = string("op_4755"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_101_cast_fp16 = reshape(shape = var_4755, x = linear_229_cast_fp16)[name = string("k_101_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_25_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354409856))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354934208))))[name = string("module_layers_25_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_25_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_25_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354999808)))];
            tensor<fp16, [1, 188, 1024]> linear_230_cast_fp16 = linear(bias = module_layers_25_self_attn_linear_v_bias_to_fp16, weight = module_layers_25_self_attn_linear_v_weight_to_fp16_quantized, x = query_51_cast_fp16)[name = string("linear_230_cast_fp16")];
            tensor<int32, [4]> var_4760 = const()[name = string("op_4760"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_51_cast_fp16 = reshape(shape = var_4760, x = linear_230_cast_fp16)[name = string("v_51_cast_fp16")];
            tensor<int32, [4]> value_53_perm_0 = const()[name = string("value_53_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_25_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355001920))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355002496))))[name = string("module_layers_25_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4772_cast_fp16 = add(x = q_151_cast_fp16, y = module_layers_25_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_4772_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_25_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355002624))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355003200))))[name = string("module_layers_25_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4774_cast_fp16 = add(x = q_151_cast_fp16, y = module_layers_25_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_4774_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_51_perm_0 = const()[name = string("q_with_bias_v_51_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_557_transpose_x_0 = const()[name = string("x_557_transpose_x_0"), val = bool(false)];
            bool x_557_transpose_y_0 = const()[name = string("x_557_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_4776_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355003328))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355195392))))[name = string("op_4776_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_51_cast_fp16 = transpose(perm = q_with_bias_v_51_perm_0, x = var_4774_cast_fp16)[name = string("transpose_241")];
            tensor<fp16, [1, 8, 188, 375]> x_557_cast_fp16 = matmul(transpose_x = x_557_transpose_x_0, transpose_y = x_557_transpose_y_0, x = q_with_bias_v_51_cast_fp16, y = op_4776_to_fp16_quantized)[name = string("x_557_cast_fp16")];
            tensor<int32, [8]> x_559_pad_0 = const()[name = string("x_559_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_559_mode_0 = const()[name = string("x_559_mode_0"), val = string("constant")];
            fp16 const_320_to_fp16 = const()[name = string("const_320_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_559_cast_fp16 = pad(constant_val = const_320_to_fp16, mode = x_559_mode_0, pad = x_559_pad_0, x = x_557_cast_fp16)[name = string("x_559_cast_fp16")];
            tensor<int32, [4]> var_4784 = const()[name = string("op_4784"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_561_cast_fp16 = reshape(shape = var_4784, x = x_559_cast_fp16)[name = string("x_561_cast_fp16")];
            tensor<int32, [4]> var_4788_begin_0 = const()[name = string("op_4788_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4788_end_0 = const()[name = string("op_4788_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4788_end_mask_0 = const()[name = string("op_4788_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4788_cast_fp16 = slice_by_index(begin = var_4788_begin_0, end = var_4788_end_0, end_mask = var_4788_end_mask_0, x = x_561_cast_fp16)[name = string("op_4788_cast_fp16")];
            tensor<int32, [4]> var_4789 = const()[name = string("op_4789"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_101_cast_fp16 = reshape(shape = var_4789, x = var_4788_cast_fp16)[name = string("matrix_bd_101_cast_fp16")];
            bool matrix_ac_51_transpose_x_0 = const()[name = string("matrix_ac_51_transpose_x_0"), val = bool(false)];
            bool matrix_ac_51_transpose_y_0 = const()[name = string("matrix_ac_51_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_178_perm_0 = const()[name = string("transpose_178_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_179_perm_0 = const()[name = string("transpose_179_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_179 = transpose(perm = transpose_179_perm_0, x = k_101_cast_fp16)[name = string("transpose_239")];
            tensor<fp16, [1, 8, 188, 128]> transpose_178 = transpose(perm = transpose_178_perm_0, x = var_4772_cast_fp16)[name = string("transpose_240")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_51_cast_fp16 = matmul(transpose_x = matrix_ac_51_transpose_x_0, transpose_y = matrix_ac_51_transpose_y_0, x = transpose_178, y = transpose_179)[name = string("matrix_ac_51_cast_fp16")];
            tensor<int32, [4]> matrix_bd_103_begin_0 = const()[name = string("matrix_bd_103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_103_end_0 = const()[name = string("matrix_bd_103_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_103_end_mask_0 = const()[name = string("matrix_bd_103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_103_cast_fp16 = slice_by_index(begin = matrix_bd_103_begin_0, end = matrix_bd_103_end_0, end_mask = matrix_bd_103_end_mask_0, x = matrix_bd_101_cast_fp16)[name = string("matrix_bd_103_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4798_cast_fp16 = add(x = matrix_ac_51_cast_fp16, y = matrix_bd_103_cast_fp16)[name = string("op_4798_cast_fp16")];
            fp16 _inversed_scores_101_y_0_to_fp16 = const()[name = string("_inversed_scores_101_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_101_cast_fp16 = mul(x = var_4798_cast_fp16, y = _inversed_scores_101_y_0_to_fp16)[name = string("_inversed_scores_101_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_103_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_101_cast_fp16, cond = mask_11)[name = string("scores_103_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4804_cast_fp16 = softmax(axis = var_23, x = scores_103_cast_fp16)[name = string("op_4804_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1333_cast_fp16 = select(a = var_11_to_fp16, b = var_4804_cast_fp16, cond = mask_11)[name = string("input_1333_cast_fp16")];
            bool x_563_transpose_x_0 = const()[name = string("x_563_transpose_x_0"), val = bool(false)];
            bool x_563_transpose_y_0 = const()[name = string("x_563_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_53_cast_fp16 = transpose(perm = value_53_perm_0, x = v_51_cast_fp16)[name = string("transpose_238")];
            tensor<fp16, [1, 8, 188, 128]> x_563_cast_fp16 = matmul(transpose_x = x_563_transpose_x_0, transpose_y = x_563_transpose_y_0, x = input_1333_cast_fp16, y = value_53_cast_fp16)[name = string("x_563_cast_fp16")];
            tensor<int32, [4]> var_4808_perm_0 = const()[name = string("op_4808_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4809 = const()[name = string("op_4809"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4808_cast_fp16 = transpose(perm = var_4808_perm_0, x = x_563_cast_fp16)[name = string("transpose_237")];
            tensor<fp16, [1, 188, 1024]> input_1335_cast_fp16 = reshape(shape = var_4809, x = var_4808_cast_fp16)[name = string("input_1335_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_25_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355198464))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355722816))))[name = string("module_layers_25_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_25_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_25_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355788416)))];
            tensor<fp16, [1, 188, 1024]> linear_232_cast_fp16 = linear(bias = module_layers_25_self_attn_linear_out_bias_to_fp16, weight = module_layers_25_self_attn_linear_out_weight_to_fp16_quantized, x = input_1335_cast_fp16)[name = string("linear_232_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1339_cast_fp16 = add(x = input_1331_cast_fp16, y = linear_232_cast_fp16)[name = string("input_1339_cast_fp16")];
            tensor<int32, [1]> x_567_axes_0 = const()[name = string("x_567_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_25_norm_conv_weight_to_fp16 = const()[name = string("module_layers_25_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355790528)))];
            tensor<fp16, [1024]> module_layers_25_norm_conv_bias_to_fp16 = const()[name = string("module_layers_25_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355792640)))];
            tensor<fp16, [1, 188, 1024]> x_567_cast_fp16 = layer_norm(axes = x_567_axes_0, beta = module_layers_25_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_25_norm_conv_weight_to_fp16, x = input_1339_cast_fp16)[name = string("x_567_cast_fp16")];
            tensor<int32, [3]> input_1341_perm_0 = const()[name = string("input_1341_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1343_pad_type_0 = const()[name = string("input_1343_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1343_strides_0 = const()[name = string("input_1343_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1343_pad_0 = const()[name = string("input_1343_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1343_dilations_0 = const()[name = string("input_1343_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1343_groups_0 = const()[name = string("input_1343_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_25_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(355794752))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356843392))))[name = string("module_layers_25_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_25_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_25_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356974528)))];
            tensor<fp16, [1, 1024, 188]> input_1341_cast_fp16 = transpose(perm = input_1341_perm_0, x = x_567_cast_fp16)[name = string("transpose_236")];
            tensor<fp16, [1, 2048, 188]> input_1343_cast_fp16 = conv(bias = module_layers_25_conv_pointwise_conv1_bias_to_fp16, dilations = input_1343_dilations_0, groups = input_1343_groups_0, pad = input_1343_pad_0, pad_type = input_1343_pad_type_0, strides = input_1343_strides_0, weight = module_layers_25_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1341_cast_fp16)[name = string("input_1343_cast_fp16")];
            int32 x_569_split_num_splits_0 = const()[name = string("x_569_split_num_splits_0"), val = int32(2)];
            int32 x_569_split_axis_0 = const()[name = string("x_569_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> x_569_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_569_split_cast_fp16_1 = split(axis = x_569_split_axis_0, num_splits = x_569_split_num_splits_0, x = input_1343_cast_fp16)[name = string("x_569_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_569_split_1_sigmoid_cast_fp16 = sigmoid(x = x_569_split_cast_fp16_1)[name = string("x_569_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_569_cast_fp16 = mul(x = x_569_split_cast_fp16_0, y = x_569_split_1_sigmoid_cast_fp16)[name = string("x_569_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1345_cast_fp16 = select(a = var_11_to_fp16, b = x_569_cast_fp16, cond = var_483)[name = string("input_1345_cast_fp16")];
            tensor<int32, [6]> input_1347_pad_0 = const()[name = string("input_1347_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1347_mode_0 = const()[name = string("input_1347_mode_0"), val = string("constant")];
            fp16 const_323_to_fp16 = const()[name = string("const_323_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1347_cast_fp16 = pad(constant_val = const_323_to_fp16, mode = input_1347_mode_0, pad = input_1347_pad_0, x = input_1345_cast_fp16)[name = string("input_1347_cast_fp16")];
            string input_1349_pad_type_0 = const()[name = string("input_1349_pad_type_0"), val = string("valid")];
            int32 input_1349_groups_0 = const()[name = string("input_1349_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1349_strides_0 = const()[name = string("input_1349_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1349_pad_0 = const()[name = string("input_1349_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1349_dilations_0 = const()[name = string("input_1349_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_434_to_fp16 = const()[name = string("const_434_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356978688)))];
            tensor<fp16, [1024]> const_435_to_fp16 = const()[name = string("const_435_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356997184)))];
            tensor<fp16, [1, 1024, 188]> input_1351_cast_fp16 = conv(bias = const_435_to_fp16, dilations = input_1349_dilations_0, groups = input_1349_groups_0, pad = input_1349_pad_0, pad_type = input_1349_pad_type_0, strides = input_1349_strides_0, weight = const_434_to_fp16, x = input_1347_cast_fp16)[name = string("input_1351_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1353_cast_fp16 = silu(x = input_1351_cast_fp16)[name = string("input_1353_cast_fp16")];
            string x_571_pad_type_0 = const()[name = string("x_571_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_571_strides_0 = const()[name = string("x_571_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_571_pad_0 = const()[name = string("x_571_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_571_dilations_0 = const()[name = string("x_571_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_571_groups_0 = const()[name = string("x_571_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_25_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356999296))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357523648))))[name = string("module_layers_25_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_25_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_25_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357589248)))];
            tensor<fp16, [1, 1024, 188]> x_571_cast_fp16 = conv(bias = module_layers_25_conv_pointwise_conv2_bias_to_fp16, dilations = x_571_dilations_0, groups = x_571_groups_0, pad = x_571_pad_0, pad_type = x_571_pad_type_0, strides = x_571_strides_0, weight = module_layers_25_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1353_cast_fp16)[name = string("x_571_cast_fp16")];
            tensor<int32, [3]> input_1355_perm_0 = const()[name = string("input_1355_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1355_cast_fp16 = transpose(perm = input_1355_perm_0, x = x_571_cast_fp16)[name = string("transpose_235")];
            tensor<fp16, [1, 188, 1024]> input_1357_cast_fp16 = add(x = input_1339_cast_fp16, y = input_1355_cast_fp16)[name = string("input_1357_cast_fp16")];
            tensor<int32, [1]> input_1359_axes_0 = const()[name = string("input_1359_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_25_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_25_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357591360)))];
            tensor<fp16, [1024]> module_layers_25_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_25_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357593472)))];
            tensor<fp16, [1, 188, 1024]> input_1359_cast_fp16 = layer_norm(axes = input_1359_axes_0, beta = module_layers_25_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_25_norm_feed_forward2_weight_to_fp16, x = input_1357_cast_fp16)[name = string("input_1359_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_25_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(357595584))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(359692800))))[name = string("module_layers_25_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_25_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_25_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(359955008)))];
            tensor<fp16, [1, 188, 4096]> linear_233_cast_fp16 = linear(bias = module_layers_25_feed_forward2_linear1_bias_to_fp16, weight = module_layers_25_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1359_cast_fp16)[name = string("linear_233_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1363_cast_fp16 = silu(x = linear_233_cast_fp16)[name = string("input_1363_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_25_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(359963264))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362060480))))[name = string("module_layers_25_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_25_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_25_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362322688)))];
            tensor<fp16, [1, 188, 1024]> linear_234_cast_fp16 = linear(bias = module_layers_25_feed_forward2_linear2_bias_to_fp16, weight = module_layers_25_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1363_cast_fp16)[name = string("linear_234_cast_fp16")];
            fp16 var_4875_to_fp16 = const()[name = string("op_4875_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4876_cast_fp16 = mul(x = linear_234_cast_fp16, y = var_4875_to_fp16)[name = string("op_4876_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1369_cast_fp16 = add(x = input_1357_cast_fp16, y = var_4876_cast_fp16)[name = string("input_1369_cast_fp16")];
            tensor<int32, [1]> input_1371_axes_0 = const()[name = string("input_1371_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_25_norm_out_weight_to_fp16 = const()[name = string("module_layers_25_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362324800)))];
            tensor<fp16, [1024]> module_layers_25_norm_out_bias_to_fp16 = const()[name = string("module_layers_25_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362326912)))];
            tensor<fp16, [1, 188, 1024]> input_1371_cast_fp16 = layer_norm(axes = input_1371_axes_0, beta = module_layers_25_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_25_norm_out_weight_to_fp16, x = input_1369_cast_fp16)[name = string("input_1371_cast_fp16")];
            tensor<int32, [1]> input_1373_axes_0 = const()[name = string("input_1373_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_26_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_26_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362329024)))];
            tensor<fp16, [1024]> module_layers_26_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_26_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362331136)))];
            tensor<fp16, [1, 188, 1024]> input_1373_cast_fp16 = layer_norm(axes = input_1373_axes_0, beta = module_layers_26_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_26_norm_feed_forward1_weight_to_fp16, x = input_1371_cast_fp16)[name = string("input_1373_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_26_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362333248))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364430464))))[name = string("module_layers_26_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_26_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_26_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364692672)))];
            tensor<fp16, [1, 188, 4096]> linear_235_cast_fp16 = linear(bias = module_layers_26_feed_forward1_linear1_bias_to_fp16, weight = module_layers_26_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1373_cast_fp16)[name = string("linear_235_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1377_cast_fp16 = silu(x = linear_235_cast_fp16)[name = string("input_1377_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_26_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364700928))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(366798144))))[name = string("module_layers_26_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_26_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_26_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367060352)))];
            tensor<fp16, [1, 188, 1024]> linear_236_cast_fp16 = linear(bias = module_layers_26_feed_forward1_linear2_bias_to_fp16, weight = module_layers_26_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1377_cast_fp16)[name = string("linear_236_cast_fp16")];
            fp16 var_4906_to_fp16 = const()[name = string("op_4906_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_4907_cast_fp16 = mul(x = linear_236_cast_fp16, y = var_4906_to_fp16)[name = string("op_4907_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1383_cast_fp16 = add(x = input_1371_cast_fp16, y = var_4907_cast_fp16)[name = string("input_1383_cast_fp16")];
            tensor<int32, [1]> query_53_axes_0 = const()[name = string("query_53_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_26_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_26_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367062464)))];
            tensor<fp16, [1024]> module_layers_26_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_26_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367064576)))];
            tensor<fp16, [1, 188, 1024]> query_53_cast_fp16 = layer_norm(axes = query_53_axes_0, beta = module_layers_26_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_26_norm_self_att_weight_to_fp16, x = input_1383_cast_fp16)[name = string("query_53_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_26_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367066688))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367591040))))[name = string("module_layers_26_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_26_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_26_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367656640)))];
            tensor<fp16, [1, 188, 1024]> linear_237_cast_fp16 = linear(bias = module_layers_26_self_attn_linear_q_bias_to_fp16, weight = module_layers_26_self_attn_linear_q_weight_to_fp16_quantized, x = query_53_cast_fp16)[name = string("linear_237_cast_fp16")];
            tensor<int32, [4]> var_4924 = const()[name = string("op_4924"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_157_cast_fp16 = reshape(shape = var_4924, x = linear_237_cast_fp16)[name = string("q_157_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_26_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(367658752))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368183104))))[name = string("module_layers_26_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_26_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_26_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368248704)))];
            tensor<fp16, [1, 188, 1024]> linear_238_cast_fp16 = linear(bias = module_layers_26_self_attn_linear_k_bias_to_fp16, weight = module_layers_26_self_attn_linear_k_weight_to_fp16_quantized, x = query_53_cast_fp16)[name = string("linear_238_cast_fp16")];
            tensor<int32, [4]> var_4929 = const()[name = string("op_4929"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_105_cast_fp16 = reshape(shape = var_4929, x = linear_238_cast_fp16)[name = string("k_105_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_26_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368250816))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368775168))))[name = string("module_layers_26_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_26_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_26_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368840768)))];
            tensor<fp16, [1, 188, 1024]> linear_239_cast_fp16 = linear(bias = module_layers_26_self_attn_linear_v_bias_to_fp16, weight = module_layers_26_self_attn_linear_v_weight_to_fp16_quantized, x = query_53_cast_fp16)[name = string("linear_239_cast_fp16")];
            tensor<int32, [4]> var_4934 = const()[name = string("op_4934"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_53_cast_fp16 = reshape(shape = var_4934, x = linear_239_cast_fp16)[name = string("v_53_cast_fp16")];
            tensor<int32, [4]> value_55_perm_0 = const()[name = string("value_55_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_26_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368842880))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368843456))))[name = string("module_layers_26_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4946_cast_fp16 = add(x = q_157_cast_fp16, y = module_layers_26_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_4946_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_26_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368843584))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368844160))))[name = string("module_layers_26_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_4948_cast_fp16 = add(x = q_157_cast_fp16, y = module_layers_26_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_4948_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_53_perm_0 = const()[name = string("q_with_bias_v_53_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_579_transpose_x_0 = const()[name = string("x_579_transpose_x_0"), val = bool(false)];
            bool x_579_transpose_y_0 = const()[name = string("x_579_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_4950_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368844288))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369036352))))[name = string("op_4950_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_53_cast_fp16 = transpose(perm = q_with_bias_v_53_perm_0, x = var_4948_cast_fp16)[name = string("transpose_234")];
            tensor<fp16, [1, 8, 188, 375]> x_579_cast_fp16 = matmul(transpose_x = x_579_transpose_x_0, transpose_y = x_579_transpose_y_0, x = q_with_bias_v_53_cast_fp16, y = op_4950_to_fp16_quantized)[name = string("x_579_cast_fp16")];
            tensor<int32, [8]> x_581_pad_0 = const()[name = string("x_581_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_581_mode_0 = const()[name = string("x_581_mode_0"), val = string("constant")];
            fp16 const_330_to_fp16 = const()[name = string("const_330_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_581_cast_fp16 = pad(constant_val = const_330_to_fp16, mode = x_581_mode_0, pad = x_581_pad_0, x = x_579_cast_fp16)[name = string("x_581_cast_fp16")];
            tensor<int32, [4]> var_4958 = const()[name = string("op_4958"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_583_cast_fp16 = reshape(shape = var_4958, x = x_581_cast_fp16)[name = string("x_583_cast_fp16")];
            tensor<int32, [4]> var_4962_begin_0 = const()[name = string("op_4962_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_4962_end_0 = const()[name = string("op_4962_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_4962_end_mask_0 = const()[name = string("op_4962_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_4962_cast_fp16 = slice_by_index(begin = var_4962_begin_0, end = var_4962_end_0, end_mask = var_4962_end_mask_0, x = x_583_cast_fp16)[name = string("op_4962_cast_fp16")];
            tensor<int32, [4]> var_4963 = const()[name = string("op_4963"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_105_cast_fp16 = reshape(shape = var_4963, x = var_4962_cast_fp16)[name = string("matrix_bd_105_cast_fp16")];
            bool matrix_ac_53_transpose_x_0 = const()[name = string("matrix_ac_53_transpose_x_0"), val = bool(false)];
            bool matrix_ac_53_transpose_y_0 = const()[name = string("matrix_ac_53_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_180_perm_0 = const()[name = string("transpose_180_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_181_perm_0 = const()[name = string("transpose_181_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_181 = transpose(perm = transpose_181_perm_0, x = k_105_cast_fp16)[name = string("transpose_232")];
            tensor<fp16, [1, 8, 188, 128]> transpose_180 = transpose(perm = transpose_180_perm_0, x = var_4946_cast_fp16)[name = string("transpose_233")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_53_cast_fp16 = matmul(transpose_x = matrix_ac_53_transpose_x_0, transpose_y = matrix_ac_53_transpose_y_0, x = transpose_180, y = transpose_181)[name = string("matrix_ac_53_cast_fp16")];
            tensor<int32, [4]> matrix_bd_107_begin_0 = const()[name = string("matrix_bd_107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_107_end_0 = const()[name = string("matrix_bd_107_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_107_end_mask_0 = const()[name = string("matrix_bd_107_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_107_cast_fp16 = slice_by_index(begin = matrix_bd_107_begin_0, end = matrix_bd_107_end_0, end_mask = matrix_bd_107_end_mask_0, x = matrix_bd_105_cast_fp16)[name = string("matrix_bd_107_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4972_cast_fp16 = add(x = matrix_ac_53_cast_fp16, y = matrix_bd_107_cast_fp16)[name = string("op_4972_cast_fp16")];
            fp16 _inversed_scores_105_y_0_to_fp16 = const()[name = string("_inversed_scores_105_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_105_cast_fp16 = mul(x = var_4972_cast_fp16, y = _inversed_scores_105_y_0_to_fp16)[name = string("_inversed_scores_105_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_107_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_105_cast_fp16, cond = mask_11)[name = string("scores_107_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_4978_cast_fp16 = softmax(axis = var_23, x = scores_107_cast_fp16)[name = string("op_4978_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1385_cast_fp16 = select(a = var_11_to_fp16, b = var_4978_cast_fp16, cond = mask_11)[name = string("input_1385_cast_fp16")];
            bool x_585_transpose_x_0 = const()[name = string("x_585_transpose_x_0"), val = bool(false)];
            bool x_585_transpose_y_0 = const()[name = string("x_585_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_55_cast_fp16 = transpose(perm = value_55_perm_0, x = v_53_cast_fp16)[name = string("transpose_231")];
            tensor<fp16, [1, 8, 188, 128]> x_585_cast_fp16 = matmul(transpose_x = x_585_transpose_x_0, transpose_y = x_585_transpose_y_0, x = input_1385_cast_fp16, y = value_55_cast_fp16)[name = string("x_585_cast_fp16")];
            tensor<int32, [4]> var_4982_perm_0 = const()[name = string("op_4982_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4983 = const()[name = string("op_4983"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_4982_cast_fp16 = transpose(perm = var_4982_perm_0, x = x_585_cast_fp16)[name = string("transpose_230")];
            tensor<fp16, [1, 188, 1024]> input_1387_cast_fp16 = reshape(shape = var_4983, x = var_4982_cast_fp16)[name = string("input_1387_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_26_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369039424))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369563776))))[name = string("module_layers_26_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_26_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_26_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369629376)))];
            tensor<fp16, [1, 188, 1024]> linear_241_cast_fp16 = linear(bias = module_layers_26_self_attn_linear_out_bias_to_fp16, weight = module_layers_26_self_attn_linear_out_weight_to_fp16_quantized, x = input_1387_cast_fp16)[name = string("linear_241_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1391_cast_fp16 = add(x = input_1383_cast_fp16, y = linear_241_cast_fp16)[name = string("input_1391_cast_fp16")];
            tensor<int32, [1]> x_589_axes_0 = const()[name = string("x_589_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_26_norm_conv_weight_to_fp16 = const()[name = string("module_layers_26_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369631488)))];
            tensor<fp16, [1024]> module_layers_26_norm_conv_bias_to_fp16 = const()[name = string("module_layers_26_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369633600)))];
            tensor<fp16, [1, 188, 1024]> x_589_cast_fp16 = layer_norm(axes = x_589_axes_0, beta = module_layers_26_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_26_norm_conv_weight_to_fp16, x = input_1391_cast_fp16)[name = string("x_589_cast_fp16")];
            tensor<int32, [3]> input_1393_perm_0 = const()[name = string("input_1393_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1395_pad_type_0 = const()[name = string("input_1395_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1395_strides_0 = const()[name = string("input_1395_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1395_pad_0 = const()[name = string("input_1395_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1395_dilations_0 = const()[name = string("input_1395_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1395_groups_0 = const()[name = string("input_1395_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_26_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369635712))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370684352))))[name = string("module_layers_26_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_26_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_26_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370815488)))];
            tensor<fp16, [1, 1024, 188]> input_1393_cast_fp16 = transpose(perm = input_1393_perm_0, x = x_589_cast_fp16)[name = string("transpose_229")];
            tensor<fp16, [1, 2048, 188]> input_1395_cast_fp16 = conv(bias = module_layers_26_conv_pointwise_conv1_bias_to_fp16, dilations = input_1395_dilations_0, groups = input_1395_groups_0, pad = input_1395_pad_0, pad_type = input_1395_pad_type_0, strides = input_1395_strides_0, weight = module_layers_26_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1393_cast_fp16)[name = string("input_1395_cast_fp16")];
            int32 x_591_split_num_splits_0 = const()[name = string("x_591_split_num_splits_0"), val = int32(2)];
            int32 x_591_split_axis_0 = const()[name = string("x_591_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> x_591_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_591_split_cast_fp16_1 = split(axis = x_591_split_axis_0, num_splits = x_591_split_num_splits_0, x = input_1395_cast_fp16)[name = string("x_591_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_591_split_1_sigmoid_cast_fp16 = sigmoid(x = x_591_split_cast_fp16_1)[name = string("x_591_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_591_cast_fp16 = mul(x = x_591_split_cast_fp16_0, y = x_591_split_1_sigmoid_cast_fp16)[name = string("x_591_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1397_cast_fp16 = select(a = var_11_to_fp16, b = x_591_cast_fp16, cond = var_483)[name = string("input_1397_cast_fp16")];
            tensor<int32, [6]> input_1399_pad_0 = const()[name = string("input_1399_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1399_mode_0 = const()[name = string("input_1399_mode_0"), val = string("constant")];
            fp16 const_333_to_fp16 = const()[name = string("const_333_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1399_cast_fp16 = pad(constant_val = const_333_to_fp16, mode = input_1399_mode_0, pad = input_1399_pad_0, x = input_1397_cast_fp16)[name = string("input_1399_cast_fp16")];
            string input_1401_pad_type_0 = const()[name = string("input_1401_pad_type_0"), val = string("valid")];
            int32 input_1401_groups_0 = const()[name = string("input_1401_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1401_strides_0 = const()[name = string("input_1401_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1401_pad_0 = const()[name = string("input_1401_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1401_dilations_0 = const()[name = string("input_1401_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_436_to_fp16 = const()[name = string("const_436_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370819648)))];
            tensor<fp16, [1024]> const_437_to_fp16 = const()[name = string("const_437_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370838144)))];
            tensor<fp16, [1, 1024, 188]> input_1403_cast_fp16 = conv(bias = const_437_to_fp16, dilations = input_1401_dilations_0, groups = input_1401_groups_0, pad = input_1401_pad_0, pad_type = input_1401_pad_type_0, strides = input_1401_strides_0, weight = const_436_to_fp16, x = input_1399_cast_fp16)[name = string("input_1403_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1405_cast_fp16 = silu(x = input_1403_cast_fp16)[name = string("input_1405_cast_fp16")];
            string x_593_pad_type_0 = const()[name = string("x_593_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_593_strides_0 = const()[name = string("x_593_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_593_pad_0 = const()[name = string("x_593_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_593_dilations_0 = const()[name = string("x_593_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_593_groups_0 = const()[name = string("x_593_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_26_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370840256))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371364608))))[name = string("module_layers_26_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_26_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_26_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371430208)))];
            tensor<fp16, [1, 1024, 188]> x_593_cast_fp16 = conv(bias = module_layers_26_conv_pointwise_conv2_bias_to_fp16, dilations = x_593_dilations_0, groups = x_593_groups_0, pad = x_593_pad_0, pad_type = x_593_pad_type_0, strides = x_593_strides_0, weight = module_layers_26_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1405_cast_fp16)[name = string("x_593_cast_fp16")];
            tensor<int32, [3]> input_1407_perm_0 = const()[name = string("input_1407_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1407_cast_fp16 = transpose(perm = input_1407_perm_0, x = x_593_cast_fp16)[name = string("transpose_228")];
            tensor<fp16, [1, 188, 1024]> input_1409_cast_fp16 = add(x = input_1391_cast_fp16, y = input_1407_cast_fp16)[name = string("input_1409_cast_fp16")];
            tensor<int32, [1]> input_1411_axes_0 = const()[name = string("input_1411_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_26_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_26_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371432320)))];
            tensor<fp16, [1024]> module_layers_26_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_26_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371434432)))];
            tensor<fp16, [1, 188, 1024]> input_1411_cast_fp16 = layer_norm(axes = input_1411_axes_0, beta = module_layers_26_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_26_norm_feed_forward2_weight_to_fp16, x = input_1409_cast_fp16)[name = string("input_1411_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_26_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371436544))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373533760))))[name = string("module_layers_26_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_26_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_26_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373795968)))];
            tensor<fp16, [1, 188, 4096]> linear_242_cast_fp16 = linear(bias = module_layers_26_feed_forward2_linear1_bias_to_fp16, weight = module_layers_26_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1411_cast_fp16)[name = string("linear_242_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1415_cast_fp16 = silu(x = linear_242_cast_fp16)[name = string("input_1415_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_26_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373804224))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375901440))))[name = string("module_layers_26_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_26_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_26_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376163648)))];
            tensor<fp16, [1, 188, 1024]> linear_243_cast_fp16 = linear(bias = module_layers_26_feed_forward2_linear2_bias_to_fp16, weight = module_layers_26_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1415_cast_fp16)[name = string("linear_243_cast_fp16")];
            fp16 var_5049_to_fp16 = const()[name = string("op_5049_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5050_cast_fp16 = mul(x = linear_243_cast_fp16, y = var_5049_to_fp16)[name = string("op_5050_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1421_cast_fp16 = add(x = input_1409_cast_fp16, y = var_5050_cast_fp16)[name = string("input_1421_cast_fp16")];
            tensor<int32, [1]> input_1423_axes_0 = const()[name = string("input_1423_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_26_norm_out_weight_to_fp16 = const()[name = string("module_layers_26_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376165760)))];
            tensor<fp16, [1024]> module_layers_26_norm_out_bias_to_fp16 = const()[name = string("module_layers_26_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376167872)))];
            tensor<fp16, [1, 188, 1024]> input_1423_cast_fp16 = layer_norm(axes = input_1423_axes_0, beta = module_layers_26_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_26_norm_out_weight_to_fp16, x = input_1421_cast_fp16)[name = string("input_1423_cast_fp16")];
            tensor<int32, [1]> input_1425_axes_0 = const()[name = string("input_1425_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_27_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_27_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376169984)))];
            tensor<fp16, [1024]> module_layers_27_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_27_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376172096)))];
            tensor<fp16, [1, 188, 1024]> input_1425_cast_fp16 = layer_norm(axes = input_1425_axes_0, beta = module_layers_27_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_27_norm_feed_forward1_weight_to_fp16, x = input_1423_cast_fp16)[name = string("input_1425_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_27_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(376174208))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378271424))))[name = string("module_layers_27_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_27_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_27_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378533632)))];
            tensor<fp16, [1, 188, 4096]> linear_244_cast_fp16 = linear(bias = module_layers_27_feed_forward1_linear1_bias_to_fp16, weight = module_layers_27_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1425_cast_fp16)[name = string("linear_244_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1429_cast_fp16 = silu(x = linear_244_cast_fp16)[name = string("input_1429_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_27_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(378541888))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380639104))))[name = string("module_layers_27_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_27_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_27_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380901312)))];
            tensor<fp16, [1, 188, 1024]> linear_245_cast_fp16 = linear(bias = module_layers_27_feed_forward1_linear2_bias_to_fp16, weight = module_layers_27_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1429_cast_fp16)[name = string("linear_245_cast_fp16")];
            fp16 var_5080_to_fp16 = const()[name = string("op_5080_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5081_cast_fp16 = mul(x = linear_245_cast_fp16, y = var_5080_to_fp16)[name = string("op_5081_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1435_cast_fp16 = add(x = input_1423_cast_fp16, y = var_5081_cast_fp16)[name = string("input_1435_cast_fp16")];
            tensor<int32, [1]> query_55_axes_0 = const()[name = string("query_55_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_27_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_27_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380903424)))];
            tensor<fp16, [1024]> module_layers_27_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_27_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380905536)))];
            tensor<fp16, [1, 188, 1024]> query_55_cast_fp16 = layer_norm(axes = query_55_axes_0, beta = module_layers_27_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_27_norm_self_att_weight_to_fp16, x = input_1435_cast_fp16)[name = string("query_55_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_27_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(380907648))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381432000))))[name = string("module_layers_27_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_27_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_27_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381497600)))];
            tensor<fp16, [1, 188, 1024]> linear_246_cast_fp16 = linear(bias = module_layers_27_self_attn_linear_q_bias_to_fp16, weight = module_layers_27_self_attn_linear_q_weight_to_fp16_quantized, x = query_55_cast_fp16)[name = string("linear_246_cast_fp16")];
            tensor<int32, [4]> var_5098 = const()[name = string("op_5098"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_163_cast_fp16 = reshape(shape = var_5098, x = linear_246_cast_fp16)[name = string("q_163_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_27_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(381499712))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382024064))))[name = string("module_layers_27_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_27_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_27_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382089664)))];
            tensor<fp16, [1, 188, 1024]> linear_247_cast_fp16 = linear(bias = module_layers_27_self_attn_linear_k_bias_to_fp16, weight = module_layers_27_self_attn_linear_k_weight_to_fp16_quantized, x = query_55_cast_fp16)[name = string("linear_247_cast_fp16")];
            tensor<int32, [4]> var_5103 = const()[name = string("op_5103"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_109_cast_fp16 = reshape(shape = var_5103, x = linear_247_cast_fp16)[name = string("k_109_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_27_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382091776))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382616128))))[name = string("module_layers_27_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_27_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_27_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382681728)))];
            tensor<fp16, [1, 188, 1024]> linear_248_cast_fp16 = linear(bias = module_layers_27_self_attn_linear_v_bias_to_fp16, weight = module_layers_27_self_attn_linear_v_weight_to_fp16_quantized, x = query_55_cast_fp16)[name = string("linear_248_cast_fp16")];
            tensor<int32, [4]> var_5108 = const()[name = string("op_5108"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_55_cast_fp16 = reshape(shape = var_5108, x = linear_248_cast_fp16)[name = string("v_55_cast_fp16")];
            tensor<int32, [4]> value_57_perm_0 = const()[name = string("value_57_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_27_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382683840))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382684416))))[name = string("module_layers_27_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5120_cast_fp16 = add(x = q_163_cast_fp16, y = module_layers_27_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_5120_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_27_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382684544))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382685120))))[name = string("module_layers_27_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5122_cast_fp16 = add(x = q_163_cast_fp16, y = module_layers_27_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_5122_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_55_perm_0 = const()[name = string("q_with_bias_v_55_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_601_transpose_x_0 = const()[name = string("x_601_transpose_x_0"), val = bool(false)];
            bool x_601_transpose_y_0 = const()[name = string("x_601_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_5124_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382685248))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382877312))))[name = string("op_5124_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_55_cast_fp16 = transpose(perm = q_with_bias_v_55_perm_0, x = var_5122_cast_fp16)[name = string("transpose_227")];
            tensor<fp16, [1, 8, 188, 375]> x_601_cast_fp16 = matmul(transpose_x = x_601_transpose_x_0, transpose_y = x_601_transpose_y_0, x = q_with_bias_v_55_cast_fp16, y = op_5124_to_fp16_quantized)[name = string("x_601_cast_fp16")];
            tensor<int32, [8]> x_603_pad_0 = const()[name = string("x_603_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_603_mode_0 = const()[name = string("x_603_mode_0"), val = string("constant")];
            fp16 const_340_to_fp16 = const()[name = string("const_340_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_603_cast_fp16 = pad(constant_val = const_340_to_fp16, mode = x_603_mode_0, pad = x_603_pad_0, x = x_601_cast_fp16)[name = string("x_603_cast_fp16")];
            tensor<int32, [4]> var_5132 = const()[name = string("op_5132"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_605_cast_fp16 = reshape(shape = var_5132, x = x_603_cast_fp16)[name = string("x_605_cast_fp16")];
            tensor<int32, [4]> var_5136_begin_0 = const()[name = string("op_5136_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5136_end_0 = const()[name = string("op_5136_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5136_end_mask_0 = const()[name = string("op_5136_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5136_cast_fp16 = slice_by_index(begin = var_5136_begin_0, end = var_5136_end_0, end_mask = var_5136_end_mask_0, x = x_605_cast_fp16)[name = string("op_5136_cast_fp16")];
            tensor<int32, [4]> var_5137 = const()[name = string("op_5137"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_109_cast_fp16 = reshape(shape = var_5137, x = var_5136_cast_fp16)[name = string("matrix_bd_109_cast_fp16")];
            bool matrix_ac_55_transpose_x_0 = const()[name = string("matrix_ac_55_transpose_x_0"), val = bool(false)];
            bool matrix_ac_55_transpose_y_0 = const()[name = string("matrix_ac_55_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_182_perm_0 = const()[name = string("transpose_182_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_183_perm_0 = const()[name = string("transpose_183_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_183 = transpose(perm = transpose_183_perm_0, x = k_109_cast_fp16)[name = string("transpose_225")];
            tensor<fp16, [1, 8, 188, 128]> transpose_182 = transpose(perm = transpose_182_perm_0, x = var_5120_cast_fp16)[name = string("transpose_226")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_55_cast_fp16 = matmul(transpose_x = matrix_ac_55_transpose_x_0, transpose_y = matrix_ac_55_transpose_y_0, x = transpose_182, y = transpose_183)[name = string("matrix_ac_55_cast_fp16")];
            tensor<int32, [4]> matrix_bd_111_begin_0 = const()[name = string("matrix_bd_111_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_111_end_0 = const()[name = string("matrix_bd_111_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_111_end_mask_0 = const()[name = string("matrix_bd_111_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_111_cast_fp16 = slice_by_index(begin = matrix_bd_111_begin_0, end = matrix_bd_111_end_0, end_mask = matrix_bd_111_end_mask_0, x = matrix_bd_109_cast_fp16)[name = string("matrix_bd_111_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5146_cast_fp16 = add(x = matrix_ac_55_cast_fp16, y = matrix_bd_111_cast_fp16)[name = string("op_5146_cast_fp16")];
            fp16 _inversed_scores_109_y_0_to_fp16 = const()[name = string("_inversed_scores_109_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_109_cast_fp16 = mul(x = var_5146_cast_fp16, y = _inversed_scores_109_y_0_to_fp16)[name = string("_inversed_scores_109_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_111_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_109_cast_fp16, cond = mask_11)[name = string("scores_111_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5152_cast_fp16 = softmax(axis = var_23, x = scores_111_cast_fp16)[name = string("op_5152_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1437_cast_fp16 = select(a = var_11_to_fp16, b = var_5152_cast_fp16, cond = mask_11)[name = string("input_1437_cast_fp16")];
            bool x_607_transpose_x_0 = const()[name = string("x_607_transpose_x_0"), val = bool(false)];
            bool x_607_transpose_y_0 = const()[name = string("x_607_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_57_cast_fp16 = transpose(perm = value_57_perm_0, x = v_55_cast_fp16)[name = string("transpose_224")];
            tensor<fp16, [1, 8, 188, 128]> x_607_cast_fp16 = matmul(transpose_x = x_607_transpose_x_0, transpose_y = x_607_transpose_y_0, x = input_1437_cast_fp16, y = value_57_cast_fp16)[name = string("x_607_cast_fp16")];
            tensor<int32, [4]> var_5156_perm_0 = const()[name = string("op_5156_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5157 = const()[name = string("op_5157"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_5156_cast_fp16 = transpose(perm = var_5156_perm_0, x = x_607_cast_fp16)[name = string("transpose_223")];
            tensor<fp16, [1, 188, 1024]> input_1439_cast_fp16 = reshape(shape = var_5157, x = var_5156_cast_fp16)[name = string("input_1439_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_27_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(382880384))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(383404736))))[name = string("module_layers_27_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_27_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_27_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(383470336)))];
            tensor<fp16, [1, 188, 1024]> linear_250_cast_fp16 = linear(bias = module_layers_27_self_attn_linear_out_bias_to_fp16, weight = module_layers_27_self_attn_linear_out_weight_to_fp16_quantized, x = input_1439_cast_fp16)[name = string("linear_250_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1443_cast_fp16 = add(x = input_1435_cast_fp16, y = linear_250_cast_fp16)[name = string("input_1443_cast_fp16")];
            tensor<int32, [1]> x_611_axes_0 = const()[name = string("x_611_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_27_norm_conv_weight_to_fp16 = const()[name = string("module_layers_27_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(383472448)))];
            tensor<fp16, [1024]> module_layers_27_norm_conv_bias_to_fp16 = const()[name = string("module_layers_27_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(383474560)))];
            tensor<fp16, [1, 188, 1024]> x_611_cast_fp16 = layer_norm(axes = x_611_axes_0, beta = module_layers_27_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_27_norm_conv_weight_to_fp16, x = input_1443_cast_fp16)[name = string("x_611_cast_fp16")];
            tensor<int32, [3]> input_1445_perm_0 = const()[name = string("input_1445_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1447_pad_type_0 = const()[name = string("input_1447_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1447_strides_0 = const()[name = string("input_1447_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1447_pad_0 = const()[name = string("input_1447_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1447_dilations_0 = const()[name = string("input_1447_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1447_groups_0 = const()[name = string("input_1447_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_27_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(383476672))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384525312))))[name = string("module_layers_27_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_27_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_27_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384656448)))];
            tensor<fp16, [1, 1024, 188]> input_1445_cast_fp16 = transpose(perm = input_1445_perm_0, x = x_611_cast_fp16)[name = string("transpose_222")];
            tensor<fp16, [1, 2048, 188]> input_1447_cast_fp16 = conv(bias = module_layers_27_conv_pointwise_conv1_bias_to_fp16, dilations = input_1447_dilations_0, groups = input_1447_groups_0, pad = input_1447_pad_0, pad_type = input_1447_pad_type_0, strides = input_1447_strides_0, weight = module_layers_27_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1445_cast_fp16)[name = string("input_1447_cast_fp16")];
            int32 x_613_split_num_splits_0 = const()[name = string("x_613_split_num_splits_0"), val = int32(2)];
            int32 x_613_split_axis_0 = const()[name = string("x_613_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> x_613_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_613_split_cast_fp16_1 = split(axis = x_613_split_axis_0, num_splits = x_613_split_num_splits_0, x = input_1447_cast_fp16)[name = string("x_613_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_613_split_1_sigmoid_cast_fp16 = sigmoid(x = x_613_split_cast_fp16_1)[name = string("x_613_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_613_cast_fp16 = mul(x = x_613_split_cast_fp16_0, y = x_613_split_1_sigmoid_cast_fp16)[name = string("x_613_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1449_cast_fp16 = select(a = var_11_to_fp16, b = x_613_cast_fp16, cond = var_483)[name = string("input_1449_cast_fp16")];
            tensor<int32, [6]> input_1451_pad_0 = const()[name = string("input_1451_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1451_mode_0 = const()[name = string("input_1451_mode_0"), val = string("constant")];
            fp16 const_343_to_fp16 = const()[name = string("const_343_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1451_cast_fp16 = pad(constant_val = const_343_to_fp16, mode = input_1451_mode_0, pad = input_1451_pad_0, x = input_1449_cast_fp16)[name = string("input_1451_cast_fp16")];
            string input_1453_pad_type_0 = const()[name = string("input_1453_pad_type_0"), val = string("valid")];
            int32 input_1453_groups_0 = const()[name = string("input_1453_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1453_strides_0 = const()[name = string("input_1453_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1453_pad_0 = const()[name = string("input_1453_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1453_dilations_0 = const()[name = string("input_1453_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_438_to_fp16 = const()[name = string("const_438_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384660608)))];
            tensor<fp16, [1024]> const_439_to_fp16 = const()[name = string("const_439_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384679104)))];
            tensor<fp16, [1, 1024, 188]> input_1455_cast_fp16 = conv(bias = const_439_to_fp16, dilations = input_1453_dilations_0, groups = input_1453_groups_0, pad = input_1453_pad_0, pad_type = input_1453_pad_type_0, strides = input_1453_strides_0, weight = const_438_to_fp16, x = input_1451_cast_fp16)[name = string("input_1455_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1457_cast_fp16 = silu(x = input_1455_cast_fp16)[name = string("input_1457_cast_fp16")];
            string x_615_pad_type_0 = const()[name = string("x_615_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_615_strides_0 = const()[name = string("x_615_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_615_pad_0 = const()[name = string("x_615_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_615_dilations_0 = const()[name = string("x_615_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_615_groups_0 = const()[name = string("x_615_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_27_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(384681216))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(385205568))))[name = string("module_layers_27_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_27_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_27_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(385271168)))];
            tensor<fp16, [1, 1024, 188]> x_615_cast_fp16 = conv(bias = module_layers_27_conv_pointwise_conv2_bias_to_fp16, dilations = x_615_dilations_0, groups = x_615_groups_0, pad = x_615_pad_0, pad_type = x_615_pad_type_0, strides = x_615_strides_0, weight = module_layers_27_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1457_cast_fp16)[name = string("x_615_cast_fp16")];
            tensor<int32, [3]> input_1459_perm_0 = const()[name = string("input_1459_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1459_cast_fp16 = transpose(perm = input_1459_perm_0, x = x_615_cast_fp16)[name = string("transpose_221")];
            tensor<fp16, [1, 188, 1024]> input_1461_cast_fp16 = add(x = input_1443_cast_fp16, y = input_1459_cast_fp16)[name = string("input_1461_cast_fp16")];
            tensor<int32, [1]> input_1463_axes_0 = const()[name = string("input_1463_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_27_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_27_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(385273280)))];
            tensor<fp16, [1024]> module_layers_27_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_27_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(385275392)))];
            tensor<fp16, [1, 188, 1024]> input_1463_cast_fp16 = layer_norm(axes = input_1463_axes_0, beta = module_layers_27_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_27_norm_feed_forward2_weight_to_fp16, x = input_1461_cast_fp16)[name = string("input_1463_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_27_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(385277504))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387374720))))[name = string("module_layers_27_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_27_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_27_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387636928)))];
            tensor<fp16, [1, 188, 4096]> linear_251_cast_fp16 = linear(bias = module_layers_27_feed_forward2_linear1_bias_to_fp16, weight = module_layers_27_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1463_cast_fp16)[name = string("linear_251_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1467_cast_fp16 = silu(x = linear_251_cast_fp16)[name = string("input_1467_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_27_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387645184))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389742400))))[name = string("module_layers_27_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_27_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_27_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390004608)))];
            tensor<fp16, [1, 188, 1024]> linear_252_cast_fp16 = linear(bias = module_layers_27_feed_forward2_linear2_bias_to_fp16, weight = module_layers_27_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1467_cast_fp16)[name = string("linear_252_cast_fp16")];
            fp16 var_5223_to_fp16 = const()[name = string("op_5223_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5224_cast_fp16 = mul(x = linear_252_cast_fp16, y = var_5223_to_fp16)[name = string("op_5224_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1473_cast_fp16 = add(x = input_1461_cast_fp16, y = var_5224_cast_fp16)[name = string("input_1473_cast_fp16")];
            tensor<int32, [1]> input_1475_axes_0 = const()[name = string("input_1475_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_27_norm_out_weight_to_fp16 = const()[name = string("module_layers_27_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390006720)))];
            tensor<fp16, [1024]> module_layers_27_norm_out_bias_to_fp16 = const()[name = string("module_layers_27_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390008832)))];
            tensor<fp16, [1, 188, 1024]> input_1475_cast_fp16 = layer_norm(axes = input_1475_axes_0, beta = module_layers_27_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_27_norm_out_weight_to_fp16, x = input_1473_cast_fp16)[name = string("input_1475_cast_fp16")];
            tensor<int32, [1]> input_1477_axes_0 = const()[name = string("input_1477_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_28_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_28_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390010944)))];
            tensor<fp16, [1024]> module_layers_28_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_28_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390013056)))];
            tensor<fp16, [1, 188, 1024]> input_1477_cast_fp16 = layer_norm(axes = input_1477_axes_0, beta = module_layers_28_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_28_norm_feed_forward1_weight_to_fp16, x = input_1475_cast_fp16)[name = string("input_1477_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_28_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(390015168))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392112384))))[name = string("module_layers_28_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_28_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_28_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392374592)))];
            tensor<fp16, [1, 188, 4096]> linear_253_cast_fp16 = linear(bias = module_layers_28_feed_forward1_linear1_bias_to_fp16, weight = module_layers_28_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1477_cast_fp16)[name = string("linear_253_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1481_cast_fp16 = silu(x = linear_253_cast_fp16)[name = string("input_1481_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_28_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(392382848))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394480064))))[name = string("module_layers_28_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_28_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_28_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394742272)))];
            tensor<fp16, [1, 188, 1024]> linear_254_cast_fp16 = linear(bias = module_layers_28_feed_forward1_linear2_bias_to_fp16, weight = module_layers_28_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1481_cast_fp16)[name = string("linear_254_cast_fp16")];
            fp16 var_5254_to_fp16 = const()[name = string("op_5254_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5255_cast_fp16 = mul(x = linear_254_cast_fp16, y = var_5254_to_fp16)[name = string("op_5255_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1487_cast_fp16 = add(x = input_1475_cast_fp16, y = var_5255_cast_fp16)[name = string("input_1487_cast_fp16")];
            tensor<int32, [1]> query_57_axes_0 = const()[name = string("query_57_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_28_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_28_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394744384)))];
            tensor<fp16, [1024]> module_layers_28_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_28_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394746496)))];
            tensor<fp16, [1, 188, 1024]> query_57_cast_fp16 = layer_norm(axes = query_57_axes_0, beta = module_layers_28_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_28_norm_self_att_weight_to_fp16, x = input_1487_cast_fp16)[name = string("query_57_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_28_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394748608))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395272960))))[name = string("module_layers_28_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_28_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_28_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395338560)))];
            tensor<fp16, [1, 188, 1024]> linear_255_cast_fp16 = linear(bias = module_layers_28_self_attn_linear_q_bias_to_fp16, weight = module_layers_28_self_attn_linear_q_weight_to_fp16_quantized, x = query_57_cast_fp16)[name = string("linear_255_cast_fp16")];
            tensor<int32, [4]> var_5272 = const()[name = string("op_5272"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_169_cast_fp16 = reshape(shape = var_5272, x = linear_255_cast_fp16)[name = string("q_169_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_28_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395340672))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395865024))))[name = string("module_layers_28_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_28_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_28_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395930624)))];
            tensor<fp16, [1, 188, 1024]> linear_256_cast_fp16 = linear(bias = module_layers_28_self_attn_linear_k_bias_to_fp16, weight = module_layers_28_self_attn_linear_k_weight_to_fp16_quantized, x = query_57_cast_fp16)[name = string("linear_256_cast_fp16")];
            tensor<int32, [4]> var_5277 = const()[name = string("op_5277"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_113_cast_fp16 = reshape(shape = var_5277, x = linear_256_cast_fp16)[name = string("k_113_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_28_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(395932736))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396457088))))[name = string("module_layers_28_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_28_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_28_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396522688)))];
            tensor<fp16, [1, 188, 1024]> linear_257_cast_fp16 = linear(bias = module_layers_28_self_attn_linear_v_bias_to_fp16, weight = module_layers_28_self_attn_linear_v_weight_to_fp16_quantized, x = query_57_cast_fp16)[name = string("linear_257_cast_fp16")];
            tensor<int32, [4]> var_5282 = const()[name = string("op_5282"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_57_cast_fp16 = reshape(shape = var_5282, x = linear_257_cast_fp16)[name = string("v_57_cast_fp16")];
            tensor<int32, [4]> value_59_perm_0 = const()[name = string("value_59_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_28_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396524800))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396525376))))[name = string("module_layers_28_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5294_cast_fp16 = add(x = q_169_cast_fp16, y = module_layers_28_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_5294_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_28_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396525504))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396526080))))[name = string("module_layers_28_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5296_cast_fp16 = add(x = q_169_cast_fp16, y = module_layers_28_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_5296_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_57_perm_0 = const()[name = string("q_with_bias_v_57_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_623_transpose_x_0 = const()[name = string("x_623_transpose_x_0"), val = bool(false)];
            bool x_623_transpose_y_0 = const()[name = string("x_623_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_5298_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396526208))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396718272))))[name = string("op_5298_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_57_cast_fp16 = transpose(perm = q_with_bias_v_57_perm_0, x = var_5296_cast_fp16)[name = string("transpose_220")];
            tensor<fp16, [1, 8, 188, 375]> x_623_cast_fp16 = matmul(transpose_x = x_623_transpose_x_0, transpose_y = x_623_transpose_y_0, x = q_with_bias_v_57_cast_fp16, y = op_5298_to_fp16_quantized)[name = string("x_623_cast_fp16")];
            tensor<int32, [8]> x_625_pad_0 = const()[name = string("x_625_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_625_mode_0 = const()[name = string("x_625_mode_0"), val = string("constant")];
            fp16 const_350_to_fp16 = const()[name = string("const_350_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_625_cast_fp16 = pad(constant_val = const_350_to_fp16, mode = x_625_mode_0, pad = x_625_pad_0, x = x_623_cast_fp16)[name = string("x_625_cast_fp16")];
            tensor<int32, [4]> var_5306 = const()[name = string("op_5306"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_627_cast_fp16 = reshape(shape = var_5306, x = x_625_cast_fp16)[name = string("x_627_cast_fp16")];
            tensor<int32, [4]> var_5310_begin_0 = const()[name = string("op_5310_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5310_end_0 = const()[name = string("op_5310_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5310_end_mask_0 = const()[name = string("op_5310_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5310_cast_fp16 = slice_by_index(begin = var_5310_begin_0, end = var_5310_end_0, end_mask = var_5310_end_mask_0, x = x_627_cast_fp16)[name = string("op_5310_cast_fp16")];
            tensor<int32, [4]> var_5311 = const()[name = string("op_5311"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_113_cast_fp16 = reshape(shape = var_5311, x = var_5310_cast_fp16)[name = string("matrix_bd_113_cast_fp16")];
            bool matrix_ac_57_transpose_x_0 = const()[name = string("matrix_ac_57_transpose_x_0"), val = bool(false)];
            bool matrix_ac_57_transpose_y_0 = const()[name = string("matrix_ac_57_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_184_perm_0 = const()[name = string("transpose_184_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_185_perm_0 = const()[name = string("transpose_185_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_185 = transpose(perm = transpose_185_perm_0, x = k_113_cast_fp16)[name = string("transpose_218")];
            tensor<fp16, [1, 8, 188, 128]> transpose_184 = transpose(perm = transpose_184_perm_0, x = var_5294_cast_fp16)[name = string("transpose_219")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_57_cast_fp16 = matmul(transpose_x = matrix_ac_57_transpose_x_0, transpose_y = matrix_ac_57_transpose_y_0, x = transpose_184, y = transpose_185)[name = string("matrix_ac_57_cast_fp16")];
            tensor<int32, [4]> matrix_bd_115_begin_0 = const()[name = string("matrix_bd_115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_115_end_0 = const()[name = string("matrix_bd_115_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_115_end_mask_0 = const()[name = string("matrix_bd_115_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_115_cast_fp16 = slice_by_index(begin = matrix_bd_115_begin_0, end = matrix_bd_115_end_0, end_mask = matrix_bd_115_end_mask_0, x = matrix_bd_113_cast_fp16)[name = string("matrix_bd_115_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5320_cast_fp16 = add(x = matrix_ac_57_cast_fp16, y = matrix_bd_115_cast_fp16)[name = string("op_5320_cast_fp16")];
            fp16 _inversed_scores_113_y_0_to_fp16 = const()[name = string("_inversed_scores_113_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_113_cast_fp16 = mul(x = var_5320_cast_fp16, y = _inversed_scores_113_y_0_to_fp16)[name = string("_inversed_scores_113_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_115_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_113_cast_fp16, cond = mask_11)[name = string("scores_115_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5326_cast_fp16 = softmax(axis = var_23, x = scores_115_cast_fp16)[name = string("op_5326_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1489_cast_fp16 = select(a = var_11_to_fp16, b = var_5326_cast_fp16, cond = mask_11)[name = string("input_1489_cast_fp16")];
            bool x_629_transpose_x_0 = const()[name = string("x_629_transpose_x_0"), val = bool(false)];
            bool x_629_transpose_y_0 = const()[name = string("x_629_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_59_cast_fp16 = transpose(perm = value_59_perm_0, x = v_57_cast_fp16)[name = string("transpose_217")];
            tensor<fp16, [1, 8, 188, 128]> x_629_cast_fp16 = matmul(transpose_x = x_629_transpose_x_0, transpose_y = x_629_transpose_y_0, x = input_1489_cast_fp16, y = value_59_cast_fp16)[name = string("x_629_cast_fp16")];
            tensor<int32, [4]> var_5330_perm_0 = const()[name = string("op_5330_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5331 = const()[name = string("op_5331"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_5330_cast_fp16 = transpose(perm = var_5330_perm_0, x = x_629_cast_fp16)[name = string("transpose_216")];
            tensor<fp16, [1, 188, 1024]> input_1491_cast_fp16 = reshape(shape = var_5331, x = var_5330_cast_fp16)[name = string("input_1491_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_28_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396721344))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(397245696))))[name = string("module_layers_28_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_28_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_28_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(397311296)))];
            tensor<fp16, [1, 188, 1024]> linear_259_cast_fp16 = linear(bias = module_layers_28_self_attn_linear_out_bias_to_fp16, weight = module_layers_28_self_attn_linear_out_weight_to_fp16_quantized, x = input_1491_cast_fp16)[name = string("linear_259_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1495_cast_fp16 = add(x = input_1487_cast_fp16, y = linear_259_cast_fp16)[name = string("input_1495_cast_fp16")];
            tensor<int32, [1]> x_633_axes_0 = const()[name = string("x_633_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_28_norm_conv_weight_to_fp16 = const()[name = string("module_layers_28_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(397313408)))];
            tensor<fp16, [1024]> module_layers_28_norm_conv_bias_to_fp16 = const()[name = string("module_layers_28_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(397315520)))];
            tensor<fp16, [1, 188, 1024]> x_633_cast_fp16 = layer_norm(axes = x_633_axes_0, beta = module_layers_28_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_28_norm_conv_weight_to_fp16, x = input_1495_cast_fp16)[name = string("x_633_cast_fp16")];
            tensor<int32, [3]> input_1497_perm_0 = const()[name = string("input_1497_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1499_pad_type_0 = const()[name = string("input_1499_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1499_strides_0 = const()[name = string("input_1499_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1499_pad_0 = const()[name = string("input_1499_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1499_dilations_0 = const()[name = string("input_1499_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1499_groups_0 = const()[name = string("input_1499_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_28_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(397317632))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(398366272))))[name = string("module_layers_28_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_28_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_28_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(398497408)))];
            tensor<fp16, [1, 1024, 188]> input_1497_cast_fp16 = transpose(perm = input_1497_perm_0, x = x_633_cast_fp16)[name = string("transpose_215")];
            tensor<fp16, [1, 2048, 188]> input_1499_cast_fp16 = conv(bias = module_layers_28_conv_pointwise_conv1_bias_to_fp16, dilations = input_1499_dilations_0, groups = input_1499_groups_0, pad = input_1499_pad_0, pad_type = input_1499_pad_type_0, strides = input_1499_strides_0, weight = module_layers_28_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1497_cast_fp16)[name = string("input_1499_cast_fp16")];
            int32 x_635_split_num_splits_0 = const()[name = string("x_635_split_num_splits_0"), val = int32(2)];
            int32 x_635_split_axis_0 = const()[name = string("x_635_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> x_635_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_635_split_cast_fp16_1 = split(axis = x_635_split_axis_0, num_splits = x_635_split_num_splits_0, x = input_1499_cast_fp16)[name = string("x_635_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_635_split_1_sigmoid_cast_fp16 = sigmoid(x = x_635_split_cast_fp16_1)[name = string("x_635_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_635_cast_fp16 = mul(x = x_635_split_cast_fp16_0, y = x_635_split_1_sigmoid_cast_fp16)[name = string("x_635_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1501_cast_fp16 = select(a = var_11_to_fp16, b = x_635_cast_fp16, cond = var_483)[name = string("input_1501_cast_fp16")];
            tensor<int32, [6]> input_1503_pad_0 = const()[name = string("input_1503_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1503_mode_0 = const()[name = string("input_1503_mode_0"), val = string("constant")];
            fp16 const_353_to_fp16 = const()[name = string("const_353_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1503_cast_fp16 = pad(constant_val = const_353_to_fp16, mode = input_1503_mode_0, pad = input_1503_pad_0, x = input_1501_cast_fp16)[name = string("input_1503_cast_fp16")];
            string input_1505_pad_type_0 = const()[name = string("input_1505_pad_type_0"), val = string("valid")];
            int32 input_1505_groups_0 = const()[name = string("input_1505_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1505_strides_0 = const()[name = string("input_1505_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1505_pad_0 = const()[name = string("input_1505_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1505_dilations_0 = const()[name = string("input_1505_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_440_to_fp16 = const()[name = string("const_440_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(398501568)))];
            tensor<fp16, [1024]> const_441_to_fp16 = const()[name = string("const_441_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(398520064)))];
            tensor<fp16, [1, 1024, 188]> input_1507_cast_fp16 = conv(bias = const_441_to_fp16, dilations = input_1505_dilations_0, groups = input_1505_groups_0, pad = input_1505_pad_0, pad_type = input_1505_pad_type_0, strides = input_1505_strides_0, weight = const_440_to_fp16, x = input_1503_cast_fp16)[name = string("input_1507_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1509_cast_fp16 = silu(x = input_1507_cast_fp16)[name = string("input_1509_cast_fp16")];
            string x_637_pad_type_0 = const()[name = string("x_637_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_637_strides_0 = const()[name = string("x_637_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_637_pad_0 = const()[name = string("x_637_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_637_dilations_0 = const()[name = string("x_637_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_637_groups_0 = const()[name = string("x_637_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_28_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(398522176))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399046528))))[name = string("module_layers_28_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_28_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_28_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399112128)))];
            tensor<fp16, [1, 1024, 188]> x_637_cast_fp16 = conv(bias = module_layers_28_conv_pointwise_conv2_bias_to_fp16, dilations = x_637_dilations_0, groups = x_637_groups_0, pad = x_637_pad_0, pad_type = x_637_pad_type_0, strides = x_637_strides_0, weight = module_layers_28_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1509_cast_fp16)[name = string("x_637_cast_fp16")];
            tensor<int32, [3]> input_1511_perm_0 = const()[name = string("input_1511_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1511_cast_fp16 = transpose(perm = input_1511_perm_0, x = x_637_cast_fp16)[name = string("transpose_214")];
            tensor<fp16, [1, 188, 1024]> input_1513_cast_fp16 = add(x = input_1495_cast_fp16, y = input_1511_cast_fp16)[name = string("input_1513_cast_fp16")];
            tensor<int32, [1]> input_1515_axes_0 = const()[name = string("input_1515_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_28_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_28_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399114240)))];
            tensor<fp16, [1024]> module_layers_28_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_28_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399116352)))];
            tensor<fp16, [1, 188, 1024]> input_1515_cast_fp16 = layer_norm(axes = input_1515_axes_0, beta = module_layers_28_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_28_norm_feed_forward2_weight_to_fp16, x = input_1513_cast_fp16)[name = string("input_1515_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_28_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(399118464))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(401215680))))[name = string("module_layers_28_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_28_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_28_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(401477888)))];
            tensor<fp16, [1, 188, 4096]> linear_260_cast_fp16 = linear(bias = module_layers_28_feed_forward2_linear1_bias_to_fp16, weight = module_layers_28_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1515_cast_fp16)[name = string("linear_260_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1519_cast_fp16 = silu(x = linear_260_cast_fp16)[name = string("input_1519_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_28_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(401486144))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403583360))))[name = string("module_layers_28_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_28_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_28_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403845568)))];
            tensor<fp16, [1, 188, 1024]> linear_261_cast_fp16 = linear(bias = module_layers_28_feed_forward2_linear2_bias_to_fp16, weight = module_layers_28_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1519_cast_fp16)[name = string("linear_261_cast_fp16")];
            fp16 var_5397_to_fp16 = const()[name = string("op_5397_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5398_cast_fp16 = mul(x = linear_261_cast_fp16, y = var_5397_to_fp16)[name = string("op_5398_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1525_cast_fp16 = add(x = input_1513_cast_fp16, y = var_5398_cast_fp16)[name = string("input_1525_cast_fp16")];
            tensor<int32, [1]> input_1527_axes_0 = const()[name = string("input_1527_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_28_norm_out_weight_to_fp16 = const()[name = string("module_layers_28_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403847680)))];
            tensor<fp16, [1024]> module_layers_28_norm_out_bias_to_fp16 = const()[name = string("module_layers_28_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403849792)))];
            tensor<fp16, [1, 188, 1024]> input_1527_cast_fp16 = layer_norm(axes = input_1527_axes_0, beta = module_layers_28_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_28_norm_out_weight_to_fp16, x = input_1525_cast_fp16)[name = string("input_1527_cast_fp16")];
            tensor<int32, [1]> input_1529_axes_0 = const()[name = string("input_1529_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_29_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_29_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403851904)))];
            tensor<fp16, [1024]> module_layers_29_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_29_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403854016)))];
            tensor<fp16, [1, 188, 1024]> input_1529_cast_fp16 = layer_norm(axes = input_1529_axes_0, beta = module_layers_29_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_29_norm_feed_forward1_weight_to_fp16, x = input_1527_cast_fp16)[name = string("input_1529_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_29_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(403856128))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(405953344))))[name = string("module_layers_29_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_29_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_29_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406215552)))];
            tensor<fp16, [1, 188, 4096]> linear_262_cast_fp16 = linear(bias = module_layers_29_feed_forward1_linear1_bias_to_fp16, weight = module_layers_29_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1529_cast_fp16)[name = string("linear_262_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1533_cast_fp16 = silu(x = linear_262_cast_fp16)[name = string("input_1533_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_29_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(406223808))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408321024))))[name = string("module_layers_29_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_29_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_29_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408583232)))];
            tensor<fp16, [1, 188, 1024]> linear_263_cast_fp16 = linear(bias = module_layers_29_feed_forward1_linear2_bias_to_fp16, weight = module_layers_29_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1533_cast_fp16)[name = string("linear_263_cast_fp16")];
            fp16 var_5428_to_fp16 = const()[name = string("op_5428_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5429_cast_fp16 = mul(x = linear_263_cast_fp16, y = var_5428_to_fp16)[name = string("op_5429_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1539_cast_fp16 = add(x = input_1527_cast_fp16, y = var_5429_cast_fp16)[name = string("input_1539_cast_fp16")];
            tensor<int32, [1]> query_59_axes_0 = const()[name = string("query_59_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_29_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_29_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408585344)))];
            tensor<fp16, [1024]> module_layers_29_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_29_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408587456)))];
            tensor<fp16, [1, 188, 1024]> query_59_cast_fp16 = layer_norm(axes = query_59_axes_0, beta = module_layers_29_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_29_norm_self_att_weight_to_fp16, x = input_1539_cast_fp16)[name = string("query_59_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_29_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(408589568))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409113920))))[name = string("module_layers_29_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_29_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_29_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409179520)))];
            tensor<fp16, [1, 188, 1024]> linear_264_cast_fp16 = linear(bias = module_layers_29_self_attn_linear_q_bias_to_fp16, weight = module_layers_29_self_attn_linear_q_weight_to_fp16_quantized, x = query_59_cast_fp16)[name = string("linear_264_cast_fp16")];
            tensor<int32, [4]> var_5446 = const()[name = string("op_5446"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_175_cast_fp16 = reshape(shape = var_5446, x = linear_264_cast_fp16)[name = string("q_175_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_29_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409181632))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409705984))))[name = string("module_layers_29_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_29_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_29_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409771584)))];
            tensor<fp16, [1, 188, 1024]> linear_265_cast_fp16 = linear(bias = module_layers_29_self_attn_linear_k_bias_to_fp16, weight = module_layers_29_self_attn_linear_k_weight_to_fp16_quantized, x = query_59_cast_fp16)[name = string("linear_265_cast_fp16")];
            tensor<int32, [4]> var_5451 = const()[name = string("op_5451"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_117_cast_fp16 = reshape(shape = var_5451, x = linear_265_cast_fp16)[name = string("k_117_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_29_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409773696))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410298048))))[name = string("module_layers_29_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_29_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_29_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410363648)))];
            tensor<fp16, [1, 188, 1024]> linear_266_cast_fp16 = linear(bias = module_layers_29_self_attn_linear_v_bias_to_fp16, weight = module_layers_29_self_attn_linear_v_weight_to_fp16_quantized, x = query_59_cast_fp16)[name = string("linear_266_cast_fp16")];
            tensor<int32, [4]> var_5456 = const()[name = string("op_5456"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_59_cast_fp16 = reshape(shape = var_5456, x = linear_266_cast_fp16)[name = string("v_59_cast_fp16")];
            tensor<int32, [4]> value_61_perm_0 = const()[name = string("value_61_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_29_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410365760))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410366336))))[name = string("module_layers_29_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5468_cast_fp16 = add(x = q_175_cast_fp16, y = module_layers_29_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_5468_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_29_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410366464))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410367040))))[name = string("module_layers_29_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5470_cast_fp16 = add(x = q_175_cast_fp16, y = module_layers_29_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_5470_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_59_perm_0 = const()[name = string("q_with_bias_v_59_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_645_transpose_x_0 = const()[name = string("x_645_transpose_x_0"), val = bool(false)];
            bool x_645_transpose_y_0 = const()[name = string("x_645_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_5472_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410367168))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410559232))))[name = string("op_5472_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_59_cast_fp16 = transpose(perm = q_with_bias_v_59_perm_0, x = var_5470_cast_fp16)[name = string("transpose_213")];
            tensor<fp16, [1, 8, 188, 375]> x_645_cast_fp16 = matmul(transpose_x = x_645_transpose_x_0, transpose_y = x_645_transpose_y_0, x = q_with_bias_v_59_cast_fp16, y = op_5472_to_fp16_quantized)[name = string("x_645_cast_fp16")];
            tensor<int32, [8]> x_647_pad_0 = const()[name = string("x_647_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_647_mode_0 = const()[name = string("x_647_mode_0"), val = string("constant")];
            fp16 const_360_to_fp16 = const()[name = string("const_360_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_647_cast_fp16 = pad(constant_val = const_360_to_fp16, mode = x_647_mode_0, pad = x_647_pad_0, x = x_645_cast_fp16)[name = string("x_647_cast_fp16")];
            tensor<int32, [4]> var_5480 = const()[name = string("op_5480"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_649_cast_fp16 = reshape(shape = var_5480, x = x_647_cast_fp16)[name = string("x_649_cast_fp16")];
            tensor<int32, [4]> var_5484_begin_0 = const()[name = string("op_5484_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5484_end_0 = const()[name = string("op_5484_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5484_end_mask_0 = const()[name = string("op_5484_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5484_cast_fp16 = slice_by_index(begin = var_5484_begin_0, end = var_5484_end_0, end_mask = var_5484_end_mask_0, x = x_649_cast_fp16)[name = string("op_5484_cast_fp16")];
            tensor<int32, [4]> var_5485 = const()[name = string("op_5485"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_117_cast_fp16 = reshape(shape = var_5485, x = var_5484_cast_fp16)[name = string("matrix_bd_117_cast_fp16")];
            bool matrix_ac_59_transpose_x_0 = const()[name = string("matrix_ac_59_transpose_x_0"), val = bool(false)];
            bool matrix_ac_59_transpose_y_0 = const()[name = string("matrix_ac_59_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_186_perm_0 = const()[name = string("transpose_186_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_187_perm_0 = const()[name = string("transpose_187_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_187 = transpose(perm = transpose_187_perm_0, x = k_117_cast_fp16)[name = string("transpose_211")];
            tensor<fp16, [1, 8, 188, 128]> transpose_186 = transpose(perm = transpose_186_perm_0, x = var_5468_cast_fp16)[name = string("transpose_212")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_59_cast_fp16 = matmul(transpose_x = matrix_ac_59_transpose_x_0, transpose_y = matrix_ac_59_transpose_y_0, x = transpose_186, y = transpose_187)[name = string("matrix_ac_59_cast_fp16")];
            tensor<int32, [4]> matrix_bd_119_begin_0 = const()[name = string("matrix_bd_119_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_119_end_0 = const()[name = string("matrix_bd_119_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_119_end_mask_0 = const()[name = string("matrix_bd_119_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_119_cast_fp16 = slice_by_index(begin = matrix_bd_119_begin_0, end = matrix_bd_119_end_0, end_mask = matrix_bd_119_end_mask_0, x = matrix_bd_117_cast_fp16)[name = string("matrix_bd_119_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5494_cast_fp16 = add(x = matrix_ac_59_cast_fp16, y = matrix_bd_119_cast_fp16)[name = string("op_5494_cast_fp16")];
            fp16 _inversed_scores_117_y_0_to_fp16 = const()[name = string("_inversed_scores_117_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_117_cast_fp16 = mul(x = var_5494_cast_fp16, y = _inversed_scores_117_y_0_to_fp16)[name = string("_inversed_scores_117_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_119_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_117_cast_fp16, cond = mask_11)[name = string("scores_119_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5500_cast_fp16 = softmax(axis = var_23, x = scores_119_cast_fp16)[name = string("op_5500_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1541_cast_fp16 = select(a = var_11_to_fp16, b = var_5500_cast_fp16, cond = mask_11)[name = string("input_1541_cast_fp16")];
            bool x_651_transpose_x_0 = const()[name = string("x_651_transpose_x_0"), val = bool(false)];
            bool x_651_transpose_y_0 = const()[name = string("x_651_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_61_cast_fp16 = transpose(perm = value_61_perm_0, x = v_59_cast_fp16)[name = string("transpose_210")];
            tensor<fp16, [1, 8, 188, 128]> x_651_cast_fp16 = matmul(transpose_x = x_651_transpose_x_0, transpose_y = x_651_transpose_y_0, x = input_1541_cast_fp16, y = value_61_cast_fp16)[name = string("x_651_cast_fp16")];
            tensor<int32, [4]> var_5504_perm_0 = const()[name = string("op_5504_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5505 = const()[name = string("op_5505"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_5504_cast_fp16 = transpose(perm = var_5504_perm_0, x = x_651_cast_fp16)[name = string("transpose_209")];
            tensor<fp16, [1, 188, 1024]> input_1543_cast_fp16 = reshape(shape = var_5505, x = var_5504_cast_fp16)[name = string("input_1543_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_29_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(410562304))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411086656))))[name = string("module_layers_29_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_29_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_29_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411152256)))];
            tensor<fp16, [1, 188, 1024]> linear_268_cast_fp16 = linear(bias = module_layers_29_self_attn_linear_out_bias_to_fp16, weight = module_layers_29_self_attn_linear_out_weight_to_fp16_quantized, x = input_1543_cast_fp16)[name = string("linear_268_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1547_cast_fp16 = add(x = input_1539_cast_fp16, y = linear_268_cast_fp16)[name = string("input_1547_cast_fp16")];
            tensor<int32, [1]> x_655_axes_0 = const()[name = string("x_655_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_29_norm_conv_weight_to_fp16 = const()[name = string("module_layers_29_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411154368)))];
            tensor<fp16, [1024]> module_layers_29_norm_conv_bias_to_fp16 = const()[name = string("module_layers_29_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411156480)))];
            tensor<fp16, [1, 188, 1024]> x_655_cast_fp16 = layer_norm(axes = x_655_axes_0, beta = module_layers_29_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_29_norm_conv_weight_to_fp16, x = input_1547_cast_fp16)[name = string("x_655_cast_fp16")];
            tensor<int32, [3]> input_1549_perm_0 = const()[name = string("input_1549_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1551_pad_type_0 = const()[name = string("input_1551_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1551_strides_0 = const()[name = string("input_1551_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1551_pad_0 = const()[name = string("input_1551_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1551_dilations_0 = const()[name = string("input_1551_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1551_groups_0 = const()[name = string("input_1551_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_29_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411158592))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412207232))))[name = string("module_layers_29_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_29_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_29_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412338368)))];
            tensor<fp16, [1, 1024, 188]> input_1549_cast_fp16 = transpose(perm = input_1549_perm_0, x = x_655_cast_fp16)[name = string("transpose_208")];
            tensor<fp16, [1, 2048, 188]> input_1551_cast_fp16 = conv(bias = module_layers_29_conv_pointwise_conv1_bias_to_fp16, dilations = input_1551_dilations_0, groups = input_1551_groups_0, pad = input_1551_pad_0, pad_type = input_1551_pad_type_0, strides = input_1551_strides_0, weight = module_layers_29_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1549_cast_fp16)[name = string("input_1551_cast_fp16")];
            int32 x_657_split_num_splits_0 = const()[name = string("x_657_split_num_splits_0"), val = int32(2)];
            int32 x_657_split_axis_0 = const()[name = string("x_657_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> x_657_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_657_split_cast_fp16_1 = split(axis = x_657_split_axis_0, num_splits = x_657_split_num_splits_0, x = input_1551_cast_fp16)[name = string("x_657_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_657_split_1_sigmoid_cast_fp16 = sigmoid(x = x_657_split_cast_fp16_1)[name = string("x_657_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_657_cast_fp16 = mul(x = x_657_split_cast_fp16_0, y = x_657_split_1_sigmoid_cast_fp16)[name = string("x_657_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1553_cast_fp16 = select(a = var_11_to_fp16, b = x_657_cast_fp16, cond = var_483)[name = string("input_1553_cast_fp16")];
            tensor<int32, [6]> input_1555_pad_0 = const()[name = string("input_1555_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1555_mode_0 = const()[name = string("input_1555_mode_0"), val = string("constant")];
            fp16 const_363_to_fp16 = const()[name = string("const_363_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1555_cast_fp16 = pad(constant_val = const_363_to_fp16, mode = input_1555_mode_0, pad = input_1555_pad_0, x = input_1553_cast_fp16)[name = string("input_1555_cast_fp16")];
            string input_1557_pad_type_0 = const()[name = string("input_1557_pad_type_0"), val = string("valid")];
            int32 input_1557_groups_0 = const()[name = string("input_1557_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1557_strides_0 = const()[name = string("input_1557_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1557_pad_0 = const()[name = string("input_1557_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1557_dilations_0 = const()[name = string("input_1557_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_442_to_fp16 = const()[name = string("const_442_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412342528)))];
            tensor<fp16, [1024]> const_443_to_fp16 = const()[name = string("const_443_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412361024)))];
            tensor<fp16, [1, 1024, 188]> input_1559_cast_fp16 = conv(bias = const_443_to_fp16, dilations = input_1557_dilations_0, groups = input_1557_groups_0, pad = input_1557_pad_0, pad_type = input_1557_pad_type_0, strides = input_1557_strides_0, weight = const_442_to_fp16, x = input_1555_cast_fp16)[name = string("input_1559_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1561_cast_fp16 = silu(x = input_1559_cast_fp16)[name = string("input_1561_cast_fp16")];
            string x_659_pad_type_0 = const()[name = string("x_659_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_659_strides_0 = const()[name = string("x_659_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_659_pad_0 = const()[name = string("x_659_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_659_dilations_0 = const()[name = string("x_659_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_659_groups_0 = const()[name = string("x_659_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_29_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412363136))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412887488))))[name = string("module_layers_29_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_29_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_29_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412953088)))];
            tensor<fp16, [1, 1024, 188]> x_659_cast_fp16 = conv(bias = module_layers_29_conv_pointwise_conv2_bias_to_fp16, dilations = x_659_dilations_0, groups = x_659_groups_0, pad = x_659_pad_0, pad_type = x_659_pad_type_0, strides = x_659_strides_0, weight = module_layers_29_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1561_cast_fp16)[name = string("x_659_cast_fp16")];
            tensor<int32, [3]> input_1563_perm_0 = const()[name = string("input_1563_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1563_cast_fp16 = transpose(perm = input_1563_perm_0, x = x_659_cast_fp16)[name = string("transpose_207")];
            tensor<fp16, [1, 188, 1024]> input_1565_cast_fp16 = add(x = input_1547_cast_fp16, y = input_1563_cast_fp16)[name = string("input_1565_cast_fp16")];
            tensor<int32, [1]> input_1567_axes_0 = const()[name = string("input_1567_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_29_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_29_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412955200)))];
            tensor<fp16, [1024]> module_layers_29_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_29_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412957312)))];
            tensor<fp16, [1, 188, 1024]> input_1567_cast_fp16 = layer_norm(axes = input_1567_axes_0, beta = module_layers_29_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_29_norm_feed_forward2_weight_to_fp16, x = input_1565_cast_fp16)[name = string("input_1567_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_29_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412959424))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415056640))))[name = string("module_layers_29_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_29_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_29_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415318848)))];
            tensor<fp16, [1, 188, 4096]> linear_269_cast_fp16 = linear(bias = module_layers_29_feed_forward2_linear1_bias_to_fp16, weight = module_layers_29_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1567_cast_fp16)[name = string("linear_269_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1571_cast_fp16 = silu(x = linear_269_cast_fp16)[name = string("input_1571_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_29_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415327104))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417424320))))[name = string("module_layers_29_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_29_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_29_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417686528)))];
            tensor<fp16, [1, 188, 1024]> linear_270_cast_fp16 = linear(bias = module_layers_29_feed_forward2_linear2_bias_to_fp16, weight = module_layers_29_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1571_cast_fp16)[name = string("linear_270_cast_fp16")];
            fp16 var_5571_to_fp16 = const()[name = string("op_5571_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5572_cast_fp16 = mul(x = linear_270_cast_fp16, y = var_5571_to_fp16)[name = string("op_5572_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1577_cast_fp16 = add(x = input_1565_cast_fp16, y = var_5572_cast_fp16)[name = string("input_1577_cast_fp16")];
            tensor<int32, [1]> input_1579_axes_0 = const()[name = string("input_1579_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_29_norm_out_weight_to_fp16 = const()[name = string("module_layers_29_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417688640)))];
            tensor<fp16, [1024]> module_layers_29_norm_out_bias_to_fp16 = const()[name = string("module_layers_29_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417690752)))];
            tensor<fp16, [1, 188, 1024]> input_1579_cast_fp16 = layer_norm(axes = input_1579_axes_0, beta = module_layers_29_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_29_norm_out_weight_to_fp16, x = input_1577_cast_fp16)[name = string("input_1579_cast_fp16")];
            tensor<int32, [1]> input_1581_axes_0 = const()[name = string("input_1581_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_30_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_30_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417692864)))];
            tensor<fp16, [1024]> module_layers_30_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_30_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417694976)))];
            tensor<fp16, [1, 188, 1024]> input_1581_cast_fp16 = layer_norm(axes = input_1581_axes_0, beta = module_layers_30_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_30_norm_feed_forward1_weight_to_fp16, x = input_1579_cast_fp16)[name = string("input_1581_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_30_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417697088))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419794304))))[name = string("module_layers_30_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_30_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_30_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(420056512)))];
            tensor<fp16, [1, 188, 4096]> linear_271_cast_fp16 = linear(bias = module_layers_30_feed_forward1_linear1_bias_to_fp16, weight = module_layers_30_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1581_cast_fp16)[name = string("linear_271_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1585_cast_fp16 = silu(x = linear_271_cast_fp16)[name = string("input_1585_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_30_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(420064768))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422161984))))[name = string("module_layers_30_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_30_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_30_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422424192)))];
            tensor<fp16, [1, 188, 1024]> linear_272_cast_fp16 = linear(bias = module_layers_30_feed_forward1_linear2_bias_to_fp16, weight = module_layers_30_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1585_cast_fp16)[name = string("linear_272_cast_fp16")];
            fp16 var_5602_to_fp16 = const()[name = string("op_5602_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5603_cast_fp16 = mul(x = linear_272_cast_fp16, y = var_5602_to_fp16)[name = string("op_5603_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1591_cast_fp16 = add(x = input_1579_cast_fp16, y = var_5603_cast_fp16)[name = string("input_1591_cast_fp16")];
            tensor<int32, [1]> query_61_axes_0 = const()[name = string("query_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_30_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_30_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422426304)))];
            tensor<fp16, [1024]> module_layers_30_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_30_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422428416)))];
            tensor<fp16, [1, 188, 1024]> query_61_cast_fp16 = layer_norm(axes = query_61_axes_0, beta = module_layers_30_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_30_norm_self_att_weight_to_fp16, x = input_1591_cast_fp16)[name = string("query_61_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_30_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422430528))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(422954880))))[name = string("module_layers_30_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_30_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_30_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(423020480)))];
            tensor<fp16, [1, 188, 1024]> linear_273_cast_fp16 = linear(bias = module_layers_30_self_attn_linear_q_bias_to_fp16, weight = module_layers_30_self_attn_linear_q_weight_to_fp16_quantized, x = query_61_cast_fp16)[name = string("linear_273_cast_fp16")];
            tensor<int32, [4]> var_5620 = const()[name = string("op_5620"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_181_cast_fp16 = reshape(shape = var_5620, x = linear_273_cast_fp16)[name = string("q_181_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_30_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(423022592))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(423546944))))[name = string("module_layers_30_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_30_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_30_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(423612544)))];
            tensor<fp16, [1, 188, 1024]> linear_274_cast_fp16 = linear(bias = module_layers_30_self_attn_linear_k_bias_to_fp16, weight = module_layers_30_self_attn_linear_k_weight_to_fp16_quantized, x = query_61_cast_fp16)[name = string("linear_274_cast_fp16")];
            tensor<int32, [4]> var_5625 = const()[name = string("op_5625"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_121_cast_fp16 = reshape(shape = var_5625, x = linear_274_cast_fp16)[name = string("k_121_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_30_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(423614656))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424139008))))[name = string("module_layers_30_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_30_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_30_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424204608)))];
            tensor<fp16, [1, 188, 1024]> linear_275_cast_fp16 = linear(bias = module_layers_30_self_attn_linear_v_bias_to_fp16, weight = module_layers_30_self_attn_linear_v_weight_to_fp16_quantized, x = query_61_cast_fp16)[name = string("linear_275_cast_fp16")];
            tensor<int32, [4]> var_5630 = const()[name = string("op_5630"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_61_cast_fp16 = reshape(shape = var_5630, x = linear_275_cast_fp16)[name = string("v_61_cast_fp16")];
            tensor<int32, [4]> value_63_perm_0 = const()[name = string("value_63_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_30_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424206720))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424207296))))[name = string("module_layers_30_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5642_cast_fp16 = add(x = q_181_cast_fp16, y = module_layers_30_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_5642_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_30_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424207424))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424208000))))[name = string("module_layers_30_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5644_cast_fp16 = add(x = q_181_cast_fp16, y = module_layers_30_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_5644_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_61_perm_0 = const()[name = string("q_with_bias_v_61_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_667_transpose_x_0 = const()[name = string("x_667_transpose_x_0"), val = bool(false)];
            bool x_667_transpose_y_0 = const()[name = string("x_667_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_5646_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424208128))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424400192))))[name = string("op_5646_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_61_cast_fp16 = transpose(perm = q_with_bias_v_61_perm_0, x = var_5644_cast_fp16)[name = string("transpose_206")];
            tensor<fp16, [1, 8, 188, 375]> x_667_cast_fp16 = matmul(transpose_x = x_667_transpose_x_0, transpose_y = x_667_transpose_y_0, x = q_with_bias_v_61_cast_fp16, y = op_5646_to_fp16_quantized)[name = string("x_667_cast_fp16")];
            tensor<int32, [8]> x_669_pad_0 = const()[name = string("x_669_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_669_mode_0 = const()[name = string("x_669_mode_0"), val = string("constant")];
            fp16 const_370_to_fp16 = const()[name = string("const_370_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_669_cast_fp16 = pad(constant_val = const_370_to_fp16, mode = x_669_mode_0, pad = x_669_pad_0, x = x_667_cast_fp16)[name = string("x_669_cast_fp16")];
            tensor<int32, [4]> var_5654 = const()[name = string("op_5654"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_671_cast_fp16 = reshape(shape = var_5654, x = x_669_cast_fp16)[name = string("x_671_cast_fp16")];
            tensor<int32, [4]> var_5658_begin_0 = const()[name = string("op_5658_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5658_end_0 = const()[name = string("op_5658_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5658_end_mask_0 = const()[name = string("op_5658_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5658_cast_fp16 = slice_by_index(begin = var_5658_begin_0, end = var_5658_end_0, end_mask = var_5658_end_mask_0, x = x_671_cast_fp16)[name = string("op_5658_cast_fp16")];
            tensor<int32, [4]> var_5659 = const()[name = string("op_5659"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_121_cast_fp16 = reshape(shape = var_5659, x = var_5658_cast_fp16)[name = string("matrix_bd_121_cast_fp16")];
            bool matrix_ac_61_transpose_x_0 = const()[name = string("matrix_ac_61_transpose_x_0"), val = bool(false)];
            bool matrix_ac_61_transpose_y_0 = const()[name = string("matrix_ac_61_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_188_perm_0 = const()[name = string("transpose_188_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_189_perm_0 = const()[name = string("transpose_189_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_189 = transpose(perm = transpose_189_perm_0, x = k_121_cast_fp16)[name = string("transpose_204")];
            tensor<fp16, [1, 8, 188, 128]> transpose_188 = transpose(perm = transpose_188_perm_0, x = var_5642_cast_fp16)[name = string("transpose_205")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_61_cast_fp16 = matmul(transpose_x = matrix_ac_61_transpose_x_0, transpose_y = matrix_ac_61_transpose_y_0, x = transpose_188, y = transpose_189)[name = string("matrix_ac_61_cast_fp16")];
            tensor<int32, [4]> matrix_bd_123_begin_0 = const()[name = string("matrix_bd_123_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_123_end_0 = const()[name = string("matrix_bd_123_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_123_end_mask_0 = const()[name = string("matrix_bd_123_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_123_cast_fp16 = slice_by_index(begin = matrix_bd_123_begin_0, end = matrix_bd_123_end_0, end_mask = matrix_bd_123_end_mask_0, x = matrix_bd_121_cast_fp16)[name = string("matrix_bd_123_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5668_cast_fp16 = add(x = matrix_ac_61_cast_fp16, y = matrix_bd_123_cast_fp16)[name = string("op_5668_cast_fp16")];
            fp16 _inversed_scores_121_y_0_to_fp16 = const()[name = string("_inversed_scores_121_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_121_cast_fp16 = mul(x = var_5668_cast_fp16, y = _inversed_scores_121_y_0_to_fp16)[name = string("_inversed_scores_121_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_123_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_121_cast_fp16, cond = mask_11)[name = string("scores_123_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5674_cast_fp16 = softmax(axis = var_23, x = scores_123_cast_fp16)[name = string("op_5674_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1593_cast_fp16 = select(a = var_11_to_fp16, b = var_5674_cast_fp16, cond = mask_11)[name = string("input_1593_cast_fp16")];
            bool x_673_transpose_x_0 = const()[name = string("x_673_transpose_x_0"), val = bool(false)];
            bool x_673_transpose_y_0 = const()[name = string("x_673_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_63_cast_fp16 = transpose(perm = value_63_perm_0, x = v_61_cast_fp16)[name = string("transpose_203")];
            tensor<fp16, [1, 8, 188, 128]> x_673_cast_fp16 = matmul(transpose_x = x_673_transpose_x_0, transpose_y = x_673_transpose_y_0, x = input_1593_cast_fp16, y = value_63_cast_fp16)[name = string("x_673_cast_fp16")];
            tensor<int32, [4]> var_5678_perm_0 = const()[name = string("op_5678_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5679 = const()[name = string("op_5679"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_5678_cast_fp16 = transpose(perm = var_5678_perm_0, x = x_673_cast_fp16)[name = string("transpose_202")];
            tensor<fp16, [1, 188, 1024]> input_1595_cast_fp16 = reshape(shape = var_5679, x = var_5678_cast_fp16)[name = string("input_1595_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_30_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424403264))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424927616))))[name = string("module_layers_30_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_30_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_30_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424993216)))];
            tensor<fp16, [1, 188, 1024]> linear_277_cast_fp16 = linear(bias = module_layers_30_self_attn_linear_out_bias_to_fp16, weight = module_layers_30_self_attn_linear_out_weight_to_fp16_quantized, x = input_1595_cast_fp16)[name = string("linear_277_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1599_cast_fp16 = add(x = input_1591_cast_fp16, y = linear_277_cast_fp16)[name = string("input_1599_cast_fp16")];
            tensor<int32, [1]> x_677_axes_0 = const()[name = string("x_677_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_30_norm_conv_weight_to_fp16 = const()[name = string("module_layers_30_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424995328)))];
            tensor<fp16, [1024]> module_layers_30_norm_conv_bias_to_fp16 = const()[name = string("module_layers_30_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424997440)))];
            tensor<fp16, [1, 188, 1024]> x_677_cast_fp16 = layer_norm(axes = x_677_axes_0, beta = module_layers_30_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_30_norm_conv_weight_to_fp16, x = input_1599_cast_fp16)[name = string("x_677_cast_fp16")];
            tensor<int32, [3]> input_1601_perm_0 = const()[name = string("input_1601_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1603_pad_type_0 = const()[name = string("input_1603_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1603_strides_0 = const()[name = string("input_1603_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1603_pad_0 = const()[name = string("input_1603_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1603_dilations_0 = const()[name = string("input_1603_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1603_groups_0 = const()[name = string("input_1603_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_30_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(424999552))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426048192))))[name = string("module_layers_30_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_30_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_30_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426179328)))];
            tensor<fp16, [1, 1024, 188]> input_1601_cast_fp16 = transpose(perm = input_1601_perm_0, x = x_677_cast_fp16)[name = string("transpose_201")];
            tensor<fp16, [1, 2048, 188]> input_1603_cast_fp16 = conv(bias = module_layers_30_conv_pointwise_conv1_bias_to_fp16, dilations = input_1603_dilations_0, groups = input_1603_groups_0, pad = input_1603_pad_0, pad_type = input_1603_pad_type_0, strides = input_1603_strides_0, weight = module_layers_30_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1601_cast_fp16)[name = string("input_1603_cast_fp16")];
            int32 x_679_split_num_splits_0 = const()[name = string("x_679_split_num_splits_0"), val = int32(2)];
            int32 x_679_split_axis_0 = const()[name = string("x_679_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> x_679_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_679_split_cast_fp16_1 = split(axis = x_679_split_axis_0, num_splits = x_679_split_num_splits_0, x = input_1603_cast_fp16)[name = string("x_679_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_679_split_1_sigmoid_cast_fp16 = sigmoid(x = x_679_split_cast_fp16_1)[name = string("x_679_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_679_cast_fp16 = mul(x = x_679_split_cast_fp16_0, y = x_679_split_1_sigmoid_cast_fp16)[name = string("x_679_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1605_cast_fp16 = select(a = var_11_to_fp16, b = x_679_cast_fp16, cond = var_483)[name = string("input_1605_cast_fp16")];
            tensor<int32, [6]> input_1607_pad_0 = const()[name = string("input_1607_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1607_mode_0 = const()[name = string("input_1607_mode_0"), val = string("constant")];
            fp16 const_373_to_fp16 = const()[name = string("const_373_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1607_cast_fp16 = pad(constant_val = const_373_to_fp16, mode = input_1607_mode_0, pad = input_1607_pad_0, x = input_1605_cast_fp16)[name = string("input_1607_cast_fp16")];
            string input_1609_pad_type_0 = const()[name = string("input_1609_pad_type_0"), val = string("valid")];
            int32 input_1609_groups_0 = const()[name = string("input_1609_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1609_strides_0 = const()[name = string("input_1609_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1609_pad_0 = const()[name = string("input_1609_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1609_dilations_0 = const()[name = string("input_1609_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_444_to_fp16 = const()[name = string("const_444_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426183488)))];
            tensor<fp16, [1024]> const_445_to_fp16 = const()[name = string("const_445_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426201984)))];
            tensor<fp16, [1, 1024, 188]> input_1611_cast_fp16 = conv(bias = const_445_to_fp16, dilations = input_1609_dilations_0, groups = input_1609_groups_0, pad = input_1609_pad_0, pad_type = input_1609_pad_type_0, strides = input_1609_strides_0, weight = const_444_to_fp16, x = input_1607_cast_fp16)[name = string("input_1611_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1613_cast_fp16 = silu(x = input_1611_cast_fp16)[name = string("input_1613_cast_fp16")];
            string x_681_pad_type_0 = const()[name = string("x_681_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_681_strides_0 = const()[name = string("x_681_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_681_pad_0 = const()[name = string("x_681_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_681_dilations_0 = const()[name = string("x_681_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_681_groups_0 = const()[name = string("x_681_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_30_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426204096))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426728448))))[name = string("module_layers_30_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_30_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_30_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426794048)))];
            tensor<fp16, [1, 1024, 188]> x_681_cast_fp16 = conv(bias = module_layers_30_conv_pointwise_conv2_bias_to_fp16, dilations = x_681_dilations_0, groups = x_681_groups_0, pad = x_681_pad_0, pad_type = x_681_pad_type_0, strides = x_681_strides_0, weight = module_layers_30_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1613_cast_fp16)[name = string("x_681_cast_fp16")];
            tensor<int32, [3]> input_1615_perm_0 = const()[name = string("input_1615_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1615_cast_fp16 = transpose(perm = input_1615_perm_0, x = x_681_cast_fp16)[name = string("transpose_200")];
            tensor<fp16, [1, 188, 1024]> input_1617_cast_fp16 = add(x = input_1599_cast_fp16, y = input_1615_cast_fp16)[name = string("input_1617_cast_fp16")];
            tensor<int32, [1]> input_1619_axes_0 = const()[name = string("input_1619_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_30_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_30_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426796160)))];
            tensor<fp16, [1024]> module_layers_30_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_30_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426798272)))];
            tensor<fp16, [1, 188, 1024]> input_1619_cast_fp16 = layer_norm(axes = input_1619_axes_0, beta = module_layers_30_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_30_norm_feed_forward2_weight_to_fp16, x = input_1617_cast_fp16)[name = string("input_1619_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_30_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(426800384))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(428897600))))[name = string("module_layers_30_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_30_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_30_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429159808)))];
            tensor<fp16, [1, 188, 4096]> linear_278_cast_fp16 = linear(bias = module_layers_30_feed_forward2_linear1_bias_to_fp16, weight = module_layers_30_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1619_cast_fp16)[name = string("linear_278_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1623_cast_fp16 = silu(x = linear_278_cast_fp16)[name = string("input_1623_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_30_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429168064))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431265280))))[name = string("module_layers_30_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_30_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_30_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431527488)))];
            tensor<fp16, [1, 188, 1024]> linear_279_cast_fp16 = linear(bias = module_layers_30_feed_forward2_linear2_bias_to_fp16, weight = module_layers_30_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1623_cast_fp16)[name = string("linear_279_cast_fp16")];
            fp16 var_5745_to_fp16 = const()[name = string("op_5745_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5746_cast_fp16 = mul(x = linear_279_cast_fp16, y = var_5745_to_fp16)[name = string("op_5746_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1629_cast_fp16 = add(x = input_1617_cast_fp16, y = var_5746_cast_fp16)[name = string("input_1629_cast_fp16")];
            tensor<int32, [1]> input_1631_axes_0 = const()[name = string("input_1631_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_30_norm_out_weight_to_fp16 = const()[name = string("module_layers_30_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431529600)))];
            tensor<fp16, [1024]> module_layers_30_norm_out_bias_to_fp16 = const()[name = string("module_layers_30_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431531712)))];
            tensor<fp16, [1, 188, 1024]> input_1631_cast_fp16 = layer_norm(axes = input_1631_axes_0, beta = module_layers_30_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_30_norm_out_weight_to_fp16, x = input_1629_cast_fp16)[name = string("input_1631_cast_fp16")];
            tensor<int32, [1]> input_1633_axes_0 = const()[name = string("input_1633_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_31_norm_feed_forward1_weight_to_fp16 = const()[name = string("module_layers_31_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431533824)))];
            tensor<fp16, [1024]> module_layers_31_norm_feed_forward1_bias_to_fp16 = const()[name = string("module_layers_31_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431535936)))];
            tensor<fp16, [1, 188, 1024]> input_1633_cast_fp16 = layer_norm(axes = input_1633_axes_0, beta = module_layers_31_norm_feed_forward1_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_31_norm_feed_forward1_weight_to_fp16, x = input_1631_cast_fp16)[name = string("input_1633_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_31_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(431538048))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(433635264))))[name = string("module_layers_31_feed_forward1_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_31_feed_forward1_linear1_bias_to_fp16 = const()[name = string("module_layers_31_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(433897472)))];
            tensor<fp16, [1, 188, 4096]> linear_280_cast_fp16 = linear(bias = module_layers_31_feed_forward1_linear1_bias_to_fp16, weight = module_layers_31_feed_forward1_linear1_weight_to_fp16_quantized, x = input_1633_cast_fp16)[name = string("linear_280_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1637_cast_fp16 = silu(x = linear_280_cast_fp16)[name = string("input_1637_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_31_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(433905728))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(436002944))))[name = string("module_layers_31_feed_forward1_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_31_feed_forward1_linear2_bias_to_fp16 = const()[name = string("module_layers_31_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(436265152)))];
            tensor<fp16, [1, 188, 1024]> linear_281_cast_fp16 = linear(bias = module_layers_31_feed_forward1_linear2_bias_to_fp16, weight = module_layers_31_feed_forward1_linear2_weight_to_fp16_quantized, x = input_1637_cast_fp16)[name = string("linear_281_cast_fp16")];
            fp16 var_5776_to_fp16 = const()[name = string("op_5776_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5777_cast_fp16 = mul(x = linear_281_cast_fp16, y = var_5776_to_fp16)[name = string("op_5777_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1643_cast_fp16 = add(x = input_1631_cast_fp16, y = var_5777_cast_fp16)[name = string("input_1643_cast_fp16")];
            tensor<int32, [1]> query_axes_0 = const()[name = string("query_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_31_norm_self_att_weight_to_fp16 = const()[name = string("module_layers_31_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(436267264)))];
            tensor<fp16, [1024]> module_layers_31_norm_self_att_bias_to_fp16 = const()[name = string("module_layers_31_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(436269376)))];
            tensor<fp16, [1, 188, 1024]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = module_layers_31_norm_self_att_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_31_norm_self_att_weight_to_fp16, x = input_1643_cast_fp16)[name = string("query_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_31_self_attn_linear_q_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(436271488))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(436795840))))[name = string("module_layers_31_self_attn_linear_q_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_31_self_attn_linear_q_bias_to_fp16 = const()[name = string("module_layers_31_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(436861440)))];
            tensor<fp16, [1, 188, 1024]> linear_282_cast_fp16 = linear(bias = module_layers_31_self_attn_linear_q_bias_to_fp16, weight = module_layers_31_self_attn_linear_q_weight_to_fp16_quantized, x = query_cast_fp16)[name = string("linear_282_cast_fp16")];
            tensor<int32, [4]> var_5794 = const()[name = string("op_5794"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> q_187_cast_fp16 = reshape(shape = var_5794, x = linear_282_cast_fp16)[name = string("q_187_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_31_self_attn_linear_k_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(436863552))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437387904))))[name = string("module_layers_31_self_attn_linear_k_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_31_self_attn_linear_k_bias_to_fp16 = const()[name = string("module_layers_31_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437453504)))];
            tensor<fp16, [1, 188, 1024]> linear_283_cast_fp16 = linear(bias = module_layers_31_self_attn_linear_k_bias_to_fp16, weight = module_layers_31_self_attn_linear_k_weight_to_fp16_quantized, x = query_cast_fp16)[name = string("linear_283_cast_fp16")];
            tensor<int32, [4]> var_5799 = const()[name = string("op_5799"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> k_125_cast_fp16 = reshape(shape = var_5799, x = linear_283_cast_fp16)[name = string("k_125_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_31_self_attn_linear_v_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437455616))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(437979968))))[name = string("module_layers_31_self_attn_linear_v_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_31_self_attn_linear_v_bias_to_fp16 = const()[name = string("module_layers_31_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438045568)))];
            tensor<fp16, [1, 188, 1024]> linear_284_cast_fp16 = linear(bias = module_layers_31_self_attn_linear_v_bias_to_fp16, weight = module_layers_31_self_attn_linear_v_weight_to_fp16_quantized, x = query_cast_fp16)[name = string("linear_284_cast_fp16")];
            tensor<int32, [4]> var_5804 = const()[name = string("op_5804"), val = tensor<int32, [4]>([1, -1, 8, 128])];
            tensor<fp16, [1, 188, 8, 128]> v_cast_fp16 = reshape(shape = var_5804, x = linear_284_cast_fp16)[name = string("v_cast_fp16")];
            tensor<int32, [4]> value_perm_0 = const()[name = string("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 128]> module_layers_31_self_attn_pos_bias_u_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438047680))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438048256))))[name = string("module_layers_31_self_attn_pos_bias_u_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5816_cast_fp16 = add(x = q_187_cast_fp16, y = module_layers_31_self_attn_pos_bias_u_to_fp16_quantized)[name = string("op_5816_cast_fp16")];
            tensor<fp16, [8, 128]> module_layers_31_self_attn_pos_bias_v_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [8, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438048384))), scale = tensor<fp16, [8, 4]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438048960))))[name = string("module_layers_31_self_attn_pos_bias_v_to_fp16_quantized")];
            tensor<fp16, [1, 188, 8, 128]> var_5818_cast_fp16 = add(x = q_187_cast_fp16, y = module_layers_31_self_attn_pos_bias_v_to_fp16_quantized)[name = string("op_5818_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_perm_0 = const()[name = string("q_with_bias_v_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            bool x_689_transpose_x_0 = const()[name = string("x_689_transpose_x_0"), val = bool(false)];
            bool x_689_transpose_y_0 = const()[name = string("x_689_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 128, 375]> op_5820_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1, 8, 128, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438049088))), scale = tensor<fp16, [1, 1, 4, 375]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438241152))))[name = string("op_5820_to_fp16_quantized")];
            tensor<fp16, [1, 8, 188, 128]> q_with_bias_v_cast_fp16 = transpose(perm = q_with_bias_v_perm_0, x = var_5818_cast_fp16)[name = string("transpose_199")];
            tensor<fp16, [1, 8, 188, 375]> x_689_cast_fp16 = matmul(transpose_x = x_689_transpose_x_0, transpose_y = x_689_transpose_y_0, x = q_with_bias_v_cast_fp16, y = op_5820_to_fp16_quantized)[name = string("x_689_cast_fp16")];
            tensor<int32, [8]> x_691_pad_0 = const()[name = string("x_691_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            string x_691_mode_0 = const()[name = string("x_691_mode_0"), val = string("constant")];
            fp16 const_380_to_fp16 = const()[name = string("const_380_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 8, 188, 376]> x_691_cast_fp16 = pad(constant_val = const_380_to_fp16, mode = x_691_mode_0, pad = x_691_pad_0, x = x_689_cast_fp16)[name = string("x_691_cast_fp16")];
            tensor<int32, [4]> var_5828 = const()[name = string("op_5828"), val = tensor<int32, [4]>([1, 8, -1, 188])];
            tensor<fp16, [1, 8, 376, 188]> x_693_cast_fp16 = reshape(shape = var_5828, x = x_691_cast_fp16)[name = string("x_693_cast_fp16")];
            tensor<int32, [4]> var_5832_begin_0 = const()[name = string("op_5832_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_5832_end_0 = const()[name = string("op_5832_end_0"), val = tensor<int32, [4]>([1, 8, 376, 188])];
            tensor<bool, [4]> var_5832_end_mask_0 = const()[name = string("op_5832_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 375, 188]> var_5832_cast_fp16 = slice_by_index(begin = var_5832_begin_0, end = var_5832_end_0, end_mask = var_5832_end_mask_0, x = x_693_cast_fp16)[name = string("op_5832_cast_fp16")];
            tensor<int32, [4]> var_5833 = const()[name = string("op_5833"), val = tensor<int32, [4]>([1, 8, 188, 375])];
            tensor<fp16, [1, 8, 188, 375]> matrix_bd_125_cast_fp16 = reshape(shape = var_5833, x = var_5832_cast_fp16)[name = string("matrix_bd_125_cast_fp16")];
            bool matrix_ac_transpose_x_0 = const()[name = string("matrix_ac_transpose_x_0"), val = bool(false)];
            bool matrix_ac_transpose_y_0 = const()[name = string("matrix_ac_transpose_y_0"), val = bool(false)];
            tensor<int32, [4]> transpose_190_perm_0 = const()[name = string("transpose_190_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_191_perm_0 = const()[name = string("transpose_191_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 128, 188]> transpose_191 = transpose(perm = transpose_191_perm_0, x = k_125_cast_fp16)[name = string("transpose_197")];
            tensor<fp16, [1, 8, 188, 128]> transpose_190 = transpose(perm = transpose_190_perm_0, x = var_5816_cast_fp16)[name = string("transpose_198")];
            tensor<fp16, [1, 8, 188, 188]> matrix_ac_cast_fp16 = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_190, y = transpose_191)[name = string("matrix_ac_cast_fp16")];
            tensor<int32, [4]> matrix_bd_begin_0 = const()[name = string("matrix_bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_end_0 = const()[name = string("matrix_bd_end_0"), val = tensor<int32, [4]>([1, 8, 188, 188])];
            tensor<bool, [4]> matrix_bd_end_mask_0 = const()[name = string("matrix_bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 188, 188]> matrix_bd_cast_fp16 = slice_by_index(begin = matrix_bd_begin_0, end = matrix_bd_end_0, end_mask = matrix_bd_end_mask_0, x = matrix_bd_125_cast_fp16)[name = string("matrix_bd_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5842_cast_fp16 = add(x = matrix_ac_cast_fp16, y = matrix_bd_cast_fp16)[name = string("op_5842_cast_fp16")];
            fp16 _inversed_scores_125_y_0_to_fp16 = const()[name = string("_inversed_scores_125_y_0_to_fp16"), val = fp16(0x1.6ap-4)];
            tensor<fp16, [1, 8, 188, 188]> _inversed_scores_125_cast_fp16 = mul(x = var_5842_cast_fp16, y = _inversed_scores_125_y_0_to_fp16)[name = string("_inversed_scores_125_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> scores_cast_fp16 = select(a = var_12_to_fp16, b = _inversed_scores_125_cast_fp16, cond = mask_11)[name = string("scores_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> var_5848_cast_fp16 = softmax(axis = var_23, x = scores_cast_fp16)[name = string("op_5848_cast_fp16")];
            tensor<fp16, [1, 8, 188, 188]> input_1645_cast_fp16 = select(a = var_11_to_fp16, b = var_5848_cast_fp16, cond = mask_11)[name = string("input_1645_cast_fp16")];
            bool x_695_transpose_x_0 = const()[name = string("x_695_transpose_x_0"), val = bool(false)];
            bool x_695_transpose_y_0 = const()[name = string("x_695_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 8, 188, 128]> value_cast_fp16 = transpose(perm = value_perm_0, x = v_cast_fp16)[name = string("transpose_196")];
            tensor<fp16, [1, 8, 188, 128]> x_695_cast_fp16 = matmul(transpose_x = x_695_transpose_x_0, transpose_y = x_695_transpose_y_0, x = input_1645_cast_fp16, y = value_cast_fp16)[name = string("x_695_cast_fp16")];
            tensor<int32, [4]> var_5852_perm_0 = const()[name = string("op_5852_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5853 = const()[name = string("op_5853"), val = tensor<int32, [3]>([1, -1, 1024])];
            tensor<fp16, [1, 188, 8, 128]> var_5852_cast_fp16 = transpose(perm = var_5852_perm_0, x = x_695_cast_fp16)[name = string("transpose_195")];
            tensor<fp16, [1, 188, 1024]> input_1647_cast_fp16 = reshape(shape = var_5853, x = var_5852_cast_fp16)[name = string("input_1647_cast_fp16")];
            tensor<fp16, [1024, 1024]> module_layers_31_self_attn_linear_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438244224))), scale = tensor<fp16, [1024, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438768576))))[name = string("module_layers_31_self_attn_linear_out_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_31_self_attn_linear_out_bias_to_fp16 = const()[name = string("module_layers_31_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438834176)))];
            tensor<fp16, [1, 188, 1024]> linear_286_cast_fp16 = linear(bias = module_layers_31_self_attn_linear_out_bias_to_fp16, weight = module_layers_31_self_attn_linear_out_weight_to_fp16_quantized, x = input_1647_cast_fp16)[name = string("linear_286_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_1651_cast_fp16 = add(x = input_1643_cast_fp16, y = linear_286_cast_fp16)[name = string("input_1651_cast_fp16")];
            tensor<int32, [1]> x_699_axes_0 = const()[name = string("x_699_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_31_norm_conv_weight_to_fp16 = const()[name = string("module_layers_31_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438836288)))];
            tensor<fp16, [1024]> module_layers_31_norm_conv_bias_to_fp16 = const()[name = string("module_layers_31_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438838400)))];
            tensor<fp16, [1, 188, 1024]> x_699_cast_fp16 = layer_norm(axes = x_699_axes_0, beta = module_layers_31_norm_conv_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_31_norm_conv_weight_to_fp16, x = input_1651_cast_fp16)[name = string("x_699_cast_fp16")];
            tensor<int32, [3]> input_1653_perm_0 = const()[name = string("input_1653_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string input_1655_pad_type_0 = const()[name = string("input_1655_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> input_1655_strides_0 = const()[name = string("input_1655_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1655_pad_0 = const()[name = string("input_1655_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1655_dilations_0 = const()[name = string("input_1655_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 input_1655_groups_0 = const()[name = string("input_1655_groups_0"), val = int32(1)];
            tensor<fp16, [2048, 1024, 1]> module_layers_31_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [2048, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438840512))), scale = tensor<fp16, [2048, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(439889152))))[name = string("module_layers_31_conv_pointwise_conv1_weight_to_fp16_quantized")];
            tensor<fp16, [2048]> module_layers_31_conv_pointwise_conv1_bias_to_fp16 = const()[name = string("module_layers_31_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440020288)))];
            tensor<fp16, [1, 1024, 188]> input_1653_cast_fp16 = transpose(perm = input_1653_perm_0, x = x_699_cast_fp16)[name = string("transpose_194")];
            tensor<fp16, [1, 2048, 188]> input_1655_cast_fp16 = conv(bias = module_layers_31_conv_pointwise_conv1_bias_to_fp16, dilations = input_1655_dilations_0, groups = input_1655_groups_0, pad = input_1655_pad_0, pad_type = input_1655_pad_type_0, strides = input_1655_strides_0, weight = module_layers_31_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_1653_cast_fp16)[name = string("input_1655_cast_fp16")];
            int32 x_701_split_num_splits_0 = const()[name = string("x_701_split_num_splits_0"), val = int32(2)];
            int32 x_701_split_axis_0 = const()[name = string("x_701_split_axis_0"), val = int32(1)];
            tensor<fp16, [1, 1024, 188]> x_701_split_cast_fp16_0, tensor<fp16, [1, 1024, 188]> x_701_split_cast_fp16_1 = split(axis = x_701_split_axis_0, num_splits = x_701_split_num_splits_0, x = input_1655_cast_fp16)[name = string("x_701_split_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_701_split_1_sigmoid_cast_fp16 = sigmoid(x = x_701_split_cast_fp16_1)[name = string("x_701_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> x_701_cast_fp16 = mul(x = x_701_split_cast_fp16_0, y = x_701_split_1_sigmoid_cast_fp16)[name = string("x_701_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1657_cast_fp16 = select(a = var_11_to_fp16, b = x_701_cast_fp16, cond = var_483)[name = string("input_1657_cast_fp16")];
            tensor<int32, [6]> input_1659_pad_0 = const()[name = string("input_1659_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            string input_1659_mode_0 = const()[name = string("input_1659_mode_0"), val = string("constant")];
            fp16 const_383_to_fp16 = const()[name = string("const_383_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 196]> input_1659_cast_fp16 = pad(constant_val = const_383_to_fp16, mode = input_1659_mode_0, pad = input_1659_pad_0, x = input_1657_cast_fp16)[name = string("input_1659_cast_fp16")];
            string input_1661_pad_type_0 = const()[name = string("input_1661_pad_type_0"), val = string("valid")];
            int32 input_1661_groups_0 = const()[name = string("input_1661_groups_0"), val = int32(1024)];
            tensor<int32, [1]> input_1661_strides_0 = const()[name = string("input_1661_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_1661_pad_0 = const()[name = string("input_1661_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_1661_dilations_0 = const()[name = string("input_1661_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1024, 1, 9]> const_446_to_fp16 = const()[name = string("const_446_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440024448)))];
            tensor<fp16, [1024]> const_447_to_fp16 = const()[name = string("const_447_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440042944)))];
            tensor<fp16, [1, 1024, 188]> input_1663_cast_fp16 = conv(bias = const_447_to_fp16, dilations = input_1661_dilations_0, groups = input_1661_groups_0, pad = input_1661_pad_0, pad_type = input_1661_pad_type_0, strides = input_1661_strides_0, weight = const_446_to_fp16, x = input_1659_cast_fp16)[name = string("input_1663_cast_fp16")];
            tensor<fp16, [1, 1024, 188]> input_1665_cast_fp16 = silu(x = input_1663_cast_fp16)[name = string("input_1665_cast_fp16")];
            string x_703_pad_type_0 = const()[name = string("x_703_pad_type_0"), val = string("valid")];
            tensor<int32, [1]> x_703_strides_0 = const()[name = string("x_703_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_703_pad_0 = const()[name = string("x_703_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_703_dilations_0 = const()[name = string("x_703_dilations_0"), val = tensor<int32, [1]>([1])];
            int32 x_703_groups_0 = const()[name = string("x_703_groups_0"), val = int32(1)];
            tensor<fp16, [1024, 1024, 1]> module_layers_31_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 1024, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440045056))), scale = tensor<fp16, [1024, 32, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440569408))))[name = string("module_layers_31_conv_pointwise_conv2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_31_conv_pointwise_conv2_bias_to_fp16 = const()[name = string("module_layers_31_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440635008)))];
            tensor<fp16, [1, 1024, 188]> x_703_cast_fp16 = conv(bias = module_layers_31_conv_pointwise_conv2_bias_to_fp16, dilations = x_703_dilations_0, groups = x_703_groups_0, pad = x_703_pad_0, pad_type = x_703_pad_type_0, strides = x_703_strides_0, weight = module_layers_31_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_1665_cast_fp16)[name = string("x_703_cast_fp16")];
            tensor<int32, [3]> input_1667_perm_0 = const()[name = string("input_1667_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 188, 1024]> input_1667_cast_fp16 = transpose(perm = input_1667_perm_0, x = x_703_cast_fp16)[name = string("transpose_193")];
            tensor<fp16, [1, 188, 1024]> input_1669_cast_fp16 = add(x = input_1651_cast_fp16, y = input_1667_cast_fp16)[name = string("input_1669_cast_fp16")];
            tensor<int32, [1]> input_1671_axes_0 = const()[name = string("input_1671_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_31_norm_feed_forward2_weight_to_fp16 = const()[name = string("module_layers_31_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440637120)))];
            tensor<fp16, [1024]> module_layers_31_norm_feed_forward2_bias_to_fp16 = const()[name = string("module_layers_31_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440639232)))];
            tensor<fp16, [1, 188, 1024]> input_1671_cast_fp16 = layer_norm(axes = input_1671_axes_0, beta = module_layers_31_norm_feed_forward2_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_31_norm_feed_forward2_weight_to_fp16, x = input_1669_cast_fp16)[name = string("input_1671_cast_fp16")];
            tensor<fp16, [4096, 1024]> module_layers_31_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440641344))), scale = tensor<fp16, [4096, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(442738560))))[name = string("module_layers_31_feed_forward2_linear1_weight_to_fp16_quantized")];
            tensor<fp16, [4096]> module_layers_31_feed_forward2_linear1_bias_to_fp16 = const()[name = string("module_layers_31_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(443000768)))];
            tensor<fp16, [1, 188, 4096]> linear_287_cast_fp16 = linear(bias = module_layers_31_feed_forward2_linear1_bias_to_fp16, weight = module_layers_31_feed_forward2_linear1_weight_to_fp16_quantized, x = input_1671_cast_fp16)[name = string("linear_287_cast_fp16")];
            tensor<fp16, [1, 188, 4096]> input_1675_cast_fp16 = silu(x = linear_287_cast_fp16)[name = string("input_1675_cast_fp16")];
            tensor<fp16, [1024, 4096]> module_layers_31_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int4, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(443009024))), scale = tensor<fp16, [1024, 128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(445106240))))[name = string("module_layers_31_feed_forward2_linear2_weight_to_fp16_quantized")];
            tensor<fp16, [1024]> module_layers_31_feed_forward2_linear2_bias_to_fp16 = const()[name = string("module_layers_31_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(445368448)))];
            tensor<fp16, [1, 188, 1024]> linear_288_cast_fp16 = linear(bias = module_layers_31_feed_forward2_linear2_bias_to_fp16, weight = module_layers_31_feed_forward2_linear2_weight_to_fp16_quantized, x = input_1675_cast_fp16)[name = string("linear_288_cast_fp16")];
            fp16 var_5919_to_fp16 = const()[name = string("op_5919_to_fp16"), val = fp16(0x1p-1)];
            tensor<fp16, [1, 188, 1024]> var_5920_cast_fp16 = mul(x = linear_288_cast_fp16, y = var_5919_to_fp16)[name = string("op_5920_cast_fp16")];
            tensor<fp16, [1, 188, 1024]> input_cast_fp16 = add(x = input_1669_cast_fp16, y = var_5920_cast_fp16)[name = string("input_cast_fp16")];
            tensor<int32, [1]> audio_signal_axes_0 = const()[name = string("audio_signal_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [1024]> module_layers_31_norm_out_weight_to_fp16 = const()[name = string("module_layers_31_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(445370560)))];
            tensor<fp16, [1024]> module_layers_31_norm_out_bias_to_fp16 = const()[name = string("module_layers_31_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(445372672)))];
            tensor<fp16, [1, 188, 1024]> audio_signal_cast_fp16 = layer_norm(axes = audio_signal_axes_0, beta = module_layers_31_norm_out_bias_to_fp16, epsilon = var_9_to_fp16, gamma = module_layers_31_norm_out_weight_to_fp16, x = input_cast_fp16)[name = string("audio_signal_cast_fp16")];
            tensor<int32, [3]> obj_1_perm_0 = const()[name = string("obj_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            string obj_1_cast_fp16_to_fp32_dtype_0 = const()[name = string("obj_1_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
            tensor<fp16, [1, 1024, 188]> obj_1_cast_fp16 = transpose(perm = obj_1_perm_0, x = audio_signal_cast_fp16)[name = string("transpose_192")];
            tensor<fp32, [1, 1024, 188]> encoder = cast(dtype = obj_1_cast_fp16_to_fp32_dtype_0, x = obj_1_cast_fp16)[name = string("cast_0")];
        } -> (encoder, encoder_length);
}