program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
{
    func main<ios17>(tensor<fp32, [1, 64, 3000]> audio_signal) {
            tensor<string, []> conv1d_pad_type_0 = const()[name = tensor<string, []>("conv1d_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_pad_0 = const()[name = tensor<string, []>("conv1d_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [1]> conv1d_strides_0 = const()[name = tensor<string, []>("conv1d_strides_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1]> conv1d_dilations_0 = const()[name = tensor<string, []>("conv1d_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_groups_0 = const()[name = tensor<string, []>("conv1d_groups_0"), val = tensor<int32, []>(1)];
            tensor<string, []> audio_signal_to_fp16_dtype_0 = const()[name = tensor<string, []>("audio_signal_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [768, 64, 5]> p_encoder_pre_encode_conv_0_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_pre_encode_conv_0_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 64, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(246720))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_pre_encode_conv_0_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_pre_encode_conv_0_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(248320)))];
            tensor<fp16, [1, 64, 3000]> audio_signal_to_fp16 = cast(dtype = audio_signal_to_fp16_dtype_0, x = audio_signal)[name = tensor<string, []>("cast_0")];
            tensor<fp16, [1, 768, 1500]> conv1d_cast_fp16 = conv(bias = p_encoder_pre_encode_conv_0_bias_to_fp16, dilations = conv1d_dilations_0, groups = conv1d_groups_0, pad = conv1d_pad_0, pad_type = conv1d_pad_type_0, strides = conv1d_strides_0, weight = p_encoder_pre_encode_conv_0_weight_to_fp16_quantized, x = audio_signal_to_fp16)[name = tensor<string, []>("conv1d_cast_fp16")];
            tensor<fp16, [1, 768, 1500]> relu_cast_fp16 = relu(x = conv1d_cast_fp16)[name = tensor<string, []>("relu_cast_fp16")];
            tensor<string, []> conv1d_1_pad_type_0 = const()[name = tensor<string, []>("conv1d_1_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_1_pad_0 = const()[name = tensor<string, []>("conv1d_1_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [1]> conv1d_1_strides_0 = const()[name = tensor<string, []>("conv1d_1_strides_0"), val = tensor<int32, [1]>([2])];
            tensor<int32, [1]> conv1d_1_dilations_0 = const()[name = tensor<string, []>("conv1d_1_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_1_groups_0 = const()[name = tensor<string, []>("conv1d_1_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 5]> p_encoder_pre_encode_conv_2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_pre_encode_conv_2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(249920))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3199104))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_pre_encode_conv_2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_pre_encode_conv_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3200704)))];
            tensor<fp16, [1, 768, 750]> conv1d_1_cast_fp16 = conv(bias = p_encoder_pre_encode_conv_2_bias_to_fp16, dilations = conv1d_1_dilations_0, groups = conv1d_1_groups_0, pad = conv1d_1_pad_0, pad_type = conv1d_1_pad_type_0, strides = conv1d_1_strides_0, weight = p_encoder_pre_encode_conv_2_weight_to_fp16_quantized, x = relu_cast_fp16)[name = tensor<string, []>("conv1d_1_cast_fp16")];
            tensor<fp16, [1, 768, 750]> relu_1_cast_fp16 = relu(x = conv1d_1_cast_fp16)[name = tensor<string, []>("relu_1_cast_fp16")];
            tensor<int32, [3]> transpose_2_perm_0 = const()[name = tensor<string, []>("transpose_2_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_axes_0 = const()[name = tensor<string, []>("layer_norm_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3202304)))];
            tensor<fp16, [768]> p_encoder_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3203904)))];
            tensor<fp16, []> layer_norm_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_2_cast_fp16 = transpose(perm = transpose_2_perm_0, x = relu_1_cast_fp16)[name = tensor<string, []>("transpose_257")];
            tensor<fp16, [1, 750, 768]> layer_norm_cast_fp16 = layer_norm(axes = layer_norm_axes_0, beta = p_encoder_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_epsilon_0_to_fp16, gamma = p_encoder_layers_0_norm_feed_forward1_weight_to_fp16, x = transpose_2_cast_fp16)[name = tensor<string, []>("layer_norm_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_0_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3205504))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5568000))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_0_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5574208)))];
            tensor<fp16, [1, 750, 3072]> linear_0_cast_fp16 = linear(bias = p_encoder_layers_0_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_0_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_cast_fp16 = silu(x = linear_0_cast_fp16)[name = tensor<string, []>("silu_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_0_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5580416))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7939776))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_0_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7941376)))];
            tensor<fp16, [1, 750, 768]> linear_1_cast_fp16 = linear(bias = p_encoder_layers_0_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_0_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<fp16, []> const_34_to_fp16 = const()[name = tensor<string, []>("const_34_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_cast_fp16 = mul(x = linear_1_cast_fp16, y = const_34_to_fp16)[name = tensor<string, []>("mul_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_4_cast_fp16 = add(x = transpose_2_cast_fp16, y = mul_cast_fp16)[name = tensor<string, []>("add_4_cast_fp16")];
            tensor<int32, [1]> layer_norm_1_axes_0 = const()[name = tensor<string, []>("layer_norm_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_0_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7942976)))];
            tensor<fp16, [768]> p_encoder_layers_0_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7944576)))];
            tensor<fp16, []> layer_norm_1_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_1_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_1_cast_fp16 = layer_norm(axes = layer_norm_1_axes_0, beta = p_encoder_layers_0_norm_self_att_bias_to_fp16, epsilon = layer_norm_1_epsilon_0_to_fp16, gamma = p_encoder_layers_0_norm_self_att_weight_to_fp16, x = add_4_cast_fp16)[name = tensor<string, []>("layer_norm_1_cast_fp16")];
            tensor<int32, [4]> const_38 = const()[name = tensor<string, []>("const_38"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_48_perm_1 = const()[name = tensor<string, []>("transpose_48_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_48 = transpose(perm = transpose_48_perm_1, x = layer_norm_1_cast_fp16)[name = tensor<string, []>("transpose_256")];
            tensor<fp16, [750, 1, 16, 48]> view_cast_fp16 = reshape(shape = const_38, x = transpose_48)[name = tensor<string, []>("view_cast_fp16")];
            tensor<fp16, [750, 1, 1, 48]> const_46_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_46_to_fp16_quantized"), quantized_data = tensor<int8, [750, 1, 1, 48]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7946176))), scale = tensor<fp16, [750]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7983104))), zero_point = tensor<int8, [750]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7982272)))];
            tensor<fp16, [750, 1, 16, 48]> mul_1_cast_fp16 = mul(x = view_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_1_cast_fp16")];
            tensor<int32, [4]> slice_3_begin_0 = const()[name = tensor<string, []>("slice_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_3_end_0 = const()[name = tensor<string, []>("slice_3_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_3_end_mask_0 = const()[name = tensor<string, []>("slice_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_3_cast_fp16 = slice_by_index(begin = slice_3_begin_0, end = slice_3_end_0, end_mask = slice_3_end_mask_0, x = view_cast_fp16)[name = tensor<string, []>("slice_3_cast_fp16")];
            tensor<int32, [4]> slice_4_begin_0 = const()[name = tensor<string, []>("slice_4_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_4_end_0 = const()[name = tensor<string, []>("slice_4_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_4_end_mask_0 = const()[name = tensor<string, []>("slice_4_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_4_cast_fp16 = slice_by_index(begin = slice_4_begin_0, end = slice_4_end_0, end_mask = slice_4_end_mask_0, x = view_cast_fp16)[name = tensor<string, []>("slice_4_cast_fp16")];
            tensor<fp16, []> const_55_promoted_to_fp16 = const()[name = tensor<string, []>("const_55_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_cast_fp16 = mul(x = slice_4_cast_fp16, y = const_55_promoted_to_fp16)[name = tensor<string, []>("neg_cast_fp16")];
            tensor<int32, []> const_56 = const()[name = tensor<string, []>("const_56"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_interleave_0 = const()[name = tensor<string, []>("cat_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_cast_fp16 = concat(axis = const_56, interleave = cat_interleave_0, values = (neg_cast_fp16, slice_3_cast_fp16))[name = tensor<string, []>("cat_cast_fp16")];
            tensor<fp16, [750, 1, 1, 48]> const_48_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_48_to_fp16_quantized"), quantized_data = tensor<int8, [750, 1, 1, 48]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7984704))), scale = tensor<fp16, [750]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8020800))), zero_point = tensor<int8, [750]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7982272)))];
            tensor<fp16, [750, 1, 16, 48]> mul_2_cast_fp16 = mul(x = cat_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_2_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_5_cast_fp16 = add(x = mul_1_cast_fp16, y = mul_2_cast_fp16)[name = tensor<string, []>("add_5_cast_fp16")];
            tensor<int32, [3]> const_65 = const()[name = tensor<string, []>("const_65"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_3_cast_fp16 = reshape(shape = const_65, x = add_5_cast_fp16)[name = tensor<string, []>("view_3_cast_fp16")];
            tensor<int32, [3]> transpose_6_perm_0 = const()[name = tensor<string, []>("transpose_6_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_0_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8022400))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8612288))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_0_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8613888)))];
            tensor<fp16, [1, 750, 768]> transpose_6_cast_fp16 = transpose(perm = transpose_6_perm_0, x = view_3_cast_fp16)[name = tensor<string, []>("transpose_255")];
            tensor<fp16, [1, 750, 768]> linear_2_cast_fp16 = linear(bias = p_encoder_layers_0_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_0_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_6_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<int32, [4]> const_74 = const()[name = tensor<string, []>("const_74"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_6_cast_fp16 = reshape(shape = const_74, x = linear_2_cast_fp16)[name = tensor<string, []>("view_6_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_0_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8615488))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9205376))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_0_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9206976)))];
            tensor<fp16, [1, 750, 768]> linear_3_cast_fp16 = linear(bias = p_encoder_layers_0_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_0_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_6_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<int32, [4]> const_75 = const()[name = tensor<string, []>("const_75"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_7_cast_fp16 = reshape(shape = const_75, x = linear_3_cast_fp16)[name = tensor<string, []>("view_7_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_0_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9208576))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9798464))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_0_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9800064)))];
            tensor<fp16, [1, 750, 768]> linear_4_cast_fp16 = linear(bias = p_encoder_layers_0_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_0_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_1_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<int32, [4]> const_76 = const()[name = tensor<string, []>("const_76"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_8_cast_fp16 = reshape(shape = const_76, x = linear_4_cast_fp16)[name = tensor<string, []>("view_8_cast_fp16")];
            tensor<int32, [4]> transpose_11_perm_0 = const()[name = tensor<string, []>("transpose_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_2_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_2_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_2_cast_fp16 = mul(x = view_7_cast_fp16, y = _inversed_div_2_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_2_cast_fp16")];
            tensor<bool, []> matmul_transpose_x_0 = const()[name = tensor<string, []>("matmul_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_transpose_y_0 = const()[name = tensor<string, []>("matmul_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_64_perm_0 = const()[name = tensor<string, []>("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_65_perm_0 = const()[name = tensor<string, []>("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_65 = transpose(perm = transpose_65_perm_0, x = _inversed_div_2_cast_fp16)[name = tensor<string, []>("transpose_253")];
            tensor<fp16, [1, 16, 750, 48]> transpose_64 = transpose(perm = transpose_64_perm_0, x = view_6_cast_fp16)[name = tensor<string, []>("transpose_254")];
            tensor<fp16, [1, 16, 750, 750]> matmul_cast_fp16 = matmul(transpose_x = matmul_transpose_x_0, transpose_y = matmul_transpose_y_0, x = transpose_64, y = transpose_65)[name = tensor<string, []>("matmul_cast_fp16")];
            tensor<int32, []> const_86 = const()[name = tensor<string, []>("const_86"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_cast_fp16 = softmax(axis = const_86, x = matmul_cast_fp16)[name = tensor<string, []>("softmax_cast_fp16")];
            tensor<bool, []> matmul_1_transpose_x_0 = const()[name = tensor<string, []>("matmul_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_1_transpose_y_0 = const()[name = tensor<string, []>("matmul_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_11_cast_fp16 = transpose(perm = transpose_11_perm_0, x = view_8_cast_fp16)[name = tensor<string, []>("transpose_252")];
            tensor<fp16, [1, 16, 750, 48]> matmul_1_cast_fp16 = matmul(transpose_x = matmul_1_transpose_x_0, transpose_y = matmul_1_transpose_y_0, x = softmax_cast_fp16, y = transpose_11_cast_fp16)[name = tensor<string, []>("matmul_1_cast_fp16")];
            tensor<int32, [4]> transpose_13_perm_0 = const()[name = tensor<string, []>("transpose_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_89 = const()[name = tensor<string, []>("const_89"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_13_cast_fp16 = transpose(perm = transpose_13_perm_0, x = matmul_1_cast_fp16)[name = tensor<string, []>("transpose_251")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_cast_fp16 = reshape(shape = const_89, x = transpose_13_cast_fp16)[name = tensor<string, []>("_unsafe_view_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_0_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9801664))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10391552))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_0_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10393152)))];
            tensor<fp16, [1, 750, 768]> linear_5_cast_fp16 = linear(bias = p_encoder_layers_0_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_0_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_7_cast_fp16 = add(x = add_4_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("add_7_cast_fp16")];
            tensor<int32, [1]> layer_norm_2_axes_0 = const()[name = tensor<string, []>("layer_norm_2_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_0_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10394752)))];
            tensor<fp16, [768]> p_encoder_layers_0_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10396352)))];
            tensor<fp16, []> layer_norm_2_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_2_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_2_cast_fp16 = layer_norm(axes = layer_norm_2_axes_0, beta = p_encoder_layers_0_norm_conv_bias_to_fp16, epsilon = layer_norm_2_epsilon_0_to_fp16, gamma = p_encoder_layers_0_norm_conv_weight_to_fp16, x = add_7_cast_fp16)[name = tensor<string, []>("layer_norm_2_cast_fp16")];
            tensor<int32, [3]> transpose_14_perm_0 = const()[name = tensor<string, []>("transpose_14_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_2_pad_type_0 = const()[name = tensor<string, []>("conv1d_2_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_2_strides_0 = const()[name = tensor<string, []>("conv1d_2_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_2_pad_0 = const()[name = tensor<string, []>("conv1d_2_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_2_dilations_0 = const()[name = tensor<string, []>("conv1d_2_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_2_groups_0 = const()[name = tensor<string, []>("conv1d_2_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10397952))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11579264))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_0_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11582400)))];
            tensor<fp16, [1, 768, 750]> transpose_14_cast_fp16 = transpose(perm = transpose_14_perm_0, x = layer_norm_2_cast_fp16)[name = tensor<string, []>("transpose_250")];
            tensor<fp16, [1, 1536, 750]> conv1d_2_cast_fp16 = conv(bias = p_encoder_layers_0_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_2_dilations_0, groups = conv1d_2_groups_0, pad = conv1d_2_pad_0, pad_type = conv1d_2_pad_type_0, strides = conv1d_2_strides_0, weight = p_encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_14_cast_fp16)[name = tensor<string, []>("conv1d_2_cast_fp16")];
            tensor<int32, []> glu_split_num_splits_0 = const()[name = tensor<string, []>("glu_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_split_axis_0 = const()[name = tensor<string, []>("glu_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_split_cast_fp16_1 = split(axis = glu_split_axis_0, num_splits = glu_split_num_splits_0, x = conv1d_2_cast_fp16)[name = tensor<string, []>("glu_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_split_cast_fp16_1)[name = tensor<string, []>("glu_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_cast_fp16 = mul(x = glu_split_cast_fp16_0, y = glu_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_cast_fp16")];
            tensor<string, []> conv1d_3_pad_type_0 = const()[name = tensor<string, []>("conv1d_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_3_pad_0 = const()[name = tensor<string, []>("conv1d_3_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_3_groups_0 = const()[name = tensor<string, []>("conv1d_3_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_3_strides_0 = const()[name = tensor<string, []>("conv1d_3_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_3_dilations_0 = const()[name = tensor<string, []>("conv1d_3_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_0_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11585536))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11589440))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_0_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11591040)))];
            tensor<fp16, [1, 768, 750]> conv1d_3_cast_fp16 = conv(bias = p_encoder_layers_0_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_3_dilations_0, groups = conv1d_3_groups_0, pad = conv1d_3_pad_0, pad_type = conv1d_3_pad_type_0, strides = conv1d_3_strides_0, weight = p_encoder_layers_0_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_cast_fp16)[name = tensor<string, []>("conv1d_3_cast_fp16")];
            tensor<int32, [3]> transpose_15_perm_0 = const()[name = tensor<string, []>("transpose_15_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_3_axes_0 = const()[name = tensor<string, []>("layer_norm_3_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_0_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11592640)))];
            tensor<fp16, [768]> p_encoder_layers_0_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11594240)))];
            tensor<fp16, []> layer_norm_3_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_3_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_15_cast_fp16 = transpose(perm = transpose_15_perm_0, x = conv1d_3_cast_fp16)[name = tensor<string, []>("transpose_249")];
            tensor<fp16, [1, 750, 768]> layer_norm_3_cast_fp16 = layer_norm(axes = layer_norm_3_axes_0, beta = p_encoder_layers_0_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_3_epsilon_0_to_fp16, gamma = p_encoder_layers_0_conv_batch_norm_weight_to_fp16, x = transpose_15_cast_fp16)[name = tensor<string, []>("layer_norm_3_cast_fp16")];
            tensor<int32, [3]> transpose_16_perm_0 = const()[name = tensor<string, []>("transpose_16_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_16_1_cast_fp16 = transpose(perm = transpose_16_perm_0, x = layer_norm_3_cast_fp16)[name = tensor<string, []>("transpose_248")];
            tensor<fp16, [1, 768, 750]> silu_1_cast_fp16 = silu(x = transpose_16_1_cast_fp16)[name = tensor<string, []>("silu_1_cast_fp16")];
            tensor<string, []> conv1d_4_pad_type_0 = const()[name = tensor<string, []>("conv1d_4_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_4_strides_0 = const()[name = tensor<string, []>("conv1d_4_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_4_pad_0 = const()[name = tensor<string, []>("conv1d_4_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_4_dilations_0 = const()[name = tensor<string, []>("conv1d_4_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_4_groups_0 = const()[name = tensor<string, []>("conv1d_4_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11595840))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12185728))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_0_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12187328)))];
            tensor<fp16, [1, 768, 750]> conv1d_4_cast_fp16 = conv(bias = p_encoder_layers_0_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_4_dilations_0, groups = conv1d_4_groups_0, pad = conv1d_4_pad_0, pad_type = conv1d_4_pad_type_0, strides = conv1d_4_strides_0, weight = p_encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_1_cast_fp16)[name = tensor<string, []>("conv1d_4_cast_fp16")];
            tensor<int32, [3]> transpose_17_perm_0 = const()[name = tensor<string, []>("transpose_17_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_17_1_cast_fp16 = transpose(perm = transpose_17_perm_0, x = conv1d_4_cast_fp16)[name = tensor<string, []>("transpose_247")];
            tensor<fp16, [1, 750, 768]> add_8_cast_fp16 = add(x = add_7_cast_fp16, y = transpose_17_1_cast_fp16)[name = tensor<string, []>("add_8_cast_fp16")];
            tensor<int32, [1]> layer_norm_4_axes_0 = const()[name = tensor<string, []>("layer_norm_4_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12188928)))];
            tensor<fp16, [768]> p_encoder_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12190528)))];
            tensor<fp16, []> layer_norm_4_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_4_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_4_cast_fp16 = layer_norm(axes = layer_norm_4_axes_0, beta = p_encoder_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_4_epsilon_0_to_fp16, gamma = p_encoder_layers_0_norm_feed_forward2_weight_to_fp16, x = add_8_cast_fp16)[name = tensor<string, []>("layer_norm_4_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_0_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12192128))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14551488))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_0_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14557696)))];
            tensor<fp16, [1, 750, 3072]> linear_6_cast_fp16 = linear(bias = p_encoder_layers_0_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_0_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_4_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_2_cast_fp16 = silu(x = linear_6_cast_fp16)[name = tensor<string, []>("silu_2_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_0_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_0_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14563904))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16923264))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_0_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16924864)))];
            tensor<fp16, [1, 750, 768]> linear_7_cast_fp16 = linear(bias = p_encoder_layers_0_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_0_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_2_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
            tensor<fp16, []> const_108_to_fp16 = const()[name = tensor<string, []>("const_108_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_5_cast_fp16 = mul(x = linear_7_cast_fp16, y = const_108_to_fp16)[name = tensor<string, []>("mul_5_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_9_cast_fp16 = add(x = add_8_cast_fp16, y = mul_5_cast_fp16)[name = tensor<string, []>("add_9_cast_fp16")];
            tensor<int32, [1]> layer_norm_5_axes_0 = const()[name = tensor<string, []>("layer_norm_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_0_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16926464)))];
            tensor<fp16, [768]> p_encoder_layers_0_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16928064)))];
            tensor<fp16, []> layer_norm_5_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_5_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_5_cast_fp16 = layer_norm(axes = layer_norm_5_axes_0, beta = p_encoder_layers_0_norm_out_bias_to_fp16, epsilon = layer_norm_5_epsilon_0_to_fp16, gamma = p_encoder_layers_0_norm_out_weight_to_fp16, x = add_9_cast_fp16)[name = tensor<string, []>("layer_norm_5_cast_fp16")];
            tensor<int32, [1]> layer_norm_6_axes_0 = const()[name = tensor<string, []>("layer_norm_6_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16929664)))];
            tensor<fp16, [768]> p_encoder_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16931264)))];
            tensor<fp16, []> layer_norm_6_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_6_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_6_cast_fp16 = layer_norm(axes = layer_norm_6_axes_0, beta = p_encoder_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_6_epsilon_0_to_fp16, gamma = p_encoder_layers_1_norm_feed_forward1_weight_to_fp16, x = layer_norm_5_cast_fp16)[name = tensor<string, []>("layer_norm_6_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_1_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16932864))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19292224))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_1_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19298432)))];
            tensor<fp16, [1, 750, 3072]> linear_8_cast_fp16 = linear(bias = p_encoder_layers_1_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_1_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_6_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_3_cast_fp16 = silu(x = linear_8_cast_fp16)[name = tensor<string, []>("silu_3_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_1_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19304640))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21664000))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_1_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21665600)))];
            tensor<fp16, [1, 750, 768]> linear_9_cast_fp16 = linear(bias = p_encoder_layers_1_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_1_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_3_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
            tensor<fp16, []> const_111_to_fp16 = const()[name = tensor<string, []>("const_111_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_6_cast_fp16 = mul(x = linear_9_cast_fp16, y = const_111_to_fp16)[name = tensor<string, []>("mul_6_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_10_cast_fp16 = add(x = layer_norm_5_cast_fp16, y = mul_6_cast_fp16)[name = tensor<string, []>("add_10_cast_fp16")];
            tensor<int32, [1]> layer_norm_7_axes_0 = const()[name = tensor<string, []>("layer_norm_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_1_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21667200)))];
            tensor<fp16, [768]> p_encoder_layers_1_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21668800)))];
            tensor<fp16, []> layer_norm_7_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_7_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_7_cast_fp16 = layer_norm(axes = layer_norm_7_axes_0, beta = p_encoder_layers_1_norm_self_att_bias_to_fp16, epsilon = layer_norm_7_epsilon_0_to_fp16, gamma = p_encoder_layers_1_norm_self_att_weight_to_fp16, x = add_10_cast_fp16)[name = tensor<string, []>("layer_norm_7_cast_fp16")];
            tensor<int32, [4]> const_115 = const()[name = tensor<string, []>("const_115"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_49_perm_1 = const()[name = tensor<string, []>("transpose_49_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_49 = transpose(perm = transpose_49_perm_1, x = layer_norm_7_cast_fp16)[name = tensor<string, []>("transpose_246")];
            tensor<fp16, [750, 1, 16, 48]> view_9_cast_fp16 = reshape(shape = const_115, x = transpose_49)[name = tensor<string, []>("view_9_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_7_cast_fp16 = mul(x = view_9_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_7_cast_fp16")];
            tensor<int32, [4]> slice_7_begin_0 = const()[name = tensor<string, []>("slice_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_7_end_0 = const()[name = tensor<string, []>("slice_7_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_7_end_mask_0 = const()[name = tensor<string, []>("slice_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_7_cast_fp16 = slice_by_index(begin = slice_7_begin_0, end = slice_7_end_0, end_mask = slice_7_end_mask_0, x = view_9_cast_fp16)[name = tensor<string, []>("slice_7_cast_fp16")];
            tensor<int32, [4]> slice_8_begin_0 = const()[name = tensor<string, []>("slice_8_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_8_end_0 = const()[name = tensor<string, []>("slice_8_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_8_end_mask_0 = const()[name = tensor<string, []>("slice_8_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_8_cast_fp16 = slice_by_index(begin = slice_8_begin_0, end = slice_8_end_0, end_mask = slice_8_end_mask_0, x = view_9_cast_fp16)[name = tensor<string, []>("slice_8_cast_fp16")];
            tensor<fp16, []> const_132_promoted_to_fp16 = const()[name = tensor<string, []>("const_132_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_2_cast_fp16 = mul(x = slice_8_cast_fp16, y = const_132_promoted_to_fp16)[name = tensor<string, []>("neg_2_cast_fp16")];
            tensor<int32, []> const_133 = const()[name = tensor<string, []>("const_133"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_2_interleave_0 = const()[name = tensor<string, []>("cat_2_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_2_cast_fp16 = concat(axis = const_133, interleave = cat_2_interleave_0, values = (neg_2_cast_fp16, slice_7_cast_fp16))[name = tensor<string, []>("cat_2_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_8_cast_fp16 = mul(x = cat_2_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_8_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_11_cast_fp16 = add(x = mul_7_cast_fp16, y = mul_8_cast_fp16)[name = tensor<string, []>("add_11_cast_fp16")];
            tensor<int32, [3]> const_142 = const()[name = tensor<string, []>("const_142"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_12_cast_fp16 = reshape(shape = const_142, x = add_11_cast_fp16)[name = tensor<string, []>("view_12_cast_fp16")];
            tensor<int32, [3]> transpose_21_perm_0 = const()[name = tensor<string, []>("transpose_21_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_1_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21670400))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22260288))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_1_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22261888)))];
            tensor<fp16, [1, 750, 768]> transpose_21_cast_fp16 = transpose(perm = transpose_21_perm_0, x = view_12_cast_fp16)[name = tensor<string, []>("transpose_245")];
            tensor<fp16, [1, 750, 768]> linear_10_cast_fp16 = linear(bias = p_encoder_layers_1_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_1_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_21_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
            tensor<int32, [4]> const_151 = const()[name = tensor<string, []>("const_151"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_15_cast_fp16 = reshape(shape = const_151, x = linear_10_cast_fp16)[name = tensor<string, []>("view_15_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_1_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22263488))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22853376))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_1_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22854976)))];
            tensor<fp16, [1, 750, 768]> linear_11_cast_fp16 = linear(bias = p_encoder_layers_1_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_1_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_21_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
            tensor<int32, [4]> const_152 = const()[name = tensor<string, []>("const_152"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_16_cast_fp16 = reshape(shape = const_152, x = linear_11_cast_fp16)[name = tensor<string, []>("view_16_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_1_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22856576))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23446464))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_1_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23448064)))];
            tensor<fp16, [1, 750, 768]> linear_12_cast_fp16 = linear(bias = p_encoder_layers_1_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_1_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_7_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<int32, [4]> const_153 = const()[name = tensor<string, []>("const_153"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_17_cast_fp16 = reshape(shape = const_153, x = linear_12_cast_fp16)[name = tensor<string, []>("view_17_cast_fp16")];
            tensor<int32, [4]> transpose_26_perm_0 = const()[name = tensor<string, []>("transpose_26_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_3_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_3_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_3_cast_fp16 = mul(x = view_16_cast_fp16, y = _inversed_div_3_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_3_cast_fp16")];
            tensor<bool, []> matmul_2_transpose_x_0 = const()[name = tensor<string, []>("matmul_2_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_2_transpose_y_0 = const()[name = tensor<string, []>("matmul_2_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_66_perm_0 = const()[name = tensor<string, []>("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_67_perm_0 = const()[name = tensor<string, []>("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_67 = transpose(perm = transpose_67_perm_0, x = _inversed_div_3_cast_fp16)[name = tensor<string, []>("transpose_243")];
            tensor<fp16, [1, 16, 750, 48]> transpose_66 = transpose(perm = transpose_66_perm_0, x = view_15_cast_fp16)[name = tensor<string, []>("transpose_244")];
            tensor<fp16, [1, 16, 750, 750]> matmul_2_cast_fp16 = matmul(transpose_x = matmul_2_transpose_x_0, transpose_y = matmul_2_transpose_y_0, x = transpose_66, y = transpose_67)[name = tensor<string, []>("matmul_2_cast_fp16")];
            tensor<int32, []> const_163 = const()[name = tensor<string, []>("const_163"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_1_cast_fp16 = softmax(axis = const_163, x = matmul_2_cast_fp16)[name = tensor<string, []>("softmax_1_cast_fp16")];
            tensor<bool, []> matmul_3_transpose_x_0 = const()[name = tensor<string, []>("matmul_3_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_3_transpose_y_0 = const()[name = tensor<string, []>("matmul_3_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_26_cast_fp16 = transpose(perm = transpose_26_perm_0, x = view_17_cast_fp16)[name = tensor<string, []>("transpose_242")];
            tensor<fp16, [1, 16, 750, 48]> matmul_3_cast_fp16 = matmul(transpose_x = matmul_3_transpose_x_0, transpose_y = matmul_3_transpose_y_0, x = softmax_1_cast_fp16, y = transpose_26_cast_fp16)[name = tensor<string, []>("matmul_3_cast_fp16")];
            tensor<int32, [4]> transpose_28_perm_0 = const()[name = tensor<string, []>("transpose_28_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_166 = const()[name = tensor<string, []>("const_166"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_28_cast_fp16 = transpose(perm = transpose_28_perm_0, x = matmul_3_cast_fp16)[name = tensor<string, []>("transpose_241")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_1_cast_fp16 = reshape(shape = const_166, x = transpose_28_cast_fp16)[name = tensor<string, []>("_unsafe_view_1_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_1_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23449664))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24039552))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_1_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24041152)))];
            tensor<fp16, [1, 750, 768]> linear_13_cast_fp16 = linear(bias = p_encoder_layers_1_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_1_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_1_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_13_cast_fp16 = add(x = add_10_cast_fp16, y = linear_13_cast_fp16)[name = tensor<string, []>("add_13_cast_fp16")];
            tensor<int32, [1]> layer_norm_8_axes_0 = const()[name = tensor<string, []>("layer_norm_8_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_1_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24042752)))];
            tensor<fp16, [768]> p_encoder_layers_1_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24044352)))];
            tensor<fp16, []> layer_norm_8_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_8_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_8_cast_fp16 = layer_norm(axes = layer_norm_8_axes_0, beta = p_encoder_layers_1_norm_conv_bias_to_fp16, epsilon = layer_norm_8_epsilon_0_to_fp16, gamma = p_encoder_layers_1_norm_conv_weight_to_fp16, x = add_13_cast_fp16)[name = tensor<string, []>("layer_norm_8_cast_fp16")];
            tensor<int32, [3]> transpose_29_perm_0 = const()[name = tensor<string, []>("transpose_29_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_5_pad_type_0 = const()[name = tensor<string, []>("conv1d_5_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_5_strides_0 = const()[name = tensor<string, []>("conv1d_5_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_5_pad_0 = const()[name = tensor<string, []>("conv1d_5_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_5_dilations_0 = const()[name = tensor<string, []>("conv1d_5_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_5_groups_0 = const()[name = tensor<string, []>("conv1d_5_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24045952))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25225664))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_1_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25228800)))];
            tensor<fp16, [1, 768, 750]> transpose_29_cast_fp16 = transpose(perm = transpose_29_perm_0, x = layer_norm_8_cast_fp16)[name = tensor<string, []>("transpose_240")];
            tensor<fp16, [1, 1536, 750]> conv1d_5_cast_fp16 = conv(bias = p_encoder_layers_1_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_5_dilations_0, groups = conv1d_5_groups_0, pad = conv1d_5_pad_0, pad_type = conv1d_5_pad_type_0, strides = conv1d_5_strides_0, weight = p_encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_29_cast_fp16)[name = tensor<string, []>("conv1d_5_cast_fp16")];
            tensor<int32, []> glu_1_split_num_splits_0 = const()[name = tensor<string, []>("glu_1_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_1_split_axis_0 = const()[name = tensor<string, []>("glu_1_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_1_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_1_split_cast_fp16_1 = split(axis = glu_1_split_axis_0, num_splits = glu_1_split_num_splits_0, x = conv1d_5_cast_fp16)[name = tensor<string, []>("glu_1_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_1_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_1_split_cast_fp16_1)[name = tensor<string, []>("glu_1_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_1_cast_fp16 = mul(x = glu_1_split_cast_fp16_0, y = glu_1_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_1_cast_fp16")];
            tensor<string, []> conv1d_6_pad_type_0 = const()[name = tensor<string, []>("conv1d_6_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_6_pad_0 = const()[name = tensor<string, []>("conv1d_6_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_6_groups_0 = const()[name = tensor<string, []>("conv1d_6_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_6_strides_0 = const()[name = tensor<string, []>("conv1d_6_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_6_dilations_0 = const()[name = tensor<string, []>("conv1d_6_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_1_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25231936))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25235840))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_1_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25237440)))];
            tensor<fp16, [1, 768, 750]> conv1d_6_cast_fp16 = conv(bias = p_encoder_layers_1_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_6_dilations_0, groups = conv1d_6_groups_0, pad = conv1d_6_pad_0, pad_type = conv1d_6_pad_type_0, strides = conv1d_6_strides_0, weight = p_encoder_layers_1_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_1_cast_fp16)[name = tensor<string, []>("conv1d_6_cast_fp16")];
            tensor<int32, [3]> transpose_30_perm_0 = const()[name = tensor<string, []>("transpose_30_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_9_axes_0 = const()[name = tensor<string, []>("layer_norm_9_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_1_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25239040)))];
            tensor<fp16, [768]> p_encoder_layers_1_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25240640)))];
            tensor<fp16, []> layer_norm_9_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_9_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_30_cast_fp16 = transpose(perm = transpose_30_perm_0, x = conv1d_6_cast_fp16)[name = tensor<string, []>("transpose_239")];
            tensor<fp16, [1, 750, 768]> layer_norm_9_cast_fp16 = layer_norm(axes = layer_norm_9_axes_0, beta = p_encoder_layers_1_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_9_epsilon_0_to_fp16, gamma = p_encoder_layers_1_conv_batch_norm_weight_to_fp16, x = transpose_30_cast_fp16)[name = tensor<string, []>("layer_norm_9_cast_fp16")];
            tensor<int32, [3]> transpose_31_perm_0 = const()[name = tensor<string, []>("transpose_31_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_31_cast_fp16 = transpose(perm = transpose_31_perm_0, x = layer_norm_9_cast_fp16)[name = tensor<string, []>("transpose_238")];
            tensor<fp16, [1, 768, 750]> silu_4_cast_fp16 = silu(x = transpose_31_cast_fp16)[name = tensor<string, []>("silu_4_cast_fp16")];
            tensor<string, []> conv1d_7_pad_type_0 = const()[name = tensor<string, []>("conv1d_7_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_7_strides_0 = const()[name = tensor<string, []>("conv1d_7_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_7_pad_0 = const()[name = tensor<string, []>("conv1d_7_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_7_dilations_0 = const()[name = tensor<string, []>("conv1d_7_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_7_groups_0 = const()[name = tensor<string, []>("conv1d_7_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25242240))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25832128))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_1_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25833728)))];
            tensor<fp16, [1, 768, 750]> conv1d_7_cast_fp16 = conv(bias = p_encoder_layers_1_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_7_dilations_0, groups = conv1d_7_groups_0, pad = conv1d_7_pad_0, pad_type = conv1d_7_pad_type_0, strides = conv1d_7_strides_0, weight = p_encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_4_cast_fp16)[name = tensor<string, []>("conv1d_7_cast_fp16")];
            tensor<int32, [3]> transpose_32_perm_0_1 = const()[name = tensor<string, []>("transpose_32_perm_0_1"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_32_cast_fp16 = transpose(perm = transpose_32_perm_0_1, x = conv1d_7_cast_fp16)[name = tensor<string, []>("transpose_237")];
            tensor<fp16, [1, 750, 768]> add_14_cast_fp16 = add(x = add_13_cast_fp16, y = transpose_32_cast_fp16)[name = tensor<string, []>("add_14_cast_fp16")];
            tensor<int32, [1]> layer_norm_10_axes_0 = const()[name = tensor<string, []>("layer_norm_10_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25835328)))];
            tensor<fp16, [768]> p_encoder_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25836928)))];
            tensor<fp16, []> layer_norm_10_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_10_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_10_cast_fp16 = layer_norm(axes = layer_norm_10_axes_0, beta = p_encoder_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_10_epsilon_0_to_fp16, gamma = p_encoder_layers_1_norm_feed_forward2_weight_to_fp16, x = add_14_cast_fp16)[name = tensor<string, []>("layer_norm_10_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_1_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25838528))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28197888))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_1_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28204096)))];
            tensor<fp16, [1, 750, 3072]> linear_14_cast_fp16 = linear(bias = p_encoder_layers_1_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_1_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_10_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_5_cast_fp16 = silu(x = linear_14_cast_fp16)[name = tensor<string, []>("silu_5_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_1_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_1_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28210304))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30569664))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_1_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30571264)))];
            tensor<fp16, [1, 750, 768]> linear_15_cast_fp16 = linear(bias = p_encoder_layers_1_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_1_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_5_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")];
            tensor<fp16, []> const_185_to_fp16 = const()[name = tensor<string, []>("const_185_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_11_cast_fp16 = mul(x = linear_15_cast_fp16, y = const_185_to_fp16)[name = tensor<string, []>("mul_11_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_15_cast_fp16 = add(x = add_14_cast_fp16, y = mul_11_cast_fp16)[name = tensor<string, []>("add_15_cast_fp16")];
            tensor<int32, [1]> layer_norm_11_axes_0 = const()[name = tensor<string, []>("layer_norm_11_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_1_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30572864)))];
            tensor<fp16, [768]> p_encoder_layers_1_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30574464)))];
            tensor<fp16, []> layer_norm_11_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_11_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_11_cast_fp16 = layer_norm(axes = layer_norm_11_axes_0, beta = p_encoder_layers_1_norm_out_bias_to_fp16, epsilon = layer_norm_11_epsilon_0_to_fp16, gamma = p_encoder_layers_1_norm_out_weight_to_fp16, x = add_15_cast_fp16)[name = tensor<string, []>("layer_norm_11_cast_fp16")];
            tensor<int32, [1]> layer_norm_12_axes_0 = const()[name = tensor<string, []>("layer_norm_12_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30576064)))];
            tensor<fp16, [768]> p_encoder_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30577664)))];
            tensor<fp16, []> layer_norm_12_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_12_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_12_cast_fp16 = layer_norm(axes = layer_norm_12_axes_0, beta = p_encoder_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_12_epsilon_0_to_fp16, gamma = p_encoder_layers_2_norm_feed_forward1_weight_to_fp16, x = layer_norm_11_cast_fp16)[name = tensor<string, []>("layer_norm_12_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_2_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30579264))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32938624))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_2_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32944832)))];
            tensor<fp16, [1, 750, 3072]> linear_16_cast_fp16 = linear(bias = p_encoder_layers_2_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_2_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_12_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_6_cast_fp16 = silu(x = linear_16_cast_fp16)[name = tensor<string, []>("silu_6_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_2_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32951040))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35310400))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_2_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35312000)))];
            tensor<fp16, [1, 750, 768]> linear_17_cast_fp16 = linear(bias = p_encoder_layers_2_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_2_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_6_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
            tensor<fp16, []> const_188_to_fp16 = const()[name = tensor<string, []>("const_188_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_12_cast_fp16 = mul(x = linear_17_cast_fp16, y = const_188_to_fp16)[name = tensor<string, []>("mul_12_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_16_cast_fp16 = add(x = layer_norm_11_cast_fp16, y = mul_12_cast_fp16)[name = tensor<string, []>("add_16_cast_fp16")];
            tensor<int32, [1]> layer_norm_13_axes_0 = const()[name = tensor<string, []>("layer_norm_13_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_2_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35313600)))];
            tensor<fp16, [768]> p_encoder_layers_2_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35315200)))];
            tensor<fp16, []> layer_norm_13_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_13_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_13_cast_fp16 = layer_norm(axes = layer_norm_13_axes_0, beta = p_encoder_layers_2_norm_self_att_bias_to_fp16, epsilon = layer_norm_13_epsilon_0_to_fp16, gamma = p_encoder_layers_2_norm_self_att_weight_to_fp16, x = add_16_cast_fp16)[name = tensor<string, []>("layer_norm_13_cast_fp16")];
            tensor<int32, [4]> const_192 = const()[name = tensor<string, []>("const_192"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_50_perm_1 = const()[name = tensor<string, []>("transpose_50_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_50 = transpose(perm = transpose_50_perm_1, x = layer_norm_13_cast_fp16)[name = tensor<string, []>("transpose_236")];
            tensor<fp16, [750, 1, 16, 48]> view_18_cast_fp16 = reshape(shape = const_192, x = transpose_50)[name = tensor<string, []>("view_18_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_13_cast_fp16 = mul(x = view_18_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_13_cast_fp16")];
            tensor<int32, [4]> slice_11_begin_0 = const()[name = tensor<string, []>("slice_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_11_end_0 = const()[name = tensor<string, []>("slice_11_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_11_end_mask_0 = const()[name = tensor<string, []>("slice_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_11_cast_fp16 = slice_by_index(begin = slice_11_begin_0, end = slice_11_end_0, end_mask = slice_11_end_mask_0, x = view_18_cast_fp16)[name = tensor<string, []>("slice_11_cast_fp16")];
            tensor<int32, [4]> slice_12_begin_0 = const()[name = tensor<string, []>("slice_12_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_12_end_0 = const()[name = tensor<string, []>("slice_12_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_12_end_mask_0 = const()[name = tensor<string, []>("slice_12_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_12_cast_fp16 = slice_by_index(begin = slice_12_begin_0, end = slice_12_end_0, end_mask = slice_12_end_mask_0, x = view_18_cast_fp16)[name = tensor<string, []>("slice_12_cast_fp16")];
            tensor<fp16, []> const_209_promoted_to_fp16 = const()[name = tensor<string, []>("const_209_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_4_cast_fp16 = mul(x = slice_12_cast_fp16, y = const_209_promoted_to_fp16)[name = tensor<string, []>("neg_4_cast_fp16")];
            tensor<int32, []> const_210 = const()[name = tensor<string, []>("const_210"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_4_interleave_0 = const()[name = tensor<string, []>("cat_4_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_4_cast_fp16 = concat(axis = const_210, interleave = cat_4_interleave_0, values = (neg_4_cast_fp16, slice_11_cast_fp16))[name = tensor<string, []>("cat_4_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_14_cast_fp16 = mul(x = cat_4_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_14_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_17_cast_fp16 = add(x = mul_13_cast_fp16, y = mul_14_cast_fp16)[name = tensor<string, []>("add_17_cast_fp16")];
            tensor<int32, [3]> const_219 = const()[name = tensor<string, []>("const_219"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_21_cast_fp16 = reshape(shape = const_219, x = add_17_cast_fp16)[name = tensor<string, []>("view_21_cast_fp16")];
            tensor<int32, [3]> transpose_36_perm_0 = const()[name = tensor<string, []>("transpose_36_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_2_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35316800))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35906688))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_2_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35908288)))];
            tensor<fp16, [1, 750, 768]> transpose_36_cast_fp16 = transpose(perm = transpose_36_perm_0, x = view_21_cast_fp16)[name = tensor<string, []>("transpose_235")];
            tensor<fp16, [1, 750, 768]> linear_18_cast_fp16 = linear(bias = p_encoder_layers_2_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_2_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_36_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<int32, [4]> const_228 = const()[name = tensor<string, []>("const_228"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_24_cast_fp16 = reshape(shape = const_228, x = linear_18_cast_fp16)[name = tensor<string, []>("view_24_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_2_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35909888))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36499776))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_2_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36501376)))];
            tensor<fp16, [1, 750, 768]> linear_19_cast_fp16 = linear(bias = p_encoder_layers_2_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_2_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_36_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<int32, [4]> const_229 = const()[name = tensor<string, []>("const_229"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_25_cast_fp16 = reshape(shape = const_229, x = linear_19_cast_fp16)[name = tensor<string, []>("view_25_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_2_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36502976))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37092864))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_2_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37094464)))];
            tensor<fp16, [1, 750, 768]> linear_20_cast_fp16 = linear(bias = p_encoder_layers_2_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_2_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_13_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<int32, [4]> const_230 = const()[name = tensor<string, []>("const_230"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_26_cast_fp16 = reshape(shape = const_230, x = linear_20_cast_fp16)[name = tensor<string, []>("view_26_cast_fp16")];
            tensor<int32, [4]> transpose_41_perm_0 = const()[name = tensor<string, []>("transpose_41_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_4_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_4_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_4_cast_fp16 = mul(x = view_25_cast_fp16, y = _inversed_div_4_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_4_cast_fp16")];
            tensor<bool, []> matmul_4_transpose_x_0 = const()[name = tensor<string, []>("matmul_4_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_4_transpose_y_0 = const()[name = tensor<string, []>("matmul_4_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_68_perm_0 = const()[name = tensor<string, []>("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_69_perm_0 = const()[name = tensor<string, []>("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_69 = transpose(perm = transpose_69_perm_0, x = _inversed_div_4_cast_fp16)[name = tensor<string, []>("transpose_233")];
            tensor<fp16, [1, 16, 750, 48]> transpose_68 = transpose(perm = transpose_68_perm_0, x = view_24_cast_fp16)[name = tensor<string, []>("transpose_234")];
            tensor<fp16, [1, 16, 750, 750]> matmul_4_cast_fp16 = matmul(transpose_x = matmul_4_transpose_x_0, transpose_y = matmul_4_transpose_y_0, x = transpose_68, y = transpose_69)[name = tensor<string, []>("matmul_4_cast_fp16")];
            tensor<int32, []> const_240 = const()[name = tensor<string, []>("const_240"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_2_cast_fp16 = softmax(axis = const_240, x = matmul_4_cast_fp16)[name = tensor<string, []>("softmax_2_cast_fp16")];
            tensor<bool, []> matmul_5_transpose_x_0 = const()[name = tensor<string, []>("matmul_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_5_transpose_y_0 = const()[name = tensor<string, []>("matmul_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_41_cast_fp16 = transpose(perm = transpose_41_perm_0, x = view_26_cast_fp16)[name = tensor<string, []>("transpose_232")];
            tensor<fp16, [1, 16, 750, 48]> matmul_5_cast_fp16 = matmul(transpose_x = matmul_5_transpose_x_0, transpose_y = matmul_5_transpose_y_0, x = softmax_2_cast_fp16, y = transpose_41_cast_fp16)[name = tensor<string, []>("matmul_5_cast_fp16")];
            tensor<int32, [4]> transpose_43_perm_0 = const()[name = tensor<string, []>("transpose_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_243 = const()[name = tensor<string, []>("const_243"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_43_cast_fp16 = transpose(perm = transpose_43_perm_0, x = matmul_5_cast_fp16)[name = tensor<string, []>("transpose_231")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_2_cast_fp16 = reshape(shape = const_243, x = transpose_43_cast_fp16)[name = tensor<string, []>("_unsafe_view_2_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_2_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37096064))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37685952))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_2_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37687552)))];
            tensor<fp16, [1, 750, 768]> linear_21_cast_fp16 = linear(bias = p_encoder_layers_2_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_2_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_2_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_19_cast_fp16 = add(x = add_16_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("add_19_cast_fp16")];
            tensor<int32, [1]> layer_norm_14_axes_0 = const()[name = tensor<string, []>("layer_norm_14_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_2_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37689152)))];
            tensor<fp16, [768]> p_encoder_layers_2_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37690752)))];
            tensor<fp16, []> layer_norm_14_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_14_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_14_cast_fp16 = layer_norm(axes = layer_norm_14_axes_0, beta = p_encoder_layers_2_norm_conv_bias_to_fp16, epsilon = layer_norm_14_epsilon_0_to_fp16, gamma = p_encoder_layers_2_norm_conv_weight_to_fp16, x = add_19_cast_fp16)[name = tensor<string, []>("layer_norm_14_cast_fp16")];
            tensor<int32, [3]> transpose_44_perm_0 = const()[name = tensor<string, []>("transpose_44_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_8_pad_type_0 = const()[name = tensor<string, []>("conv1d_8_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_8_strides_0 = const()[name = tensor<string, []>("conv1d_8_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_8_pad_0 = const()[name = tensor<string, []>("conv1d_8_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_8_dilations_0 = const()[name = tensor<string, []>("conv1d_8_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_8_groups_0 = const()[name = tensor<string, []>("conv1d_8_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37692352))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38872064))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_2_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38875200)))];
            tensor<fp16, [1, 768, 750]> transpose_44_cast_fp16 = transpose(perm = transpose_44_perm_0, x = layer_norm_14_cast_fp16)[name = tensor<string, []>("transpose_230")];
            tensor<fp16, [1, 1536, 750]> conv1d_8_cast_fp16 = conv(bias = p_encoder_layers_2_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_8_dilations_0, groups = conv1d_8_groups_0, pad = conv1d_8_pad_0, pad_type = conv1d_8_pad_type_0, strides = conv1d_8_strides_0, weight = p_encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_44_cast_fp16)[name = tensor<string, []>("conv1d_8_cast_fp16")];
            tensor<int32, []> glu_2_split_num_splits_0 = const()[name = tensor<string, []>("glu_2_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_2_split_axis_0 = const()[name = tensor<string, []>("glu_2_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_2_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_2_split_cast_fp16_1 = split(axis = glu_2_split_axis_0, num_splits = glu_2_split_num_splits_0, x = conv1d_8_cast_fp16)[name = tensor<string, []>("glu_2_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_2_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_2_split_cast_fp16_1)[name = tensor<string, []>("glu_2_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_2_cast_fp16 = mul(x = glu_2_split_cast_fp16_0, y = glu_2_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_2_cast_fp16")];
            tensor<string, []> conv1d_9_pad_type_0 = const()[name = tensor<string, []>("conv1d_9_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_9_pad_0 = const()[name = tensor<string, []>("conv1d_9_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_9_groups_0 = const()[name = tensor<string, []>("conv1d_9_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_9_strides_0 = const()[name = tensor<string, []>("conv1d_9_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_9_dilations_0 = const()[name = tensor<string, []>("conv1d_9_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_2_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38878336))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38882240))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_2_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38883840)))];
            tensor<fp16, [1, 768, 750]> conv1d_9_cast_fp16 = conv(bias = p_encoder_layers_2_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_9_dilations_0, groups = conv1d_9_groups_0, pad = conv1d_9_pad_0, pad_type = conv1d_9_pad_type_0, strides = conv1d_9_strides_0, weight = p_encoder_layers_2_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_2_cast_fp16)[name = tensor<string, []>("conv1d_9_cast_fp16")];
            tensor<int32, [3]> transpose_45_perm_0 = const()[name = tensor<string, []>("transpose_45_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_15_axes_0 = const()[name = tensor<string, []>("layer_norm_15_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_2_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38885440)))];
            tensor<fp16, [768]> p_encoder_layers_2_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38887040)))];
            tensor<fp16, []> layer_norm_15_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_15_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_45_cast_fp16 = transpose(perm = transpose_45_perm_0, x = conv1d_9_cast_fp16)[name = tensor<string, []>("transpose_229")];
            tensor<fp16, [1, 750, 768]> layer_norm_15_cast_fp16 = layer_norm(axes = layer_norm_15_axes_0, beta = p_encoder_layers_2_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_15_epsilon_0_to_fp16, gamma = p_encoder_layers_2_conv_batch_norm_weight_to_fp16, x = transpose_45_cast_fp16)[name = tensor<string, []>("layer_norm_15_cast_fp16")];
            tensor<int32, [3]> transpose_46_perm_0 = const()[name = tensor<string, []>("transpose_46_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_46_cast_fp16 = transpose(perm = transpose_46_perm_0, x = layer_norm_15_cast_fp16)[name = tensor<string, []>("transpose_228")];
            tensor<fp16, [1, 768, 750]> silu_7_cast_fp16 = silu(x = transpose_46_cast_fp16)[name = tensor<string, []>("silu_7_cast_fp16")];
            tensor<string, []> conv1d_10_pad_type_0 = const()[name = tensor<string, []>("conv1d_10_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_10_strides_0 = const()[name = tensor<string, []>("conv1d_10_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_10_pad_0 = const()[name = tensor<string, []>("conv1d_10_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_10_dilations_0 = const()[name = tensor<string, []>("conv1d_10_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_10_groups_0 = const()[name = tensor<string, []>("conv1d_10_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38888640))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39478528))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_2_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39480128)))];
            tensor<fp16, [1, 768, 750]> conv1d_10_cast_fp16 = conv(bias = p_encoder_layers_2_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_10_dilations_0, groups = conv1d_10_groups_0, pad = conv1d_10_pad_0, pad_type = conv1d_10_pad_type_0, strides = conv1d_10_strides_0, weight = p_encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_7_cast_fp16)[name = tensor<string, []>("conv1d_10_cast_fp16")];
            tensor<int32, [3]> transpose_47_perm_0 = const()[name = tensor<string, []>("transpose_47_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_47_cast_fp16 = transpose(perm = transpose_47_perm_0, x = conv1d_10_cast_fp16)[name = tensor<string, []>("transpose_227")];
            tensor<fp16, [1, 750, 768]> add_20_cast_fp16 = add(x = add_19_cast_fp16, y = transpose_47_cast_fp16)[name = tensor<string, []>("add_20_cast_fp16")];
            tensor<int32, [1]> layer_norm_16_axes_0 = const()[name = tensor<string, []>("layer_norm_16_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39481728)))];
            tensor<fp16, [768]> p_encoder_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39483328)))];
            tensor<fp16, []> layer_norm_16_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_16_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_16_cast_fp16 = layer_norm(axes = layer_norm_16_axes_0, beta = p_encoder_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_16_epsilon_0_to_fp16, gamma = p_encoder_layers_2_norm_feed_forward2_weight_to_fp16, x = add_20_cast_fp16)[name = tensor<string, []>("layer_norm_16_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_2_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39484928))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41844288))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_2_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41850496)))];
            tensor<fp16, [1, 750, 3072]> linear_22_cast_fp16 = linear(bias = p_encoder_layers_2_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_2_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_16_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_8_cast_fp16 = silu(x = linear_22_cast_fp16)[name = tensor<string, []>("silu_8_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_2_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_2_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41856704))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44216064))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_2_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44217664)))];
            tensor<fp16, [1, 750, 768]> linear_23_cast_fp16 = linear(bias = p_encoder_layers_2_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_2_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_8_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
            tensor<fp16, []> const_262_to_fp16 = const()[name = tensor<string, []>("const_262_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_17_cast_fp16 = mul(x = linear_23_cast_fp16, y = const_262_to_fp16)[name = tensor<string, []>("mul_17_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_21_cast_fp16 = add(x = add_20_cast_fp16, y = mul_17_cast_fp16)[name = tensor<string, []>("add_21_cast_fp16")];
            tensor<int32, [1]> layer_norm_17_axes_0 = const()[name = tensor<string, []>("layer_norm_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_2_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44219264)))];
            tensor<fp16, [768]> p_encoder_layers_2_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44220864)))];
            tensor<fp16, []> layer_norm_17_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_17_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_17_cast_fp16 = layer_norm(axes = layer_norm_17_axes_0, beta = p_encoder_layers_2_norm_out_bias_to_fp16, epsilon = layer_norm_17_epsilon_0_to_fp16, gamma = p_encoder_layers_2_norm_out_weight_to_fp16, x = add_21_cast_fp16)[name = tensor<string, []>("layer_norm_17_cast_fp16")];
            tensor<int32, [1]> layer_norm_18_axes_0 = const()[name = tensor<string, []>("layer_norm_18_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44222464)))];
            tensor<fp16, [768]> p_encoder_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44224064)))];
            tensor<fp16, []> layer_norm_18_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_18_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_18_cast_fp16 = layer_norm(axes = layer_norm_18_axes_0, beta = p_encoder_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_18_epsilon_0_to_fp16, gamma = p_encoder_layers_3_norm_feed_forward1_weight_to_fp16, x = layer_norm_17_cast_fp16)[name = tensor<string, []>("layer_norm_18_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_3_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44225664))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46585024))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_3_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46591232)))];
            tensor<fp16, [1, 750, 3072]> linear_24_cast_fp16 = linear(bias = p_encoder_layers_3_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_3_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_18_cast_fp16)[name = tensor<string, []>("linear_24_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_9_cast_fp16 = silu(x = linear_24_cast_fp16)[name = tensor<string, []>("silu_9_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_3_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46597440))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48956800))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_3_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48958400)))];
            tensor<fp16, [1, 750, 768]> linear_25_cast_fp16 = linear(bias = p_encoder_layers_3_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_3_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_9_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<fp16, []> const_265_to_fp16 = const()[name = tensor<string, []>("const_265_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_18_cast_fp16 = mul(x = linear_25_cast_fp16, y = const_265_to_fp16)[name = tensor<string, []>("mul_18_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_22_cast_fp16 = add(x = layer_norm_17_cast_fp16, y = mul_18_cast_fp16)[name = tensor<string, []>("add_22_cast_fp16")];
            tensor<int32, [1]> layer_norm_19_axes_0 = const()[name = tensor<string, []>("layer_norm_19_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_3_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48960000)))];
            tensor<fp16, [768]> p_encoder_layers_3_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48961600)))];
            tensor<fp16, []> layer_norm_19_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_19_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_19_cast_fp16 = layer_norm(axes = layer_norm_19_axes_0, beta = p_encoder_layers_3_norm_self_att_bias_to_fp16, epsilon = layer_norm_19_epsilon_0_to_fp16, gamma = p_encoder_layers_3_norm_self_att_weight_to_fp16, x = add_22_cast_fp16)[name = tensor<string, []>("layer_norm_19_cast_fp16")];
            tensor<int32, [4]> const_269 = const()[name = tensor<string, []>("const_269"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_51_perm_1 = const()[name = tensor<string, []>("transpose_51_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_51 = transpose(perm = transpose_51_perm_1, x = layer_norm_19_cast_fp16)[name = tensor<string, []>("transpose_226")];
            tensor<fp16, [750, 1, 16, 48]> view_27_cast_fp16 = reshape(shape = const_269, x = transpose_51)[name = tensor<string, []>("view_27_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_19_cast_fp16 = mul(x = view_27_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_19_cast_fp16")];
            tensor<int32, [4]> slice_15_begin_0 = const()[name = tensor<string, []>("slice_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_15_end_0 = const()[name = tensor<string, []>("slice_15_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_15_end_mask_0 = const()[name = tensor<string, []>("slice_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_15_cast_fp16 = slice_by_index(begin = slice_15_begin_0, end = slice_15_end_0, end_mask = slice_15_end_mask_0, x = view_27_cast_fp16)[name = tensor<string, []>("slice_15_cast_fp16")];
            tensor<int32, [4]> slice_16_begin_0 = const()[name = tensor<string, []>("slice_16_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_16_end_0 = const()[name = tensor<string, []>("slice_16_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_16_end_mask_0 = const()[name = tensor<string, []>("slice_16_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_16_cast_fp16 = slice_by_index(begin = slice_16_begin_0, end = slice_16_end_0, end_mask = slice_16_end_mask_0, x = view_27_cast_fp16)[name = tensor<string, []>("slice_16_cast_fp16")];
            tensor<fp16, []> const_286_promoted_to_fp16 = const()[name = tensor<string, []>("const_286_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_6_cast_fp16 = mul(x = slice_16_cast_fp16, y = const_286_promoted_to_fp16)[name = tensor<string, []>("neg_6_cast_fp16")];
            tensor<int32, []> const_287 = const()[name = tensor<string, []>("const_287"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_6_interleave_0 = const()[name = tensor<string, []>("cat_6_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_6_cast_fp16 = concat(axis = const_287, interleave = cat_6_interleave_0, values = (neg_6_cast_fp16, slice_15_cast_fp16))[name = tensor<string, []>("cat_6_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_20_cast_fp16 = mul(x = cat_6_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_20_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_23_cast_fp16 = add(x = mul_19_cast_fp16, y = mul_20_cast_fp16)[name = tensor<string, []>("add_23_cast_fp16")];
            tensor<int32, [3]> const_296 = const()[name = tensor<string, []>("const_296"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_30_cast_fp16 = reshape(shape = const_296, x = add_23_cast_fp16)[name = tensor<string, []>("view_30_cast_fp16")];
            tensor<int32, [3]> transpose_51_perm_0 = const()[name = tensor<string, []>("transpose_51_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_3_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48963200))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49553088))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_3_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49554688)))];
            tensor<fp16, [1, 750, 768]> transpose_51_cast_fp16 = transpose(perm = transpose_51_perm_0, x = view_30_cast_fp16)[name = tensor<string, []>("transpose_225")];
            tensor<fp16, [1, 750, 768]> linear_26_cast_fp16 = linear(bias = p_encoder_layers_3_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_3_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_51_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<int32, [4]> const_305 = const()[name = tensor<string, []>("const_305"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_33_cast_fp16 = reshape(shape = const_305, x = linear_26_cast_fp16)[name = tensor<string, []>("view_33_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_3_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49556288))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50146176))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_3_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50147776)))];
            tensor<fp16, [1, 750, 768]> linear_27_cast_fp16 = linear(bias = p_encoder_layers_3_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_3_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_51_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
            tensor<int32, [4]> const_306 = const()[name = tensor<string, []>("const_306"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_34_cast_fp16 = reshape(shape = const_306, x = linear_27_cast_fp16)[name = tensor<string, []>("view_34_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_3_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50149376))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50739264))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_3_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50740864)))];
            tensor<fp16, [1, 750, 768]> linear_28_cast_fp16 = linear(bias = p_encoder_layers_3_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_3_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_19_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
            tensor<int32, [4]> const_307 = const()[name = tensor<string, []>("const_307"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_35_cast_fp16 = reshape(shape = const_307, x = linear_28_cast_fp16)[name = tensor<string, []>("view_35_cast_fp16")];
            tensor<int32, [4]> transpose_56_perm_0 = const()[name = tensor<string, []>("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_5_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_5_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_5_cast_fp16 = mul(x = view_34_cast_fp16, y = _inversed_div_5_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_5_cast_fp16")];
            tensor<bool, []> matmul_6_transpose_x_0 = const()[name = tensor<string, []>("matmul_6_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_6_transpose_y_0 = const()[name = tensor<string, []>("matmul_6_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_70_perm_0 = const()[name = tensor<string, []>("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_71_perm_0 = const()[name = tensor<string, []>("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_71 = transpose(perm = transpose_71_perm_0, x = _inversed_div_5_cast_fp16)[name = tensor<string, []>("transpose_223")];
            tensor<fp16, [1, 16, 750, 48]> transpose_70 = transpose(perm = transpose_70_perm_0, x = view_33_cast_fp16)[name = tensor<string, []>("transpose_224")];
            tensor<fp16, [1, 16, 750, 750]> matmul_6_cast_fp16 = matmul(transpose_x = matmul_6_transpose_x_0, transpose_y = matmul_6_transpose_y_0, x = transpose_70, y = transpose_71)[name = tensor<string, []>("matmul_6_cast_fp16")];
            tensor<int32, []> const_317 = const()[name = tensor<string, []>("const_317"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_3_cast_fp16 = softmax(axis = const_317, x = matmul_6_cast_fp16)[name = tensor<string, []>("softmax_3_cast_fp16")];
            tensor<bool, []> matmul_7_transpose_x_0 = const()[name = tensor<string, []>("matmul_7_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_7_transpose_y_0 = const()[name = tensor<string, []>("matmul_7_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_56_cast_fp16 = transpose(perm = transpose_56_perm_0, x = view_35_cast_fp16)[name = tensor<string, []>("transpose_222")];
            tensor<fp16, [1, 16, 750, 48]> matmul_7_cast_fp16 = matmul(transpose_x = matmul_7_transpose_x_0, transpose_y = matmul_7_transpose_y_0, x = softmax_3_cast_fp16, y = transpose_56_cast_fp16)[name = tensor<string, []>("matmul_7_cast_fp16")];
            tensor<int32, [4]> transpose_58_perm_0 = const()[name = tensor<string, []>("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_320 = const()[name = tensor<string, []>("const_320"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_58_cast_fp16 = transpose(perm = transpose_58_perm_0, x = matmul_7_cast_fp16)[name = tensor<string, []>("transpose_221")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_3_cast_fp16 = reshape(shape = const_320, x = transpose_58_cast_fp16)[name = tensor<string, []>("_unsafe_view_3_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_3_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50742464))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51332352))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_3_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51333952)))];
            tensor<fp16, [1, 750, 768]> linear_29_cast_fp16 = linear(bias = p_encoder_layers_3_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_3_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_3_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_25_cast_fp16 = add(x = add_22_cast_fp16, y = linear_29_cast_fp16)[name = tensor<string, []>("add_25_cast_fp16")];
            tensor<int32, [1]> layer_norm_20_axes_0 = const()[name = tensor<string, []>("layer_norm_20_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_3_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51335552)))];
            tensor<fp16, [768]> p_encoder_layers_3_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51337152)))];
            tensor<fp16, []> layer_norm_20_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_20_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_20_cast_fp16 = layer_norm(axes = layer_norm_20_axes_0, beta = p_encoder_layers_3_norm_conv_bias_to_fp16, epsilon = layer_norm_20_epsilon_0_to_fp16, gamma = p_encoder_layers_3_norm_conv_weight_to_fp16, x = add_25_cast_fp16)[name = tensor<string, []>("layer_norm_20_cast_fp16")];
            tensor<int32, [3]> transpose_59_perm_0 = const()[name = tensor<string, []>("transpose_59_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_11_pad_type_0 = const()[name = tensor<string, []>("conv1d_11_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_11_strides_0 = const()[name = tensor<string, []>("conv1d_11_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_11_pad_0 = const()[name = tensor<string, []>("conv1d_11_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_11_dilations_0 = const()[name = tensor<string, []>("conv1d_11_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_11_groups_0 = const()[name = tensor<string, []>("conv1d_11_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51338752))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52518464))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_3_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52521600)))];
            tensor<fp16, [1, 768, 750]> transpose_59_cast_fp16 = transpose(perm = transpose_59_perm_0, x = layer_norm_20_cast_fp16)[name = tensor<string, []>("transpose_220")];
            tensor<fp16, [1, 1536, 750]> conv1d_11_cast_fp16 = conv(bias = p_encoder_layers_3_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_11_dilations_0, groups = conv1d_11_groups_0, pad = conv1d_11_pad_0, pad_type = conv1d_11_pad_type_0, strides = conv1d_11_strides_0, weight = p_encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_59_cast_fp16)[name = tensor<string, []>("conv1d_11_cast_fp16")];
            tensor<int32, []> glu_3_split_num_splits_0 = const()[name = tensor<string, []>("glu_3_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_3_split_axis_0 = const()[name = tensor<string, []>("glu_3_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_3_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_3_split_cast_fp16_1 = split(axis = glu_3_split_axis_0, num_splits = glu_3_split_num_splits_0, x = conv1d_11_cast_fp16)[name = tensor<string, []>("glu_3_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_3_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_3_split_cast_fp16_1)[name = tensor<string, []>("glu_3_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_3_cast_fp16 = mul(x = glu_3_split_cast_fp16_0, y = glu_3_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_3_cast_fp16")];
            tensor<string, []> conv1d_12_pad_type_0 = const()[name = tensor<string, []>("conv1d_12_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_12_pad_0 = const()[name = tensor<string, []>("conv1d_12_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_12_groups_0 = const()[name = tensor<string, []>("conv1d_12_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_12_strides_0 = const()[name = tensor<string, []>("conv1d_12_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_12_dilations_0 = const()[name = tensor<string, []>("conv1d_12_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_3_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52524736))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52528640))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_3_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52530240)))];
            tensor<fp16, [1, 768, 750]> conv1d_12_cast_fp16 = conv(bias = p_encoder_layers_3_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_12_dilations_0, groups = conv1d_12_groups_0, pad = conv1d_12_pad_0, pad_type = conv1d_12_pad_type_0, strides = conv1d_12_strides_0, weight = p_encoder_layers_3_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_3_cast_fp16)[name = tensor<string, []>("conv1d_12_cast_fp16")];
            tensor<int32, [3]> transpose_60_perm_0 = const()[name = tensor<string, []>("transpose_60_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_21_axes_0 = const()[name = tensor<string, []>("layer_norm_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_3_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52531840)))];
            tensor<fp16, [768]> p_encoder_layers_3_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52533440)))];
            tensor<fp16, []> layer_norm_21_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_21_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_60_cast_fp16 = transpose(perm = transpose_60_perm_0, x = conv1d_12_cast_fp16)[name = tensor<string, []>("transpose_219")];
            tensor<fp16, [1, 750, 768]> layer_norm_21_cast_fp16 = layer_norm(axes = layer_norm_21_axes_0, beta = p_encoder_layers_3_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_21_epsilon_0_to_fp16, gamma = p_encoder_layers_3_conv_batch_norm_weight_to_fp16, x = transpose_60_cast_fp16)[name = tensor<string, []>("layer_norm_21_cast_fp16")];
            tensor<int32, [3]> transpose_61_perm_0 = const()[name = tensor<string, []>("transpose_61_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_61_cast_fp16 = transpose(perm = transpose_61_perm_0, x = layer_norm_21_cast_fp16)[name = tensor<string, []>("transpose_218")];
            tensor<fp16, [1, 768, 750]> silu_10_cast_fp16 = silu(x = transpose_61_cast_fp16)[name = tensor<string, []>("silu_10_cast_fp16")];
            tensor<string, []> conv1d_13_pad_type_0 = const()[name = tensor<string, []>("conv1d_13_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_13_strides_0 = const()[name = tensor<string, []>("conv1d_13_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_13_pad_0 = const()[name = tensor<string, []>("conv1d_13_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_13_dilations_0 = const()[name = tensor<string, []>("conv1d_13_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_13_groups_0 = const()[name = tensor<string, []>("conv1d_13_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52535040))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53124928))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_3_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53126528)))];
            tensor<fp16, [1, 768, 750]> conv1d_13_cast_fp16 = conv(bias = p_encoder_layers_3_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_13_dilations_0, groups = conv1d_13_groups_0, pad = conv1d_13_pad_0, pad_type = conv1d_13_pad_type_0, strides = conv1d_13_strides_0, weight = p_encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_10_cast_fp16)[name = tensor<string, []>("conv1d_13_cast_fp16")];
            tensor<int32, [3]> transpose_62_perm_0 = const()[name = tensor<string, []>("transpose_62_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_62_cast_fp16 = transpose(perm = transpose_62_perm_0, x = conv1d_13_cast_fp16)[name = tensor<string, []>("transpose_217")];
            tensor<fp16, [1, 750, 768]> add_26_cast_fp16 = add(x = add_25_cast_fp16, y = transpose_62_cast_fp16)[name = tensor<string, []>("add_26_cast_fp16")];
            tensor<int32, [1]> layer_norm_22_axes_0 = const()[name = tensor<string, []>("layer_norm_22_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53128128)))];
            tensor<fp16, [768]> p_encoder_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53129728)))];
            tensor<fp16, []> layer_norm_22_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_22_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_22_cast_fp16 = layer_norm(axes = layer_norm_22_axes_0, beta = p_encoder_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_22_epsilon_0_to_fp16, gamma = p_encoder_layers_3_norm_feed_forward2_weight_to_fp16, x = add_26_cast_fp16)[name = tensor<string, []>("layer_norm_22_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_3_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53131328))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55490688))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_3_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55496896)))];
            tensor<fp16, [1, 750, 3072]> linear_30_cast_fp16 = linear(bias = p_encoder_layers_3_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_3_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_22_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_11_cast_fp16 = silu(x = linear_30_cast_fp16)[name = tensor<string, []>("silu_11_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_3_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_3_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55503104))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57862464))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_3_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57864064)))];
            tensor<fp16, [1, 750, 768]> linear_31_cast_fp16 = linear(bias = p_encoder_layers_3_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_3_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_11_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
            tensor<fp16, []> const_339_to_fp16 = const()[name = tensor<string, []>("const_339_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_23_cast_fp16 = mul(x = linear_31_cast_fp16, y = const_339_to_fp16)[name = tensor<string, []>("mul_23_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_27_cast_fp16 = add(x = add_26_cast_fp16, y = mul_23_cast_fp16)[name = tensor<string, []>("add_27_cast_fp16")];
            tensor<int32, [1]> layer_norm_23_axes_0 = const()[name = tensor<string, []>("layer_norm_23_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_3_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57865664)))];
            tensor<fp16, [768]> p_encoder_layers_3_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57867264)))];
            tensor<fp16, []> layer_norm_23_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_23_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_23_cast_fp16 = layer_norm(axes = layer_norm_23_axes_0, beta = p_encoder_layers_3_norm_out_bias_to_fp16, epsilon = layer_norm_23_epsilon_0_to_fp16, gamma = p_encoder_layers_3_norm_out_weight_to_fp16, x = add_27_cast_fp16)[name = tensor<string, []>("layer_norm_23_cast_fp16")];
            tensor<int32, [1]> layer_norm_24_axes_0 = const()[name = tensor<string, []>("layer_norm_24_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57868864)))];
            tensor<fp16, [768]> p_encoder_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57870464)))];
            tensor<fp16, []> layer_norm_24_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_24_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_24_cast_fp16 = layer_norm(axes = layer_norm_24_axes_0, beta = p_encoder_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_24_epsilon_0_to_fp16, gamma = p_encoder_layers_4_norm_feed_forward1_weight_to_fp16, x = layer_norm_23_cast_fp16)[name = tensor<string, []>("layer_norm_24_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_4_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57872064))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60231424))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_4_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60237632)))];
            tensor<fp16, [1, 750, 3072]> linear_32_cast_fp16 = linear(bias = p_encoder_layers_4_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_4_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_24_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_12_cast_fp16 = silu(x = linear_32_cast_fp16)[name = tensor<string, []>("silu_12_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_4_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60243840))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62603200))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_4_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62604800)))];
            tensor<fp16, [1, 750, 768]> linear_33_cast_fp16 = linear(bias = p_encoder_layers_4_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_4_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_12_cast_fp16)[name = tensor<string, []>("linear_33_cast_fp16")];
            tensor<fp16, []> const_342_to_fp16 = const()[name = tensor<string, []>("const_342_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_24_cast_fp16 = mul(x = linear_33_cast_fp16, y = const_342_to_fp16)[name = tensor<string, []>("mul_24_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_28_cast_fp16 = add(x = layer_norm_23_cast_fp16, y = mul_24_cast_fp16)[name = tensor<string, []>("add_28_cast_fp16")];
            tensor<int32, [1]> layer_norm_25_axes_0 = const()[name = tensor<string, []>("layer_norm_25_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_4_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62606400)))];
            tensor<fp16, [768]> p_encoder_layers_4_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62608000)))];
            tensor<fp16, []> layer_norm_25_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_25_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_25_cast_fp16 = layer_norm(axes = layer_norm_25_axes_0, beta = p_encoder_layers_4_norm_self_att_bias_to_fp16, epsilon = layer_norm_25_epsilon_0_to_fp16, gamma = p_encoder_layers_4_norm_self_att_weight_to_fp16, x = add_28_cast_fp16)[name = tensor<string, []>("layer_norm_25_cast_fp16")];
            tensor<int32, [4]> const_346 = const()[name = tensor<string, []>("const_346"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_52_perm_1 = const()[name = tensor<string, []>("transpose_52_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_52 = transpose(perm = transpose_52_perm_1, x = layer_norm_25_cast_fp16)[name = tensor<string, []>("transpose_216")];
            tensor<fp16, [750, 1, 16, 48]> view_36_cast_fp16 = reshape(shape = const_346, x = transpose_52)[name = tensor<string, []>("view_36_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_25_cast_fp16 = mul(x = view_36_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_25_cast_fp16")];
            tensor<int32, [4]> slice_19_begin_0 = const()[name = tensor<string, []>("slice_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_19_end_0 = const()[name = tensor<string, []>("slice_19_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_19_end_mask_0 = const()[name = tensor<string, []>("slice_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_19_cast_fp16 = slice_by_index(begin = slice_19_begin_0, end = slice_19_end_0, end_mask = slice_19_end_mask_0, x = view_36_cast_fp16)[name = tensor<string, []>("slice_19_cast_fp16")];
            tensor<int32, [4]> slice_20_begin_0 = const()[name = tensor<string, []>("slice_20_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_20_end_0 = const()[name = tensor<string, []>("slice_20_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_20_end_mask_0 = const()[name = tensor<string, []>("slice_20_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_20_cast_fp16 = slice_by_index(begin = slice_20_begin_0, end = slice_20_end_0, end_mask = slice_20_end_mask_0, x = view_36_cast_fp16)[name = tensor<string, []>("slice_20_cast_fp16")];
            tensor<fp16, []> const_363_promoted_to_fp16 = const()[name = tensor<string, []>("const_363_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_8_cast_fp16 = mul(x = slice_20_cast_fp16, y = const_363_promoted_to_fp16)[name = tensor<string, []>("neg_8_cast_fp16")];
            tensor<int32, []> const_364 = const()[name = tensor<string, []>("const_364"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_8_interleave_0 = const()[name = tensor<string, []>("cat_8_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_8_cast_fp16 = concat(axis = const_364, interleave = cat_8_interleave_0, values = (neg_8_cast_fp16, slice_19_cast_fp16))[name = tensor<string, []>("cat_8_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_26_cast_fp16 = mul(x = cat_8_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_26_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_29_cast_fp16 = add(x = mul_25_cast_fp16, y = mul_26_cast_fp16)[name = tensor<string, []>("add_29_cast_fp16")];
            tensor<int32, [3]> const_373 = const()[name = tensor<string, []>("const_373"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_39_cast_fp16 = reshape(shape = const_373, x = add_29_cast_fp16)[name = tensor<string, []>("view_39_cast_fp16")];
            tensor<int32, [3]> transpose_66_perm_0_1 = const()[name = tensor<string, []>("transpose_66_perm_0_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_4_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62609600))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63199488))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_4_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63201088)))];
            tensor<fp16, [1, 750, 768]> transpose_66_cast_fp16 = transpose(perm = transpose_66_perm_0_1, x = view_39_cast_fp16)[name = tensor<string, []>("transpose_215")];
            tensor<fp16, [1, 750, 768]> linear_34_cast_fp16 = linear(bias = p_encoder_layers_4_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_4_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_66_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
            tensor<int32, [4]> const_382 = const()[name = tensor<string, []>("const_382"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_42_cast_fp16 = reshape(shape = const_382, x = linear_34_cast_fp16)[name = tensor<string, []>("view_42_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_4_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63202688))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63792576))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_4_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63794176)))];
            tensor<fp16, [1, 750, 768]> linear_35_cast_fp16 = linear(bias = p_encoder_layers_4_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_4_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_66_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
            tensor<int32, [4]> const_383 = const()[name = tensor<string, []>("const_383"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_43_cast_fp16 = reshape(shape = const_383, x = linear_35_cast_fp16)[name = tensor<string, []>("view_43_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_4_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63795776))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64385664))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_4_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64387264)))];
            tensor<fp16, [1, 750, 768]> linear_36_cast_fp16 = linear(bias = p_encoder_layers_4_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_4_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_25_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<int32, [4]> const_384 = const()[name = tensor<string, []>("const_384"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_44_cast_fp16 = reshape(shape = const_384, x = linear_36_cast_fp16)[name = tensor<string, []>("view_44_cast_fp16")];
            tensor<int32, [4]> transpose_71_perm_0_1 = const()[name = tensor<string, []>("transpose_71_perm_0_1"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_6_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_6_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_6_cast_fp16 = mul(x = view_43_cast_fp16, y = _inversed_div_6_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_6_cast_fp16")];
            tensor<bool, []> matmul_8_transpose_x_0 = const()[name = tensor<string, []>("matmul_8_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_8_transpose_y_0 = const()[name = tensor<string, []>("matmul_8_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_72_perm_0 = const()[name = tensor<string, []>("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_73_perm_0 = const()[name = tensor<string, []>("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_73 = transpose(perm = transpose_73_perm_0, x = _inversed_div_6_cast_fp16)[name = tensor<string, []>("transpose_213")];
            tensor<fp16, [1, 16, 750, 48]> transpose_72 = transpose(perm = transpose_72_perm_0, x = view_42_cast_fp16)[name = tensor<string, []>("transpose_214")];
            tensor<fp16, [1, 16, 750, 750]> matmul_8_cast_fp16 = matmul(transpose_x = matmul_8_transpose_x_0, transpose_y = matmul_8_transpose_y_0, x = transpose_72, y = transpose_73)[name = tensor<string, []>("matmul_8_cast_fp16")];
            tensor<int32, []> const_394 = const()[name = tensor<string, []>("const_394"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_4_cast_fp16 = softmax(axis = const_394, x = matmul_8_cast_fp16)[name = tensor<string, []>("softmax_4_cast_fp16")];
            tensor<bool, []> matmul_9_transpose_x_0 = const()[name = tensor<string, []>("matmul_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_9_transpose_y_0 = const()[name = tensor<string, []>("matmul_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_71_cast_fp16 = transpose(perm = transpose_71_perm_0_1, x = view_44_cast_fp16)[name = tensor<string, []>("transpose_212")];
            tensor<fp16, [1, 16, 750, 48]> matmul_9_cast_fp16 = matmul(transpose_x = matmul_9_transpose_x_0, transpose_y = matmul_9_transpose_y_0, x = softmax_4_cast_fp16, y = transpose_71_cast_fp16)[name = tensor<string, []>("matmul_9_cast_fp16")];
            tensor<int32, [4]> transpose_73_perm_0_1 = const()[name = tensor<string, []>("transpose_73_perm_0_1"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_397 = const()[name = tensor<string, []>("const_397"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_73_cast_fp16 = transpose(perm = transpose_73_perm_0_1, x = matmul_9_cast_fp16)[name = tensor<string, []>("transpose_211")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_4_cast_fp16 = reshape(shape = const_397, x = transpose_73_cast_fp16)[name = tensor<string, []>("_unsafe_view_4_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_4_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64388864))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64978752))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_4_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64980352)))];
            tensor<fp16, [1, 750, 768]> linear_37_cast_fp16 = linear(bias = p_encoder_layers_4_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_4_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_4_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_31_cast_fp16 = add(x = add_28_cast_fp16, y = linear_37_cast_fp16)[name = tensor<string, []>("add_31_cast_fp16")];
            tensor<int32, [1]> layer_norm_26_axes_0 = const()[name = tensor<string, []>("layer_norm_26_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_4_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64981952)))];
            tensor<fp16, [768]> p_encoder_layers_4_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64983552)))];
            tensor<fp16, []> layer_norm_26_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_26_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_26_cast_fp16 = layer_norm(axes = layer_norm_26_axes_0, beta = p_encoder_layers_4_norm_conv_bias_to_fp16, epsilon = layer_norm_26_epsilon_0_to_fp16, gamma = p_encoder_layers_4_norm_conv_weight_to_fp16, x = add_31_cast_fp16)[name = tensor<string, []>("layer_norm_26_cast_fp16")];
            tensor<int32, [3]> transpose_74_perm_0 = const()[name = tensor<string, []>("transpose_74_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_14_pad_type_0 = const()[name = tensor<string, []>("conv1d_14_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_14_strides_0 = const()[name = tensor<string, []>("conv1d_14_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_14_pad_0 = const()[name = tensor<string, []>("conv1d_14_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_14_dilations_0 = const()[name = tensor<string, []>("conv1d_14_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_14_groups_0 = const()[name = tensor<string, []>("conv1d_14_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64985152))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66164864))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_4_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66168000)))];
            tensor<fp16, [1, 768, 750]> transpose_74_cast_fp16 = transpose(perm = transpose_74_perm_0, x = layer_norm_26_cast_fp16)[name = tensor<string, []>("transpose_210")];
            tensor<fp16, [1, 1536, 750]> conv1d_14_cast_fp16 = conv(bias = p_encoder_layers_4_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_14_dilations_0, groups = conv1d_14_groups_0, pad = conv1d_14_pad_0, pad_type = conv1d_14_pad_type_0, strides = conv1d_14_strides_0, weight = p_encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_74_cast_fp16)[name = tensor<string, []>("conv1d_14_cast_fp16")];
            tensor<int32, []> glu_4_split_num_splits_0 = const()[name = tensor<string, []>("glu_4_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_4_split_axis_0 = const()[name = tensor<string, []>("glu_4_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_4_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_4_split_cast_fp16_1 = split(axis = glu_4_split_axis_0, num_splits = glu_4_split_num_splits_0, x = conv1d_14_cast_fp16)[name = tensor<string, []>("glu_4_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_4_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_4_split_cast_fp16_1)[name = tensor<string, []>("glu_4_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_4_cast_fp16 = mul(x = glu_4_split_cast_fp16_0, y = glu_4_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_4_cast_fp16")];
            tensor<string, []> conv1d_15_pad_type_0 = const()[name = tensor<string, []>("conv1d_15_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_15_pad_0 = const()[name = tensor<string, []>("conv1d_15_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_15_groups_0 = const()[name = tensor<string, []>("conv1d_15_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_15_strides_0 = const()[name = tensor<string, []>("conv1d_15_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_15_dilations_0 = const()[name = tensor<string, []>("conv1d_15_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_4_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66171136))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66175040))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_4_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66176640)))];
            tensor<fp16, [1, 768, 750]> conv1d_15_cast_fp16 = conv(bias = p_encoder_layers_4_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_15_dilations_0, groups = conv1d_15_groups_0, pad = conv1d_15_pad_0, pad_type = conv1d_15_pad_type_0, strides = conv1d_15_strides_0, weight = p_encoder_layers_4_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_4_cast_fp16)[name = tensor<string, []>("conv1d_15_cast_fp16")];
            tensor<int32, [3]> transpose_75_perm_0 = const()[name = tensor<string, []>("transpose_75_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_27_axes_0 = const()[name = tensor<string, []>("layer_norm_27_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_4_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66178240)))];
            tensor<fp16, [768]> p_encoder_layers_4_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66179840)))];
            tensor<fp16, []> layer_norm_27_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_27_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_75_cast_fp16 = transpose(perm = transpose_75_perm_0, x = conv1d_15_cast_fp16)[name = tensor<string, []>("transpose_209")];
            tensor<fp16, [1, 750, 768]> layer_norm_27_cast_fp16 = layer_norm(axes = layer_norm_27_axes_0, beta = p_encoder_layers_4_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_27_epsilon_0_to_fp16, gamma = p_encoder_layers_4_conv_batch_norm_weight_to_fp16, x = transpose_75_cast_fp16)[name = tensor<string, []>("layer_norm_27_cast_fp16")];
            tensor<int32, [3]> transpose_76_perm_0 = const()[name = tensor<string, []>("transpose_76_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_76_cast_fp16 = transpose(perm = transpose_76_perm_0, x = layer_norm_27_cast_fp16)[name = tensor<string, []>("transpose_208")];
            tensor<fp16, [1, 768, 750]> silu_13_cast_fp16 = silu(x = transpose_76_cast_fp16)[name = tensor<string, []>("silu_13_cast_fp16")];
            tensor<string, []> conv1d_16_pad_type_0 = const()[name = tensor<string, []>("conv1d_16_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_16_strides_0 = const()[name = tensor<string, []>("conv1d_16_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_16_pad_0 = const()[name = tensor<string, []>("conv1d_16_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_16_dilations_0 = const()[name = tensor<string, []>("conv1d_16_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_16_groups_0 = const()[name = tensor<string, []>("conv1d_16_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66181440))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66771328))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_4_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66772928)))];
            tensor<fp16, [1, 768, 750]> conv1d_16_cast_fp16 = conv(bias = p_encoder_layers_4_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_16_dilations_0, groups = conv1d_16_groups_0, pad = conv1d_16_pad_0, pad_type = conv1d_16_pad_type_0, strides = conv1d_16_strides_0, weight = p_encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_13_cast_fp16)[name = tensor<string, []>("conv1d_16_cast_fp16")];
            tensor<int32, [3]> transpose_77_perm_0 = const()[name = tensor<string, []>("transpose_77_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_77_cast_fp16 = transpose(perm = transpose_77_perm_0, x = conv1d_16_cast_fp16)[name = tensor<string, []>("transpose_207")];
            tensor<fp16, [1, 750, 768]> add_32_cast_fp16 = add(x = add_31_cast_fp16, y = transpose_77_cast_fp16)[name = tensor<string, []>("add_32_cast_fp16")];
            tensor<int32, [1]> layer_norm_28_axes_0 = const()[name = tensor<string, []>("layer_norm_28_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66774528)))];
            tensor<fp16, [768]> p_encoder_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66776128)))];
            tensor<fp16, []> layer_norm_28_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_28_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_28_cast_fp16 = layer_norm(axes = layer_norm_28_axes_0, beta = p_encoder_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_28_epsilon_0_to_fp16, gamma = p_encoder_layers_4_norm_feed_forward2_weight_to_fp16, x = add_32_cast_fp16)[name = tensor<string, []>("layer_norm_28_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_4_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66777728))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69137088))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_4_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69143296)))];
            tensor<fp16, [1, 750, 3072]> linear_38_cast_fp16 = linear(bias = p_encoder_layers_4_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_4_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_28_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_14_cast_fp16 = silu(x = linear_38_cast_fp16)[name = tensor<string, []>("silu_14_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_4_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_4_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69149504))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71508864))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_4_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71510464)))];
            tensor<fp16, [1, 750, 768]> linear_39_cast_fp16 = linear(bias = p_encoder_layers_4_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_4_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_14_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
            tensor<fp16, []> const_416_to_fp16 = const()[name = tensor<string, []>("const_416_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_29_cast_fp16 = mul(x = linear_39_cast_fp16, y = const_416_to_fp16)[name = tensor<string, []>("mul_29_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_33_cast_fp16 = add(x = add_32_cast_fp16, y = mul_29_cast_fp16)[name = tensor<string, []>("add_33_cast_fp16")];
            tensor<int32, [1]> layer_norm_29_axes_0 = const()[name = tensor<string, []>("layer_norm_29_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_4_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71512064)))];
            tensor<fp16, [768]> p_encoder_layers_4_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71513664)))];
            tensor<fp16, []> layer_norm_29_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_29_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_29_cast_fp16 = layer_norm(axes = layer_norm_29_axes_0, beta = p_encoder_layers_4_norm_out_bias_to_fp16, epsilon = layer_norm_29_epsilon_0_to_fp16, gamma = p_encoder_layers_4_norm_out_weight_to_fp16, x = add_33_cast_fp16)[name = tensor<string, []>("layer_norm_29_cast_fp16")];
            tensor<int32, [1]> layer_norm_30_axes_0 = const()[name = tensor<string, []>("layer_norm_30_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71515264)))];
            tensor<fp16, [768]> p_encoder_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71516864)))];
            tensor<fp16, []> layer_norm_30_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_30_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_30_cast_fp16 = layer_norm(axes = layer_norm_30_axes_0, beta = p_encoder_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_30_epsilon_0_to_fp16, gamma = p_encoder_layers_5_norm_feed_forward1_weight_to_fp16, x = layer_norm_29_cast_fp16)[name = tensor<string, []>("layer_norm_30_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_5_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71518464))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73877824))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_5_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73884032)))];
            tensor<fp16, [1, 750, 3072]> linear_40_cast_fp16 = linear(bias = p_encoder_layers_5_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_5_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_30_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_15_cast_fp16 = silu(x = linear_40_cast_fp16)[name = tensor<string, []>("silu_15_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_5_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73890240))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76249600))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_5_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76251200)))];
            tensor<fp16, [1, 750, 768]> linear_41_cast_fp16 = linear(bias = p_encoder_layers_5_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_5_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_15_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
            tensor<fp16, []> const_419_to_fp16 = const()[name = tensor<string, []>("const_419_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_30_cast_fp16 = mul(x = linear_41_cast_fp16, y = const_419_to_fp16)[name = tensor<string, []>("mul_30_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_34_cast_fp16 = add(x = layer_norm_29_cast_fp16, y = mul_30_cast_fp16)[name = tensor<string, []>("add_34_cast_fp16")];
            tensor<int32, [1]> layer_norm_31_axes_0 = const()[name = tensor<string, []>("layer_norm_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_5_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76252800)))];
            tensor<fp16, [768]> p_encoder_layers_5_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76254400)))];
            tensor<fp16, []> layer_norm_31_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_31_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_31_cast_fp16 = layer_norm(axes = layer_norm_31_axes_0, beta = p_encoder_layers_5_norm_self_att_bias_to_fp16, epsilon = layer_norm_31_epsilon_0_to_fp16, gamma = p_encoder_layers_5_norm_self_att_weight_to_fp16, x = add_34_cast_fp16)[name = tensor<string, []>("layer_norm_31_cast_fp16")];
            tensor<int32, [4]> const_423 = const()[name = tensor<string, []>("const_423"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_53_perm_1 = const()[name = tensor<string, []>("transpose_53_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_53 = transpose(perm = transpose_53_perm_1, x = layer_norm_31_cast_fp16)[name = tensor<string, []>("transpose_206")];
            tensor<fp16, [750, 1, 16, 48]> view_45_cast_fp16 = reshape(shape = const_423, x = transpose_53)[name = tensor<string, []>("view_45_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_31_cast_fp16 = mul(x = view_45_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_31_cast_fp16")];
            tensor<int32, [4]> slice_23_begin_0 = const()[name = tensor<string, []>("slice_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_23_end_0 = const()[name = tensor<string, []>("slice_23_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_23_end_mask_0 = const()[name = tensor<string, []>("slice_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_23_cast_fp16 = slice_by_index(begin = slice_23_begin_0, end = slice_23_end_0, end_mask = slice_23_end_mask_0, x = view_45_cast_fp16)[name = tensor<string, []>("slice_23_cast_fp16")];
            tensor<int32, [4]> slice_24_begin_0 = const()[name = tensor<string, []>("slice_24_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_24_end_0 = const()[name = tensor<string, []>("slice_24_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_24_end_mask_0 = const()[name = tensor<string, []>("slice_24_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_24_cast_fp16 = slice_by_index(begin = slice_24_begin_0, end = slice_24_end_0, end_mask = slice_24_end_mask_0, x = view_45_cast_fp16)[name = tensor<string, []>("slice_24_cast_fp16")];
            tensor<fp16, []> const_440_promoted_to_fp16 = const()[name = tensor<string, []>("const_440_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_10_cast_fp16 = mul(x = slice_24_cast_fp16, y = const_440_promoted_to_fp16)[name = tensor<string, []>("neg_10_cast_fp16")];
            tensor<int32, []> const_441 = const()[name = tensor<string, []>("const_441"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_10_interleave_0 = const()[name = tensor<string, []>("cat_10_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_10_cast_fp16 = concat(axis = const_441, interleave = cat_10_interleave_0, values = (neg_10_cast_fp16, slice_23_cast_fp16))[name = tensor<string, []>("cat_10_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_32_cast_fp16 = mul(x = cat_10_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_32_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_35_cast_fp16 = add(x = mul_31_cast_fp16, y = mul_32_cast_fp16)[name = tensor<string, []>("add_35_cast_fp16")];
            tensor<int32, [3]> const_450 = const()[name = tensor<string, []>("const_450"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_48_cast_fp16 = reshape(shape = const_450, x = add_35_cast_fp16)[name = tensor<string, []>("view_48_cast_fp16")];
            tensor<int32, [3]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_5_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76256000))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76845888))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_5_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76847488)))];
            tensor<fp16, [1, 750, 768]> transpose_81_cast_fp16 = transpose(perm = transpose_81_perm_0, x = view_48_cast_fp16)[name = tensor<string, []>("transpose_205")];
            tensor<fp16, [1, 750, 768]> linear_42_cast_fp16 = linear(bias = p_encoder_layers_5_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_5_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_81_cast_fp16)[name = tensor<string, []>("linear_42_cast_fp16")];
            tensor<int32, [4]> const_459 = const()[name = tensor<string, []>("const_459"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_51_cast_fp16 = reshape(shape = const_459, x = linear_42_cast_fp16)[name = tensor<string, []>("view_51_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_5_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76849088))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77438976))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_5_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77440576)))];
            tensor<fp16, [1, 750, 768]> linear_43_cast_fp16 = linear(bias = p_encoder_layers_5_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_5_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_81_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<int32, [4]> const_460 = const()[name = tensor<string, []>("const_460"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_52_cast_fp16 = reshape(shape = const_460, x = linear_43_cast_fp16)[name = tensor<string, []>("view_52_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_5_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77442176))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78032064))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_5_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78033664)))];
            tensor<fp16, [1, 750, 768]> linear_44_cast_fp16 = linear(bias = p_encoder_layers_5_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_5_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_31_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<int32, [4]> const_461 = const()[name = tensor<string, []>("const_461"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_53_cast_fp16 = reshape(shape = const_461, x = linear_44_cast_fp16)[name = tensor<string, []>("view_53_cast_fp16")];
            tensor<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_7_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_7_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_7_cast_fp16 = mul(x = view_52_cast_fp16, y = _inversed_div_7_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_7_cast_fp16")];
            tensor<bool, []> matmul_10_transpose_x_0 = const()[name = tensor<string, []>("matmul_10_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_10_transpose_y_0 = const()[name = tensor<string, []>("matmul_10_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_74_perm_0_1 = const()[name = tensor<string, []>("transpose_74_perm_0_1"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_75_perm_0_1 = const()[name = tensor<string, []>("transpose_75_perm_0_1"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_75 = transpose(perm = transpose_75_perm_0_1, x = _inversed_div_7_cast_fp16)[name = tensor<string, []>("transpose_203")];
            tensor<fp16, [1, 16, 750, 48]> transpose_74 = transpose(perm = transpose_74_perm_0_1, x = view_51_cast_fp16)[name = tensor<string, []>("transpose_204")];
            tensor<fp16, [1, 16, 750, 750]> matmul_10_cast_fp16 = matmul(transpose_x = matmul_10_transpose_x_0, transpose_y = matmul_10_transpose_y_0, x = transpose_74, y = transpose_75)[name = tensor<string, []>("matmul_10_cast_fp16")];
            tensor<int32, []> const_471 = const()[name = tensor<string, []>("const_471"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_5_cast_fp16 = softmax(axis = const_471, x = matmul_10_cast_fp16)[name = tensor<string, []>("softmax_5_cast_fp16")];
            tensor<bool, []> matmul_11_transpose_x_0 = const()[name = tensor<string, []>("matmul_11_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_11_transpose_y_0 = const()[name = tensor<string, []>("matmul_11_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_86_cast_fp16 = transpose(perm = transpose_86_perm_0, x = view_53_cast_fp16)[name = tensor<string, []>("transpose_202")];
            tensor<fp16, [1, 16, 750, 48]> matmul_11_cast_fp16 = matmul(transpose_x = matmul_11_transpose_x_0, transpose_y = matmul_11_transpose_y_0, x = softmax_5_cast_fp16, y = transpose_86_cast_fp16)[name = tensor<string, []>("matmul_11_cast_fp16")];
            tensor<int32, [4]> transpose_88_perm_0 = const()[name = tensor<string, []>("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_474 = const()[name = tensor<string, []>("const_474"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_88_cast_fp16 = transpose(perm = transpose_88_perm_0, x = matmul_11_cast_fp16)[name = tensor<string, []>("transpose_201")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_5_cast_fp16 = reshape(shape = const_474, x = transpose_88_cast_fp16)[name = tensor<string, []>("_unsafe_view_5_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_5_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78035264))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78625152))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_5_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78626752)))];
            tensor<fp16, [1, 750, 768]> linear_45_cast_fp16 = linear(bias = p_encoder_layers_5_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_5_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_5_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_37_cast_fp16 = add(x = add_34_cast_fp16, y = linear_45_cast_fp16)[name = tensor<string, []>("add_37_cast_fp16")];
            tensor<int32, [1]> layer_norm_32_axes_0 = const()[name = tensor<string, []>("layer_norm_32_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_5_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78628352)))];
            tensor<fp16, [768]> p_encoder_layers_5_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78629952)))];
            tensor<fp16, []> layer_norm_32_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_32_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_32_cast_fp16 = layer_norm(axes = layer_norm_32_axes_0, beta = p_encoder_layers_5_norm_conv_bias_to_fp16, epsilon = layer_norm_32_epsilon_0_to_fp16, gamma = p_encoder_layers_5_norm_conv_weight_to_fp16, x = add_37_cast_fp16)[name = tensor<string, []>("layer_norm_32_cast_fp16")];
            tensor<int32, [3]> transpose_89_perm_0 = const()[name = tensor<string, []>("transpose_89_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_17_pad_type_0 = const()[name = tensor<string, []>("conv1d_17_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_17_strides_0 = const()[name = tensor<string, []>("conv1d_17_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_17_pad_0 = const()[name = tensor<string, []>("conv1d_17_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_17_dilations_0 = const()[name = tensor<string, []>("conv1d_17_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_17_groups_0 = const()[name = tensor<string, []>("conv1d_17_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78631552))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79811264))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_5_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79814400)))];
            tensor<fp16, [1, 768, 750]> transpose_89_cast_fp16 = transpose(perm = transpose_89_perm_0, x = layer_norm_32_cast_fp16)[name = tensor<string, []>("transpose_200")];
            tensor<fp16, [1, 1536, 750]> conv1d_17_cast_fp16 = conv(bias = p_encoder_layers_5_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_17_dilations_0, groups = conv1d_17_groups_0, pad = conv1d_17_pad_0, pad_type = conv1d_17_pad_type_0, strides = conv1d_17_strides_0, weight = p_encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_89_cast_fp16)[name = tensor<string, []>("conv1d_17_cast_fp16")];
            tensor<int32, []> glu_5_split_num_splits_0 = const()[name = tensor<string, []>("glu_5_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_5_split_axis_0 = const()[name = tensor<string, []>("glu_5_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_5_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_5_split_cast_fp16_1 = split(axis = glu_5_split_axis_0, num_splits = glu_5_split_num_splits_0, x = conv1d_17_cast_fp16)[name = tensor<string, []>("glu_5_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_5_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_5_split_cast_fp16_1)[name = tensor<string, []>("glu_5_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_5_cast_fp16 = mul(x = glu_5_split_cast_fp16_0, y = glu_5_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_5_cast_fp16")];
            tensor<string, []> conv1d_18_pad_type_0 = const()[name = tensor<string, []>("conv1d_18_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_18_pad_0 = const()[name = tensor<string, []>("conv1d_18_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_18_groups_0 = const()[name = tensor<string, []>("conv1d_18_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_18_strides_0 = const()[name = tensor<string, []>("conv1d_18_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_18_dilations_0 = const()[name = tensor<string, []>("conv1d_18_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_5_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79817536))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79821440))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_5_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79823040)))];
            tensor<fp16, [1, 768, 750]> conv1d_18_cast_fp16 = conv(bias = p_encoder_layers_5_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_18_dilations_0, groups = conv1d_18_groups_0, pad = conv1d_18_pad_0, pad_type = conv1d_18_pad_type_0, strides = conv1d_18_strides_0, weight = p_encoder_layers_5_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_5_cast_fp16)[name = tensor<string, []>("conv1d_18_cast_fp16")];
            tensor<int32, [3]> transpose_90_perm_0 = const()[name = tensor<string, []>("transpose_90_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_33_axes_0 = const()[name = tensor<string, []>("layer_norm_33_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_5_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79824640)))];
            tensor<fp16, [768]> p_encoder_layers_5_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79826240)))];
            tensor<fp16, []> layer_norm_33_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_33_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_90_cast_fp16 = transpose(perm = transpose_90_perm_0, x = conv1d_18_cast_fp16)[name = tensor<string, []>("transpose_199")];
            tensor<fp16, [1, 750, 768]> layer_norm_33_cast_fp16 = layer_norm(axes = layer_norm_33_axes_0, beta = p_encoder_layers_5_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_33_epsilon_0_to_fp16, gamma = p_encoder_layers_5_conv_batch_norm_weight_to_fp16, x = transpose_90_cast_fp16)[name = tensor<string, []>("layer_norm_33_cast_fp16")];
            tensor<int32, [3]> transpose_91_perm_0 = const()[name = tensor<string, []>("transpose_91_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_91_cast_fp16 = transpose(perm = transpose_91_perm_0, x = layer_norm_33_cast_fp16)[name = tensor<string, []>("transpose_198")];
            tensor<fp16, [1, 768, 750]> silu_16_cast_fp16 = silu(x = transpose_91_cast_fp16)[name = tensor<string, []>("silu_16_cast_fp16")];
            tensor<string, []> conv1d_19_pad_type_0 = const()[name = tensor<string, []>("conv1d_19_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_19_strides_0 = const()[name = tensor<string, []>("conv1d_19_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_19_pad_0 = const()[name = tensor<string, []>("conv1d_19_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_19_dilations_0 = const()[name = tensor<string, []>("conv1d_19_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_19_groups_0 = const()[name = tensor<string, []>("conv1d_19_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79827840))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80417728))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_5_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80419328)))];
            tensor<fp16, [1, 768, 750]> conv1d_19_cast_fp16 = conv(bias = p_encoder_layers_5_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_19_dilations_0, groups = conv1d_19_groups_0, pad = conv1d_19_pad_0, pad_type = conv1d_19_pad_type_0, strides = conv1d_19_strides_0, weight = p_encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_16_cast_fp16)[name = tensor<string, []>("conv1d_19_cast_fp16")];
            tensor<int32, [3]> transpose_92_perm_0 = const()[name = tensor<string, []>("transpose_92_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_92_cast_fp16 = transpose(perm = transpose_92_perm_0, x = conv1d_19_cast_fp16)[name = tensor<string, []>("transpose_197")];
            tensor<fp16, [1, 750, 768]> add_38_cast_fp16 = add(x = add_37_cast_fp16, y = transpose_92_cast_fp16)[name = tensor<string, []>("add_38_cast_fp16")];
            tensor<int32, [1]> layer_norm_34_axes_0 = const()[name = tensor<string, []>("layer_norm_34_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80420928)))];
            tensor<fp16, [768]> p_encoder_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80422528)))];
            tensor<fp16, []> layer_norm_34_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_34_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_34_cast_fp16 = layer_norm(axes = layer_norm_34_axes_0, beta = p_encoder_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_34_epsilon_0_to_fp16, gamma = p_encoder_layers_5_norm_feed_forward2_weight_to_fp16, x = add_38_cast_fp16)[name = tensor<string, []>("layer_norm_34_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_5_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80424128))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82783488))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_5_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82789696)))];
            tensor<fp16, [1, 750, 3072]> linear_46_cast_fp16 = linear(bias = p_encoder_layers_5_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_5_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_34_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_17_cast_fp16 = silu(x = linear_46_cast_fp16)[name = tensor<string, []>("silu_17_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_5_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_5_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(82795904))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85155264))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_5_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85156864)))];
            tensor<fp16, [1, 750, 768]> linear_47_cast_fp16 = linear(bias = p_encoder_layers_5_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_5_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_17_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
            tensor<fp16, []> const_493_to_fp16 = const()[name = tensor<string, []>("const_493_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_35_cast_fp16 = mul(x = linear_47_cast_fp16, y = const_493_to_fp16)[name = tensor<string, []>("mul_35_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_39_cast_fp16 = add(x = add_38_cast_fp16, y = mul_35_cast_fp16)[name = tensor<string, []>("add_39_cast_fp16")];
            tensor<int32, [1]> layer_norm_35_axes_0 = const()[name = tensor<string, []>("layer_norm_35_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_5_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85158464)))];
            tensor<fp16, [768]> p_encoder_layers_5_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85160064)))];
            tensor<fp16, []> layer_norm_35_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_35_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_35_cast_fp16 = layer_norm(axes = layer_norm_35_axes_0, beta = p_encoder_layers_5_norm_out_bias_to_fp16, epsilon = layer_norm_35_epsilon_0_to_fp16, gamma = p_encoder_layers_5_norm_out_weight_to_fp16, x = add_39_cast_fp16)[name = tensor<string, []>("layer_norm_35_cast_fp16")];
            tensor<int32, [1]> layer_norm_36_axes_0 = const()[name = tensor<string, []>("layer_norm_36_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85161664)))];
            tensor<fp16, [768]> p_encoder_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85163264)))];
            tensor<fp16, []> layer_norm_36_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_36_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_36_cast_fp16 = layer_norm(axes = layer_norm_36_axes_0, beta = p_encoder_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_36_epsilon_0_to_fp16, gamma = p_encoder_layers_6_norm_feed_forward1_weight_to_fp16, x = layer_norm_35_cast_fp16)[name = tensor<string, []>("layer_norm_36_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_6_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85164864))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87524224))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_6_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87530432)))];
            tensor<fp16, [1, 750, 3072]> linear_48_cast_fp16 = linear(bias = p_encoder_layers_6_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_6_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_36_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_18_cast_fp16 = silu(x = linear_48_cast_fp16)[name = tensor<string, []>("silu_18_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_6_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87536640))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89896000))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_6_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89897600)))];
            tensor<fp16, [1, 750, 768]> linear_49_cast_fp16 = linear(bias = p_encoder_layers_6_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_6_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_18_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<fp16, []> const_496_to_fp16 = const()[name = tensor<string, []>("const_496_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_36_cast_fp16 = mul(x = linear_49_cast_fp16, y = const_496_to_fp16)[name = tensor<string, []>("mul_36_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_40_cast_fp16 = add(x = layer_norm_35_cast_fp16, y = mul_36_cast_fp16)[name = tensor<string, []>("add_40_cast_fp16")];
            tensor<int32, [1]> layer_norm_37_axes_0 = const()[name = tensor<string, []>("layer_norm_37_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_6_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89899200)))];
            tensor<fp16, [768]> p_encoder_layers_6_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89900800)))];
            tensor<fp16, []> layer_norm_37_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_37_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_37_cast_fp16 = layer_norm(axes = layer_norm_37_axes_0, beta = p_encoder_layers_6_norm_self_att_bias_to_fp16, epsilon = layer_norm_37_epsilon_0_to_fp16, gamma = p_encoder_layers_6_norm_self_att_weight_to_fp16, x = add_40_cast_fp16)[name = tensor<string, []>("layer_norm_37_cast_fp16")];
            tensor<int32, [4]> const_500 = const()[name = tensor<string, []>("const_500"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_54_perm_1 = const()[name = tensor<string, []>("transpose_54_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_54 = transpose(perm = transpose_54_perm_1, x = layer_norm_37_cast_fp16)[name = tensor<string, []>("transpose_196")];
            tensor<fp16, [750, 1, 16, 48]> view_54_cast_fp16 = reshape(shape = const_500, x = transpose_54)[name = tensor<string, []>("view_54_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_37_cast_fp16 = mul(x = view_54_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_37_cast_fp16")];
            tensor<int32, [4]> slice_27_begin_0 = const()[name = tensor<string, []>("slice_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_27_end_0 = const()[name = tensor<string, []>("slice_27_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_27_end_mask_0 = const()[name = tensor<string, []>("slice_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_27_cast_fp16 = slice_by_index(begin = slice_27_begin_0, end = slice_27_end_0, end_mask = slice_27_end_mask_0, x = view_54_cast_fp16)[name = tensor<string, []>("slice_27_cast_fp16")];
            tensor<int32, [4]> slice_28_begin_0 = const()[name = tensor<string, []>("slice_28_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_28_end_0 = const()[name = tensor<string, []>("slice_28_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_28_end_mask_0 = const()[name = tensor<string, []>("slice_28_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_28_cast_fp16 = slice_by_index(begin = slice_28_begin_0, end = slice_28_end_0, end_mask = slice_28_end_mask_0, x = view_54_cast_fp16)[name = tensor<string, []>("slice_28_cast_fp16")];
            tensor<fp16, []> const_517_promoted_to_fp16 = const()[name = tensor<string, []>("const_517_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_12_cast_fp16 = mul(x = slice_28_cast_fp16, y = const_517_promoted_to_fp16)[name = tensor<string, []>("neg_12_cast_fp16")];
            tensor<int32, []> const_518 = const()[name = tensor<string, []>("const_518"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_12_interleave_0 = const()[name = tensor<string, []>("cat_12_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_12_cast_fp16 = concat(axis = const_518, interleave = cat_12_interleave_0, values = (neg_12_cast_fp16, slice_27_cast_fp16))[name = tensor<string, []>("cat_12_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_38_cast_fp16 = mul(x = cat_12_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_38_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_41_cast_fp16 = add(x = mul_37_cast_fp16, y = mul_38_cast_fp16)[name = tensor<string, []>("add_41_cast_fp16")];
            tensor<int32, [3]> const_527 = const()[name = tensor<string, []>("const_527"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_57_cast_fp16 = reshape(shape = const_527, x = add_41_cast_fp16)[name = tensor<string, []>("view_57_cast_fp16")];
            tensor<int32, [3]> transpose_96_perm_0 = const()[name = tensor<string, []>("transpose_96_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_6_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89902400))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90492288))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_6_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90493888)))];
            tensor<fp16, [1, 750, 768]> transpose_96_cast_fp16 = transpose(perm = transpose_96_perm_0, x = view_57_cast_fp16)[name = tensor<string, []>("transpose_195")];
            tensor<fp16, [1, 750, 768]> linear_50_cast_fp16 = linear(bias = p_encoder_layers_6_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_6_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_96_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> const_536 = const()[name = tensor<string, []>("const_536"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_60_cast_fp16 = reshape(shape = const_536, x = linear_50_cast_fp16)[name = tensor<string, []>("view_60_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_6_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90495488))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91085376))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_6_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91086976)))];
            tensor<fp16, [1, 750, 768]> linear_51_cast_fp16 = linear(bias = p_encoder_layers_6_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_6_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_96_cast_fp16)[name = tensor<string, []>("linear_51_cast_fp16")];
            tensor<int32, [4]> const_537 = const()[name = tensor<string, []>("const_537"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_61_cast_fp16 = reshape(shape = const_537, x = linear_51_cast_fp16)[name = tensor<string, []>("view_61_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_6_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91088576))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91678464))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_6_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91680064)))];
            tensor<fp16, [1, 750, 768]> linear_52_cast_fp16 = linear(bias = p_encoder_layers_6_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_6_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_37_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
            tensor<int32, [4]> const_538 = const()[name = tensor<string, []>("const_538"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_62_cast_fp16 = reshape(shape = const_538, x = linear_52_cast_fp16)[name = tensor<string, []>("view_62_cast_fp16")];
            tensor<int32, [4]> transpose_101_perm_0 = const()[name = tensor<string, []>("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_8_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_8_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_8_cast_fp16 = mul(x = view_61_cast_fp16, y = _inversed_div_8_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_8_cast_fp16")];
            tensor<bool, []> matmul_12_transpose_x_0 = const()[name = tensor<string, []>("matmul_12_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_12_transpose_y_0 = const()[name = tensor<string, []>("matmul_12_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_76_perm_0_1 = const()[name = tensor<string, []>("transpose_76_perm_0_1"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_77_perm_0_1 = const()[name = tensor<string, []>("transpose_77_perm_0_1"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_77 = transpose(perm = transpose_77_perm_0_1, x = _inversed_div_8_cast_fp16)[name = tensor<string, []>("transpose_193")];
            tensor<fp16, [1, 16, 750, 48]> transpose_76 = transpose(perm = transpose_76_perm_0_1, x = view_60_cast_fp16)[name = tensor<string, []>("transpose_194")];
            tensor<fp16, [1, 16, 750, 750]> matmul_12_cast_fp16 = matmul(transpose_x = matmul_12_transpose_x_0, transpose_y = matmul_12_transpose_y_0, x = transpose_76, y = transpose_77)[name = tensor<string, []>("matmul_12_cast_fp16")];
            tensor<int32, []> const_548 = const()[name = tensor<string, []>("const_548"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_6_cast_fp16 = softmax(axis = const_548, x = matmul_12_cast_fp16)[name = tensor<string, []>("softmax_6_cast_fp16")];
            tensor<bool, []> matmul_13_transpose_x_0 = const()[name = tensor<string, []>("matmul_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_13_transpose_y_0 = const()[name = tensor<string, []>("matmul_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_101_cast_fp16 = transpose(perm = transpose_101_perm_0, x = view_62_cast_fp16)[name = tensor<string, []>("transpose_192")];
            tensor<fp16, [1, 16, 750, 48]> matmul_13_cast_fp16 = matmul(transpose_x = matmul_13_transpose_x_0, transpose_y = matmul_13_transpose_y_0, x = softmax_6_cast_fp16, y = transpose_101_cast_fp16)[name = tensor<string, []>("matmul_13_cast_fp16")];
            tensor<int32, [4]> transpose_103_perm_0 = const()[name = tensor<string, []>("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_551 = const()[name = tensor<string, []>("const_551"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_103_cast_fp16 = transpose(perm = transpose_103_perm_0, x = matmul_13_cast_fp16)[name = tensor<string, []>("transpose_191")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_6_cast_fp16 = reshape(shape = const_551, x = transpose_103_cast_fp16)[name = tensor<string, []>("_unsafe_view_6_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_6_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91681664))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92271552))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_6_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92273152)))];
            tensor<fp16, [1, 750, 768]> linear_53_cast_fp16 = linear(bias = p_encoder_layers_6_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_6_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_6_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_43_cast_fp16 = add(x = add_40_cast_fp16, y = linear_53_cast_fp16)[name = tensor<string, []>("add_43_cast_fp16")];
            tensor<int32, [1]> layer_norm_38_axes_0 = const()[name = tensor<string, []>("layer_norm_38_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_6_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92274752)))];
            tensor<fp16, [768]> p_encoder_layers_6_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92276352)))];
            tensor<fp16, []> layer_norm_38_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_38_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_38_cast_fp16 = layer_norm(axes = layer_norm_38_axes_0, beta = p_encoder_layers_6_norm_conv_bias_to_fp16, epsilon = layer_norm_38_epsilon_0_to_fp16, gamma = p_encoder_layers_6_norm_conv_weight_to_fp16, x = add_43_cast_fp16)[name = tensor<string, []>("layer_norm_38_cast_fp16")];
            tensor<int32, [3]> transpose_104_perm_0 = const()[name = tensor<string, []>("transpose_104_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_20_pad_type_0 = const()[name = tensor<string, []>("conv1d_20_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_20_strides_0 = const()[name = tensor<string, []>("conv1d_20_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_20_pad_0 = const()[name = tensor<string, []>("conv1d_20_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_20_dilations_0 = const()[name = tensor<string, []>("conv1d_20_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_20_groups_0 = const()[name = tensor<string, []>("conv1d_20_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92277952))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93457664))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_6_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93460800)))];
            tensor<fp16, [1, 768, 750]> transpose_104_cast_fp16 = transpose(perm = transpose_104_perm_0, x = layer_norm_38_cast_fp16)[name = tensor<string, []>("transpose_190")];
            tensor<fp16, [1, 1536, 750]> conv1d_20_cast_fp16 = conv(bias = p_encoder_layers_6_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_20_dilations_0, groups = conv1d_20_groups_0, pad = conv1d_20_pad_0, pad_type = conv1d_20_pad_type_0, strides = conv1d_20_strides_0, weight = p_encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_104_cast_fp16)[name = tensor<string, []>("conv1d_20_cast_fp16")];
            tensor<int32, []> glu_6_split_num_splits_0 = const()[name = tensor<string, []>("glu_6_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_6_split_axis_0 = const()[name = tensor<string, []>("glu_6_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_6_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_6_split_cast_fp16_1 = split(axis = glu_6_split_axis_0, num_splits = glu_6_split_num_splits_0, x = conv1d_20_cast_fp16)[name = tensor<string, []>("glu_6_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_6_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_6_split_cast_fp16_1)[name = tensor<string, []>("glu_6_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_6_cast_fp16 = mul(x = glu_6_split_cast_fp16_0, y = glu_6_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_6_cast_fp16")];
            tensor<string, []> conv1d_21_pad_type_0 = const()[name = tensor<string, []>("conv1d_21_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_21_pad_0 = const()[name = tensor<string, []>("conv1d_21_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_21_groups_0 = const()[name = tensor<string, []>("conv1d_21_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_21_strides_0 = const()[name = tensor<string, []>("conv1d_21_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_21_dilations_0 = const()[name = tensor<string, []>("conv1d_21_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_6_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93463936))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93467840))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_6_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93469440)))];
            tensor<fp16, [1, 768, 750]> conv1d_21_cast_fp16 = conv(bias = p_encoder_layers_6_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_21_dilations_0, groups = conv1d_21_groups_0, pad = conv1d_21_pad_0, pad_type = conv1d_21_pad_type_0, strides = conv1d_21_strides_0, weight = p_encoder_layers_6_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_6_cast_fp16)[name = tensor<string, []>("conv1d_21_cast_fp16")];
            tensor<int32, [3]> transpose_105_perm_0 = const()[name = tensor<string, []>("transpose_105_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_39_axes_0 = const()[name = tensor<string, []>("layer_norm_39_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_6_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93471040)))];
            tensor<fp16, [768]> p_encoder_layers_6_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93472640)))];
            tensor<fp16, []> layer_norm_39_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_39_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_105_cast_fp16 = transpose(perm = transpose_105_perm_0, x = conv1d_21_cast_fp16)[name = tensor<string, []>("transpose_189")];
            tensor<fp16, [1, 750, 768]> layer_norm_39_cast_fp16 = layer_norm(axes = layer_norm_39_axes_0, beta = p_encoder_layers_6_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_39_epsilon_0_to_fp16, gamma = p_encoder_layers_6_conv_batch_norm_weight_to_fp16, x = transpose_105_cast_fp16)[name = tensor<string, []>("layer_norm_39_cast_fp16")];
            tensor<int32, [3]> transpose_106_perm_0 = const()[name = tensor<string, []>("transpose_106_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_106_cast_fp16 = transpose(perm = transpose_106_perm_0, x = layer_norm_39_cast_fp16)[name = tensor<string, []>("transpose_188")];
            tensor<fp16, [1, 768, 750]> silu_19_cast_fp16 = silu(x = transpose_106_cast_fp16)[name = tensor<string, []>("silu_19_cast_fp16")];
            tensor<string, []> conv1d_22_pad_type_0 = const()[name = tensor<string, []>("conv1d_22_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_22_strides_0 = const()[name = tensor<string, []>("conv1d_22_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_22_pad_0 = const()[name = tensor<string, []>("conv1d_22_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_22_dilations_0 = const()[name = tensor<string, []>("conv1d_22_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_22_groups_0 = const()[name = tensor<string, []>("conv1d_22_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93474240))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94064128))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_6_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94065728)))];
            tensor<fp16, [1, 768, 750]> conv1d_22_cast_fp16 = conv(bias = p_encoder_layers_6_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_22_dilations_0, groups = conv1d_22_groups_0, pad = conv1d_22_pad_0, pad_type = conv1d_22_pad_type_0, strides = conv1d_22_strides_0, weight = p_encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_19_cast_fp16)[name = tensor<string, []>("conv1d_22_cast_fp16")];
            tensor<int32, [3]> transpose_107_perm_0 = const()[name = tensor<string, []>("transpose_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_107_cast_fp16 = transpose(perm = transpose_107_perm_0, x = conv1d_22_cast_fp16)[name = tensor<string, []>("transpose_187")];
            tensor<fp16, [1, 750, 768]> add_44_cast_fp16 = add(x = add_43_cast_fp16, y = transpose_107_cast_fp16)[name = tensor<string, []>("add_44_cast_fp16")];
            tensor<int32, [1]> layer_norm_40_axes_0 = const()[name = tensor<string, []>("layer_norm_40_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94067328)))];
            tensor<fp16, [768]> p_encoder_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94068928)))];
            tensor<fp16, []> layer_norm_40_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_40_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_40_cast_fp16 = layer_norm(axes = layer_norm_40_axes_0, beta = p_encoder_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_40_epsilon_0_to_fp16, gamma = p_encoder_layers_6_norm_feed_forward2_weight_to_fp16, x = add_44_cast_fp16)[name = tensor<string, []>("layer_norm_40_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_6_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94070528))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96429888))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_6_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96436096)))];
            tensor<fp16, [1, 750, 3072]> linear_54_cast_fp16 = linear(bias = p_encoder_layers_6_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_6_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_40_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_20_cast_fp16 = silu(x = linear_54_cast_fp16)[name = tensor<string, []>("silu_20_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_6_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_6_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96442304))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98801664))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_6_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98803264)))];
            tensor<fp16, [1, 750, 768]> linear_55_cast_fp16 = linear(bias = p_encoder_layers_6_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_6_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_20_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
            tensor<fp16, []> const_570_to_fp16 = const()[name = tensor<string, []>("const_570_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_41_cast_fp16 = mul(x = linear_55_cast_fp16, y = const_570_to_fp16)[name = tensor<string, []>("mul_41_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_45_cast_fp16 = add(x = add_44_cast_fp16, y = mul_41_cast_fp16)[name = tensor<string, []>("add_45_cast_fp16")];
            tensor<int32, [1]> layer_norm_41_axes_0 = const()[name = tensor<string, []>("layer_norm_41_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_6_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98804864)))];
            tensor<fp16, [768]> p_encoder_layers_6_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98806464)))];
            tensor<fp16, []> layer_norm_41_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_41_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_41_cast_fp16 = layer_norm(axes = layer_norm_41_axes_0, beta = p_encoder_layers_6_norm_out_bias_to_fp16, epsilon = layer_norm_41_epsilon_0_to_fp16, gamma = p_encoder_layers_6_norm_out_weight_to_fp16, x = add_45_cast_fp16)[name = tensor<string, []>("layer_norm_41_cast_fp16")];
            tensor<int32, [1]> layer_norm_42_axes_0 = const()[name = tensor<string, []>("layer_norm_42_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98808064)))];
            tensor<fp16, [768]> p_encoder_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98809664)))];
            tensor<fp16, []> layer_norm_42_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_42_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_42_cast_fp16 = layer_norm(axes = layer_norm_42_axes_0, beta = p_encoder_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_42_epsilon_0_to_fp16, gamma = p_encoder_layers_7_norm_feed_forward1_weight_to_fp16, x = layer_norm_41_cast_fp16)[name = tensor<string, []>("layer_norm_42_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_7_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98811264))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101170624))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_7_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101176832)))];
            tensor<fp16, [1, 750, 3072]> linear_56_cast_fp16 = linear(bias = p_encoder_layers_7_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_7_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_42_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_21_cast_fp16 = silu(x = linear_56_cast_fp16)[name = tensor<string, []>("silu_21_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_7_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101183040))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103542400))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_7_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103544000)))];
            tensor<fp16, [1, 750, 768]> linear_57_cast_fp16 = linear(bias = p_encoder_layers_7_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_7_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_21_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
            tensor<fp16, []> const_573_to_fp16 = const()[name = tensor<string, []>("const_573_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_42_cast_fp16 = mul(x = linear_57_cast_fp16, y = const_573_to_fp16)[name = tensor<string, []>("mul_42_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_46_cast_fp16 = add(x = layer_norm_41_cast_fp16, y = mul_42_cast_fp16)[name = tensor<string, []>("add_46_cast_fp16")];
            tensor<int32, [1]> layer_norm_43_axes_0 = const()[name = tensor<string, []>("layer_norm_43_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_7_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103545600)))];
            tensor<fp16, [768]> p_encoder_layers_7_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103547200)))];
            tensor<fp16, []> layer_norm_43_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_43_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_43_cast_fp16 = layer_norm(axes = layer_norm_43_axes_0, beta = p_encoder_layers_7_norm_self_att_bias_to_fp16, epsilon = layer_norm_43_epsilon_0_to_fp16, gamma = p_encoder_layers_7_norm_self_att_weight_to_fp16, x = add_46_cast_fp16)[name = tensor<string, []>("layer_norm_43_cast_fp16")];
            tensor<int32, [4]> const_577 = const()[name = tensor<string, []>("const_577"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_55_perm_1 = const()[name = tensor<string, []>("transpose_55_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_55 = transpose(perm = transpose_55_perm_1, x = layer_norm_43_cast_fp16)[name = tensor<string, []>("transpose_186")];
            tensor<fp16, [750, 1, 16, 48]> view_63_cast_fp16 = reshape(shape = const_577, x = transpose_55)[name = tensor<string, []>("view_63_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_43_cast_fp16 = mul(x = view_63_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_43_cast_fp16")];
            tensor<int32, [4]> slice_31_begin_0 = const()[name = tensor<string, []>("slice_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_31_end_0 = const()[name = tensor<string, []>("slice_31_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_31_end_mask_0 = const()[name = tensor<string, []>("slice_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_31_cast_fp16 = slice_by_index(begin = slice_31_begin_0, end = slice_31_end_0, end_mask = slice_31_end_mask_0, x = view_63_cast_fp16)[name = tensor<string, []>("slice_31_cast_fp16")];
            tensor<int32, [4]> slice_32_begin_0 = const()[name = tensor<string, []>("slice_32_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_32_end_0 = const()[name = tensor<string, []>("slice_32_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_32_end_mask_0 = const()[name = tensor<string, []>("slice_32_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_32_cast_fp16 = slice_by_index(begin = slice_32_begin_0, end = slice_32_end_0, end_mask = slice_32_end_mask_0, x = view_63_cast_fp16)[name = tensor<string, []>("slice_32_cast_fp16")];
            tensor<fp16, []> const_594_promoted_to_fp16 = const()[name = tensor<string, []>("const_594_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_14_cast_fp16 = mul(x = slice_32_cast_fp16, y = const_594_promoted_to_fp16)[name = tensor<string, []>("neg_14_cast_fp16")];
            tensor<int32, []> const_595 = const()[name = tensor<string, []>("const_595"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_14_interleave_0 = const()[name = tensor<string, []>("cat_14_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_14_cast_fp16 = concat(axis = const_595, interleave = cat_14_interleave_0, values = (neg_14_cast_fp16, slice_31_cast_fp16))[name = tensor<string, []>("cat_14_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_44_cast_fp16 = mul(x = cat_14_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_44_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_47_cast_fp16 = add(x = mul_43_cast_fp16, y = mul_44_cast_fp16)[name = tensor<string, []>("add_47_cast_fp16")];
            tensor<int32, [3]> const_604 = const()[name = tensor<string, []>("const_604"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_66_cast_fp16 = reshape(shape = const_604, x = add_47_cast_fp16)[name = tensor<string, []>("view_66_cast_fp16")];
            tensor<int32, [3]> transpose_111_perm_0 = const()[name = tensor<string, []>("transpose_111_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_7_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103548800))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104138688))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_7_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104140288)))];
            tensor<fp16, [1, 750, 768]> transpose_111_cast_fp16 = transpose(perm = transpose_111_perm_0, x = view_66_cast_fp16)[name = tensor<string, []>("transpose_185")];
            tensor<fp16, [1, 750, 768]> linear_58_cast_fp16 = linear(bias = p_encoder_layers_7_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_7_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_111_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
            tensor<int32, [4]> const_613 = const()[name = tensor<string, []>("const_613"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_69_cast_fp16 = reshape(shape = const_613, x = linear_58_cast_fp16)[name = tensor<string, []>("view_69_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_7_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104141888))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104731776))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_7_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104733376)))];
            tensor<fp16, [1, 750, 768]> linear_59_cast_fp16 = linear(bias = p_encoder_layers_7_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_7_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_111_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
            tensor<int32, [4]> const_614 = const()[name = tensor<string, []>("const_614"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_70_cast_fp16 = reshape(shape = const_614, x = linear_59_cast_fp16)[name = tensor<string, []>("view_70_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_7_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104734976))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105324864))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_7_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105326464)))];
            tensor<fp16, [1, 750, 768]> linear_60_cast_fp16 = linear(bias = p_encoder_layers_7_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_7_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_43_cast_fp16)[name = tensor<string, []>("linear_60_cast_fp16")];
            tensor<int32, [4]> const_615 = const()[name = tensor<string, []>("const_615"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_71_cast_fp16 = reshape(shape = const_615, x = linear_60_cast_fp16)[name = tensor<string, []>("view_71_cast_fp16")];
            tensor<int32, [4]> transpose_116_perm_0 = const()[name = tensor<string, []>("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_9_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_9_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_9_cast_fp16 = mul(x = view_70_cast_fp16, y = _inversed_div_9_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_9_cast_fp16")];
            tensor<bool, []> matmul_14_transpose_x_0 = const()[name = tensor<string, []>("matmul_14_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_14_transpose_y_0 = const()[name = tensor<string, []>("matmul_14_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_78_perm_0 = const()[name = tensor<string, []>("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_79_perm_0 = const()[name = tensor<string, []>("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_79 = transpose(perm = transpose_79_perm_0, x = _inversed_div_9_cast_fp16)[name = tensor<string, []>("transpose_183")];
            tensor<fp16, [1, 16, 750, 48]> transpose_78 = transpose(perm = transpose_78_perm_0, x = view_69_cast_fp16)[name = tensor<string, []>("transpose_184")];
            tensor<fp16, [1, 16, 750, 750]> matmul_14_cast_fp16 = matmul(transpose_x = matmul_14_transpose_x_0, transpose_y = matmul_14_transpose_y_0, x = transpose_78, y = transpose_79)[name = tensor<string, []>("matmul_14_cast_fp16")];
            tensor<int32, []> const_625 = const()[name = tensor<string, []>("const_625"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_7_cast_fp16 = softmax(axis = const_625, x = matmul_14_cast_fp16)[name = tensor<string, []>("softmax_7_cast_fp16")];
            tensor<bool, []> matmul_15_transpose_x_0 = const()[name = tensor<string, []>("matmul_15_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_15_transpose_y_0 = const()[name = tensor<string, []>("matmul_15_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_116_cast_fp16 = transpose(perm = transpose_116_perm_0, x = view_71_cast_fp16)[name = tensor<string, []>("transpose_182")];
            tensor<fp16, [1, 16, 750, 48]> matmul_15_cast_fp16 = matmul(transpose_x = matmul_15_transpose_x_0, transpose_y = matmul_15_transpose_y_0, x = softmax_7_cast_fp16, y = transpose_116_cast_fp16)[name = tensor<string, []>("matmul_15_cast_fp16")];
            tensor<int32, [4]> transpose_118_perm_0 = const()[name = tensor<string, []>("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_628 = const()[name = tensor<string, []>("const_628"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_118_cast_fp16 = transpose(perm = transpose_118_perm_0, x = matmul_15_cast_fp16)[name = tensor<string, []>("transpose_181")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_7_cast_fp16 = reshape(shape = const_628, x = transpose_118_cast_fp16)[name = tensor<string, []>("_unsafe_view_7_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_7_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105328064))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105917952))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_7_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105919552)))];
            tensor<fp16, [1, 750, 768]> linear_61_cast_fp16 = linear(bias = p_encoder_layers_7_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_7_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_7_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_49_cast_fp16 = add(x = add_46_cast_fp16, y = linear_61_cast_fp16)[name = tensor<string, []>("add_49_cast_fp16")];
            tensor<int32, [1]> layer_norm_44_axes_0 = const()[name = tensor<string, []>("layer_norm_44_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_7_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105921152)))];
            tensor<fp16, [768]> p_encoder_layers_7_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105922752)))];
            tensor<fp16, []> layer_norm_44_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_44_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_44_cast_fp16 = layer_norm(axes = layer_norm_44_axes_0, beta = p_encoder_layers_7_norm_conv_bias_to_fp16, epsilon = layer_norm_44_epsilon_0_to_fp16, gamma = p_encoder_layers_7_norm_conv_weight_to_fp16, x = add_49_cast_fp16)[name = tensor<string, []>("layer_norm_44_cast_fp16")];
            tensor<int32, [3]> transpose_119_perm_0 = const()[name = tensor<string, []>("transpose_119_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_23_pad_type_0 = const()[name = tensor<string, []>("conv1d_23_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_23_strides_0 = const()[name = tensor<string, []>("conv1d_23_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_23_pad_0 = const()[name = tensor<string, []>("conv1d_23_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_23_dilations_0 = const()[name = tensor<string, []>("conv1d_23_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_23_groups_0 = const()[name = tensor<string, []>("conv1d_23_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105924352))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107104064))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_7_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107107200)))];
            tensor<fp16, [1, 768, 750]> transpose_119_cast_fp16 = transpose(perm = transpose_119_perm_0, x = layer_norm_44_cast_fp16)[name = tensor<string, []>("transpose_180")];
            tensor<fp16, [1, 1536, 750]> conv1d_23_cast_fp16 = conv(bias = p_encoder_layers_7_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_23_dilations_0, groups = conv1d_23_groups_0, pad = conv1d_23_pad_0, pad_type = conv1d_23_pad_type_0, strides = conv1d_23_strides_0, weight = p_encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_119_cast_fp16)[name = tensor<string, []>("conv1d_23_cast_fp16")];
            tensor<int32, []> glu_7_split_num_splits_0 = const()[name = tensor<string, []>("glu_7_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_7_split_axis_0 = const()[name = tensor<string, []>("glu_7_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_7_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_7_split_cast_fp16_1 = split(axis = glu_7_split_axis_0, num_splits = glu_7_split_num_splits_0, x = conv1d_23_cast_fp16)[name = tensor<string, []>("glu_7_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_7_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_7_split_cast_fp16_1)[name = tensor<string, []>("glu_7_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_7_cast_fp16 = mul(x = glu_7_split_cast_fp16_0, y = glu_7_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_7_cast_fp16")];
            tensor<string, []> conv1d_24_pad_type_0 = const()[name = tensor<string, []>("conv1d_24_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_24_pad_0 = const()[name = tensor<string, []>("conv1d_24_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_24_groups_0 = const()[name = tensor<string, []>("conv1d_24_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_24_strides_0 = const()[name = tensor<string, []>("conv1d_24_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_24_dilations_0 = const()[name = tensor<string, []>("conv1d_24_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_7_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107110336))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107114240))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_7_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107115840)))];
            tensor<fp16, [1, 768, 750]> conv1d_24_cast_fp16 = conv(bias = p_encoder_layers_7_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_24_dilations_0, groups = conv1d_24_groups_0, pad = conv1d_24_pad_0, pad_type = conv1d_24_pad_type_0, strides = conv1d_24_strides_0, weight = p_encoder_layers_7_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_7_cast_fp16)[name = tensor<string, []>("conv1d_24_cast_fp16")];
            tensor<int32, [3]> transpose_120_perm_0 = const()[name = tensor<string, []>("transpose_120_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_45_axes_0 = const()[name = tensor<string, []>("layer_norm_45_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_7_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107117440)))];
            tensor<fp16, [768]> p_encoder_layers_7_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107119040)))];
            tensor<fp16, []> layer_norm_45_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_45_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_120_cast_fp16 = transpose(perm = transpose_120_perm_0, x = conv1d_24_cast_fp16)[name = tensor<string, []>("transpose_179")];
            tensor<fp16, [1, 750, 768]> layer_norm_45_cast_fp16 = layer_norm(axes = layer_norm_45_axes_0, beta = p_encoder_layers_7_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_45_epsilon_0_to_fp16, gamma = p_encoder_layers_7_conv_batch_norm_weight_to_fp16, x = transpose_120_cast_fp16)[name = tensor<string, []>("layer_norm_45_cast_fp16")];
            tensor<int32, [3]> transpose_121_perm_0 = const()[name = tensor<string, []>("transpose_121_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_121_cast_fp16 = transpose(perm = transpose_121_perm_0, x = layer_norm_45_cast_fp16)[name = tensor<string, []>("transpose_178")];
            tensor<fp16, [1, 768, 750]> silu_22_cast_fp16 = silu(x = transpose_121_cast_fp16)[name = tensor<string, []>("silu_22_cast_fp16")];
            tensor<string, []> conv1d_25_pad_type_0 = const()[name = tensor<string, []>("conv1d_25_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_25_strides_0 = const()[name = tensor<string, []>("conv1d_25_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_25_pad_0 = const()[name = tensor<string, []>("conv1d_25_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_25_dilations_0 = const()[name = tensor<string, []>("conv1d_25_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_25_groups_0 = const()[name = tensor<string, []>("conv1d_25_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107120640))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107710528))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_7_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107712128)))];
            tensor<fp16, [1, 768, 750]> conv1d_25_cast_fp16 = conv(bias = p_encoder_layers_7_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_25_dilations_0, groups = conv1d_25_groups_0, pad = conv1d_25_pad_0, pad_type = conv1d_25_pad_type_0, strides = conv1d_25_strides_0, weight = p_encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_22_cast_fp16)[name = tensor<string, []>("conv1d_25_cast_fp16")];
            tensor<int32, [3]> transpose_122_perm_0 = const()[name = tensor<string, []>("transpose_122_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_122_cast_fp16 = transpose(perm = transpose_122_perm_0, x = conv1d_25_cast_fp16)[name = tensor<string, []>("transpose_177")];
            tensor<fp16, [1, 750, 768]> add_50_cast_fp16 = add(x = add_49_cast_fp16, y = transpose_122_cast_fp16)[name = tensor<string, []>("add_50_cast_fp16")];
            tensor<int32, [1]> layer_norm_46_axes_0 = const()[name = tensor<string, []>("layer_norm_46_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107713728)))];
            tensor<fp16, [768]> p_encoder_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107715328)))];
            tensor<fp16, []> layer_norm_46_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_46_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_46_cast_fp16 = layer_norm(axes = layer_norm_46_axes_0, beta = p_encoder_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_46_epsilon_0_to_fp16, gamma = p_encoder_layers_7_norm_feed_forward2_weight_to_fp16, x = add_50_cast_fp16)[name = tensor<string, []>("layer_norm_46_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_7_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107716928))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110076288))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_7_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110082496)))];
            tensor<fp16, [1, 750, 3072]> linear_62_cast_fp16 = linear(bias = p_encoder_layers_7_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_7_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_46_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_23_cast_fp16 = silu(x = linear_62_cast_fp16)[name = tensor<string, []>("silu_23_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_7_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_7_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110088704))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112448064))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_7_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112449664)))];
            tensor<fp16, [1, 750, 768]> linear_63_cast_fp16 = linear(bias = p_encoder_layers_7_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_7_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_23_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
            tensor<fp16, []> const_647_to_fp16 = const()[name = tensor<string, []>("const_647_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_47_cast_fp16 = mul(x = linear_63_cast_fp16, y = const_647_to_fp16)[name = tensor<string, []>("mul_47_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_51_cast_fp16 = add(x = add_50_cast_fp16, y = mul_47_cast_fp16)[name = tensor<string, []>("add_51_cast_fp16")];
            tensor<int32, [1]> layer_norm_47_axes_0 = const()[name = tensor<string, []>("layer_norm_47_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_7_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112451264)))];
            tensor<fp16, [768]> p_encoder_layers_7_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112452864)))];
            tensor<fp16, []> layer_norm_47_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_47_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_47_cast_fp16 = layer_norm(axes = layer_norm_47_axes_0, beta = p_encoder_layers_7_norm_out_bias_to_fp16, epsilon = layer_norm_47_epsilon_0_to_fp16, gamma = p_encoder_layers_7_norm_out_weight_to_fp16, x = add_51_cast_fp16)[name = tensor<string, []>("layer_norm_47_cast_fp16")];
            tensor<int32, [1]> layer_norm_48_axes_0 = const()[name = tensor<string, []>("layer_norm_48_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112454464)))];
            tensor<fp16, [768]> p_encoder_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112456064)))];
            tensor<fp16, []> layer_norm_48_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_48_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_48_cast_fp16 = layer_norm(axes = layer_norm_48_axes_0, beta = p_encoder_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_48_epsilon_0_to_fp16, gamma = p_encoder_layers_8_norm_feed_forward1_weight_to_fp16, x = layer_norm_47_cast_fp16)[name = tensor<string, []>("layer_norm_48_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_8_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(112457664))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114817024))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_8_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114823232)))];
            tensor<fp16, [1, 750, 3072]> linear_64_cast_fp16 = linear(bias = p_encoder_layers_8_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_8_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_48_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_24_cast_fp16 = silu(x = linear_64_cast_fp16)[name = tensor<string, []>("silu_24_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_8_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114829440))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117188800))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_8_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117190400)))];
            tensor<fp16, [1, 750, 768]> linear_65_cast_fp16 = linear(bias = p_encoder_layers_8_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_8_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_24_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
            tensor<fp16, []> const_650_to_fp16 = const()[name = tensor<string, []>("const_650_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_48_cast_fp16 = mul(x = linear_65_cast_fp16, y = const_650_to_fp16)[name = tensor<string, []>("mul_48_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_52_cast_fp16 = add(x = layer_norm_47_cast_fp16, y = mul_48_cast_fp16)[name = tensor<string, []>("add_52_cast_fp16")];
            tensor<int32, [1]> layer_norm_49_axes_0 = const()[name = tensor<string, []>("layer_norm_49_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_8_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117192000)))];
            tensor<fp16, [768]> p_encoder_layers_8_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117193600)))];
            tensor<fp16, []> layer_norm_49_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_49_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_49_cast_fp16 = layer_norm(axes = layer_norm_49_axes_0, beta = p_encoder_layers_8_norm_self_att_bias_to_fp16, epsilon = layer_norm_49_epsilon_0_to_fp16, gamma = p_encoder_layers_8_norm_self_att_weight_to_fp16, x = add_52_cast_fp16)[name = tensor<string, []>("layer_norm_49_cast_fp16")];
            tensor<int32, [4]> const_654 = const()[name = tensor<string, []>("const_654"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_56_perm_1 = const()[name = tensor<string, []>("transpose_56_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_56 = transpose(perm = transpose_56_perm_1, x = layer_norm_49_cast_fp16)[name = tensor<string, []>("transpose_176")];
            tensor<fp16, [750, 1, 16, 48]> view_72_cast_fp16 = reshape(shape = const_654, x = transpose_56)[name = tensor<string, []>("view_72_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_49_cast_fp16 = mul(x = view_72_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_49_cast_fp16")];
            tensor<int32, [4]> slice_35_begin_0 = const()[name = tensor<string, []>("slice_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_35_end_0 = const()[name = tensor<string, []>("slice_35_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_35_end_mask_0 = const()[name = tensor<string, []>("slice_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_35_cast_fp16 = slice_by_index(begin = slice_35_begin_0, end = slice_35_end_0, end_mask = slice_35_end_mask_0, x = view_72_cast_fp16)[name = tensor<string, []>("slice_35_cast_fp16")];
            tensor<int32, [4]> slice_36_begin_0 = const()[name = tensor<string, []>("slice_36_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_36_end_0 = const()[name = tensor<string, []>("slice_36_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_36_end_mask_0 = const()[name = tensor<string, []>("slice_36_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_36_cast_fp16 = slice_by_index(begin = slice_36_begin_0, end = slice_36_end_0, end_mask = slice_36_end_mask_0, x = view_72_cast_fp16)[name = tensor<string, []>("slice_36_cast_fp16")];
            tensor<fp16, []> const_671_promoted_to_fp16 = const()[name = tensor<string, []>("const_671_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_16_cast_fp16 = mul(x = slice_36_cast_fp16, y = const_671_promoted_to_fp16)[name = tensor<string, []>("neg_16_cast_fp16")];
            tensor<int32, []> const_672 = const()[name = tensor<string, []>("const_672"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_16_interleave_0 = const()[name = tensor<string, []>("cat_16_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_16_cast_fp16 = concat(axis = const_672, interleave = cat_16_interleave_0, values = (neg_16_cast_fp16, slice_35_cast_fp16))[name = tensor<string, []>("cat_16_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_50_cast_fp16 = mul(x = cat_16_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_50_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_53_cast_fp16 = add(x = mul_49_cast_fp16, y = mul_50_cast_fp16)[name = tensor<string, []>("add_53_cast_fp16")];
            tensor<int32, [3]> const_681 = const()[name = tensor<string, []>("const_681"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_75_cast_fp16 = reshape(shape = const_681, x = add_53_cast_fp16)[name = tensor<string, []>("view_75_cast_fp16")];
            tensor<int32, [3]> transpose_126_perm_0 = const()[name = tensor<string, []>("transpose_126_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_8_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117195200))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117785088))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_8_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117786688)))];
            tensor<fp16, [1, 750, 768]> transpose_126_cast_fp16 = transpose(perm = transpose_126_perm_0, x = view_75_cast_fp16)[name = tensor<string, []>("transpose_175")];
            tensor<fp16, [1, 750, 768]> linear_66_cast_fp16 = linear(bias = p_encoder_layers_8_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_8_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_126_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<int32, [4]> const_690 = const()[name = tensor<string, []>("const_690"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_78_cast_fp16 = reshape(shape = const_690, x = linear_66_cast_fp16)[name = tensor<string, []>("view_78_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_8_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(117788288))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118378176))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_8_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118379776)))];
            tensor<fp16, [1, 750, 768]> linear_67_cast_fp16 = linear(bias = p_encoder_layers_8_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_8_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_126_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<int32, [4]> const_691 = const()[name = tensor<string, []>("const_691"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_79_cast_fp16 = reshape(shape = const_691, x = linear_67_cast_fp16)[name = tensor<string, []>("view_79_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_8_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118381376))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118971264))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_8_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118972864)))];
            tensor<fp16, [1, 750, 768]> linear_68_cast_fp16 = linear(bias = p_encoder_layers_8_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_8_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_49_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<int32, [4]> const_692 = const()[name = tensor<string, []>("const_692"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_80_cast_fp16 = reshape(shape = const_692, x = linear_68_cast_fp16)[name = tensor<string, []>("view_80_cast_fp16")];
            tensor<int32, [4]> transpose_131_perm_0 = const()[name = tensor<string, []>("transpose_131_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_10_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_10_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_10_cast_fp16 = mul(x = view_79_cast_fp16, y = _inversed_div_10_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_10_cast_fp16")];
            tensor<bool, []> matmul_16_transpose_x_0 = const()[name = tensor<string, []>("matmul_16_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_16_transpose_y_0 = const()[name = tensor<string, []>("matmul_16_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_80_perm_0 = const()[name = tensor<string, []>("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_81_perm_0_1 = const()[name = tensor<string, []>("transpose_81_perm_0_1"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_81 = transpose(perm = transpose_81_perm_0_1, x = _inversed_div_10_cast_fp16)[name = tensor<string, []>("transpose_173")];
            tensor<fp16, [1, 16, 750, 48]> transpose_80 = transpose(perm = transpose_80_perm_0, x = view_78_cast_fp16)[name = tensor<string, []>("transpose_174")];
            tensor<fp16, [1, 16, 750, 750]> matmul_16_cast_fp16 = matmul(transpose_x = matmul_16_transpose_x_0, transpose_y = matmul_16_transpose_y_0, x = transpose_80, y = transpose_81)[name = tensor<string, []>("matmul_16_cast_fp16")];
            tensor<int32, []> const_702 = const()[name = tensor<string, []>("const_702"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_8_cast_fp16 = softmax(axis = const_702, x = matmul_16_cast_fp16)[name = tensor<string, []>("softmax_8_cast_fp16")];
            tensor<bool, []> matmul_17_transpose_x_0 = const()[name = tensor<string, []>("matmul_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_17_transpose_y_0 = const()[name = tensor<string, []>("matmul_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_131_cast_fp16 = transpose(perm = transpose_131_perm_0, x = view_80_cast_fp16)[name = tensor<string, []>("transpose_172")];
            tensor<fp16, [1, 16, 750, 48]> matmul_17_cast_fp16 = matmul(transpose_x = matmul_17_transpose_x_0, transpose_y = matmul_17_transpose_y_0, x = softmax_8_cast_fp16, y = transpose_131_cast_fp16)[name = tensor<string, []>("matmul_17_cast_fp16")];
            tensor<int32, [4]> transpose_133_perm_0 = const()[name = tensor<string, []>("transpose_133_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_705 = const()[name = tensor<string, []>("const_705"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_133_cast_fp16 = transpose(perm = transpose_133_perm_0, x = matmul_17_cast_fp16)[name = tensor<string, []>("transpose_171")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_8_cast_fp16 = reshape(shape = const_705, x = transpose_133_cast_fp16)[name = tensor<string, []>("_unsafe_view_8_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_8_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(118974464))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119564352))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_8_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119565952)))];
            tensor<fp16, [1, 750, 768]> linear_69_cast_fp16 = linear(bias = p_encoder_layers_8_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_8_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_8_cast_fp16)[name = tensor<string, []>("linear_69_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_55_cast_fp16 = add(x = add_52_cast_fp16, y = linear_69_cast_fp16)[name = tensor<string, []>("add_55_cast_fp16")];
            tensor<int32, [1]> layer_norm_50_axes_0 = const()[name = tensor<string, []>("layer_norm_50_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_8_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119567552)))];
            tensor<fp16, [768]> p_encoder_layers_8_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119569152)))];
            tensor<fp16, []> layer_norm_50_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_50_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_50_cast_fp16 = layer_norm(axes = layer_norm_50_axes_0, beta = p_encoder_layers_8_norm_conv_bias_to_fp16, epsilon = layer_norm_50_epsilon_0_to_fp16, gamma = p_encoder_layers_8_norm_conv_weight_to_fp16, x = add_55_cast_fp16)[name = tensor<string, []>("layer_norm_50_cast_fp16")];
            tensor<int32, [3]> transpose_134_perm_0 = const()[name = tensor<string, []>("transpose_134_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_26_pad_type_0 = const()[name = tensor<string, []>("conv1d_26_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_26_strides_0 = const()[name = tensor<string, []>("conv1d_26_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_26_pad_0 = const()[name = tensor<string, []>("conv1d_26_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_26_dilations_0 = const()[name = tensor<string, []>("conv1d_26_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_26_groups_0 = const()[name = tensor<string, []>("conv1d_26_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119570752))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120750464))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_8_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120753600)))];
            tensor<fp16, [1, 768, 750]> transpose_134_cast_fp16 = transpose(perm = transpose_134_perm_0, x = layer_norm_50_cast_fp16)[name = tensor<string, []>("transpose_170")];
            tensor<fp16, [1, 1536, 750]> conv1d_26_cast_fp16 = conv(bias = p_encoder_layers_8_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_26_dilations_0, groups = conv1d_26_groups_0, pad = conv1d_26_pad_0, pad_type = conv1d_26_pad_type_0, strides = conv1d_26_strides_0, weight = p_encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_134_cast_fp16)[name = tensor<string, []>("conv1d_26_cast_fp16")];
            tensor<int32, []> glu_8_split_num_splits_0 = const()[name = tensor<string, []>("glu_8_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_8_split_axis_0 = const()[name = tensor<string, []>("glu_8_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_8_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_8_split_cast_fp16_1 = split(axis = glu_8_split_axis_0, num_splits = glu_8_split_num_splits_0, x = conv1d_26_cast_fp16)[name = tensor<string, []>("glu_8_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_8_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_8_split_cast_fp16_1)[name = tensor<string, []>("glu_8_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_8_cast_fp16 = mul(x = glu_8_split_cast_fp16_0, y = glu_8_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_8_cast_fp16")];
            tensor<string, []> conv1d_27_pad_type_0 = const()[name = tensor<string, []>("conv1d_27_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_27_pad_0 = const()[name = tensor<string, []>("conv1d_27_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_27_groups_0 = const()[name = tensor<string, []>("conv1d_27_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_27_strides_0 = const()[name = tensor<string, []>("conv1d_27_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_27_dilations_0 = const()[name = tensor<string, []>("conv1d_27_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_8_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120756736))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120760640))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_8_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120762240)))];
            tensor<fp16, [1, 768, 750]> conv1d_27_cast_fp16 = conv(bias = p_encoder_layers_8_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_27_dilations_0, groups = conv1d_27_groups_0, pad = conv1d_27_pad_0, pad_type = conv1d_27_pad_type_0, strides = conv1d_27_strides_0, weight = p_encoder_layers_8_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_8_cast_fp16)[name = tensor<string, []>("conv1d_27_cast_fp16")];
            tensor<int32, [3]> transpose_135_perm_0 = const()[name = tensor<string, []>("transpose_135_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_51_axes_0 = const()[name = tensor<string, []>("layer_norm_51_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_8_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120763840)))];
            tensor<fp16, [768]> p_encoder_layers_8_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120765440)))];
            tensor<fp16, []> layer_norm_51_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_51_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_135_cast_fp16 = transpose(perm = transpose_135_perm_0, x = conv1d_27_cast_fp16)[name = tensor<string, []>("transpose_169")];
            tensor<fp16, [1, 750, 768]> layer_norm_51_cast_fp16 = layer_norm(axes = layer_norm_51_axes_0, beta = p_encoder_layers_8_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_51_epsilon_0_to_fp16, gamma = p_encoder_layers_8_conv_batch_norm_weight_to_fp16, x = transpose_135_cast_fp16)[name = tensor<string, []>("layer_norm_51_cast_fp16")];
            tensor<int32, [3]> transpose_136_perm_0 = const()[name = tensor<string, []>("transpose_136_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_136_cast_fp16 = transpose(perm = transpose_136_perm_0, x = layer_norm_51_cast_fp16)[name = tensor<string, []>("transpose_168")];
            tensor<fp16, [1, 768, 750]> silu_25_cast_fp16 = silu(x = transpose_136_cast_fp16)[name = tensor<string, []>("silu_25_cast_fp16")];
            tensor<string, []> conv1d_28_pad_type_0 = const()[name = tensor<string, []>("conv1d_28_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_28_strides_0 = const()[name = tensor<string, []>("conv1d_28_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_28_pad_0 = const()[name = tensor<string, []>("conv1d_28_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_28_dilations_0 = const()[name = tensor<string, []>("conv1d_28_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_28_groups_0 = const()[name = tensor<string, []>("conv1d_28_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120767040))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121356928))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_8_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121358528)))];
            tensor<fp16, [1, 768, 750]> conv1d_28_cast_fp16 = conv(bias = p_encoder_layers_8_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_28_dilations_0, groups = conv1d_28_groups_0, pad = conv1d_28_pad_0, pad_type = conv1d_28_pad_type_0, strides = conv1d_28_strides_0, weight = p_encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_25_cast_fp16)[name = tensor<string, []>("conv1d_28_cast_fp16")];
            tensor<int32, [3]> transpose_137_perm_0 = const()[name = tensor<string, []>("transpose_137_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_137_cast_fp16 = transpose(perm = transpose_137_perm_0, x = conv1d_28_cast_fp16)[name = tensor<string, []>("transpose_167")];
            tensor<fp16, [1, 750, 768]> add_56_cast_fp16 = add(x = add_55_cast_fp16, y = transpose_137_cast_fp16)[name = tensor<string, []>("add_56_cast_fp16")];
            tensor<int32, [1]> layer_norm_52_axes_0 = const()[name = tensor<string, []>("layer_norm_52_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121360128)))];
            tensor<fp16, [768]> p_encoder_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121361728)))];
            tensor<fp16, []> layer_norm_52_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_52_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_52_cast_fp16 = layer_norm(axes = layer_norm_52_axes_0, beta = p_encoder_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_52_epsilon_0_to_fp16, gamma = p_encoder_layers_8_norm_feed_forward2_weight_to_fp16, x = add_56_cast_fp16)[name = tensor<string, []>("layer_norm_52_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_8_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121363328))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123722688))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_8_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123728896)))];
            tensor<fp16, [1, 750, 3072]> linear_70_cast_fp16 = linear(bias = p_encoder_layers_8_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_8_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_52_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_26_cast_fp16 = silu(x = linear_70_cast_fp16)[name = tensor<string, []>("silu_26_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_8_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_8_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123735104))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126094464))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_8_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126096064)))];
            tensor<fp16, [1, 750, 768]> linear_71_cast_fp16 = linear(bias = p_encoder_layers_8_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_8_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_26_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
            tensor<fp16, []> const_724_to_fp16 = const()[name = tensor<string, []>("const_724_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_53_cast_fp16 = mul(x = linear_71_cast_fp16, y = const_724_to_fp16)[name = tensor<string, []>("mul_53_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_57_cast_fp16 = add(x = add_56_cast_fp16, y = mul_53_cast_fp16)[name = tensor<string, []>("add_57_cast_fp16")];
            tensor<int32, [1]> layer_norm_53_axes_0 = const()[name = tensor<string, []>("layer_norm_53_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_8_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126097664)))];
            tensor<fp16, [768]> p_encoder_layers_8_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126099264)))];
            tensor<fp16, []> layer_norm_53_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_53_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_53_cast_fp16 = layer_norm(axes = layer_norm_53_axes_0, beta = p_encoder_layers_8_norm_out_bias_to_fp16, epsilon = layer_norm_53_epsilon_0_to_fp16, gamma = p_encoder_layers_8_norm_out_weight_to_fp16, x = add_57_cast_fp16)[name = tensor<string, []>("layer_norm_53_cast_fp16")];
            tensor<int32, [1]> layer_norm_54_axes_0 = const()[name = tensor<string, []>("layer_norm_54_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126100864)))];
            tensor<fp16, [768]> p_encoder_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126102464)))];
            tensor<fp16, []> layer_norm_54_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_54_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_54_cast_fp16 = layer_norm(axes = layer_norm_54_axes_0, beta = p_encoder_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_54_epsilon_0_to_fp16, gamma = p_encoder_layers_9_norm_feed_forward1_weight_to_fp16, x = layer_norm_53_cast_fp16)[name = tensor<string, []>("layer_norm_54_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_9_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(126104064))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128463424))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_9_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128469632)))];
            tensor<fp16, [1, 750, 3072]> linear_72_cast_fp16 = linear(bias = p_encoder_layers_9_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_9_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_54_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_27_cast_fp16 = silu(x = linear_72_cast_fp16)[name = tensor<string, []>("silu_27_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_9_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(128475840))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130835200))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_9_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130836800)))];
            tensor<fp16, [1, 750, 768]> linear_73_cast_fp16 = linear(bias = p_encoder_layers_9_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_9_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_27_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
            tensor<fp16, []> const_727_to_fp16 = const()[name = tensor<string, []>("const_727_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_54_cast_fp16 = mul(x = linear_73_cast_fp16, y = const_727_to_fp16)[name = tensor<string, []>("mul_54_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_58_cast_fp16 = add(x = layer_norm_53_cast_fp16, y = mul_54_cast_fp16)[name = tensor<string, []>("add_58_cast_fp16")];
            tensor<int32, [1]> layer_norm_55_axes_0 = const()[name = tensor<string, []>("layer_norm_55_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_9_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130838400)))];
            tensor<fp16, [768]> p_encoder_layers_9_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130840000)))];
            tensor<fp16, []> layer_norm_55_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_55_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_55_cast_fp16 = layer_norm(axes = layer_norm_55_axes_0, beta = p_encoder_layers_9_norm_self_att_bias_to_fp16, epsilon = layer_norm_55_epsilon_0_to_fp16, gamma = p_encoder_layers_9_norm_self_att_weight_to_fp16, x = add_58_cast_fp16)[name = tensor<string, []>("layer_norm_55_cast_fp16")];
            tensor<int32, [4]> const_731 = const()[name = tensor<string, []>("const_731"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_57_perm_1 = const()[name = tensor<string, []>("transpose_57_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_57 = transpose(perm = transpose_57_perm_1, x = layer_norm_55_cast_fp16)[name = tensor<string, []>("transpose_166")];
            tensor<fp16, [750, 1, 16, 48]> view_81_cast_fp16 = reshape(shape = const_731, x = transpose_57)[name = tensor<string, []>("view_81_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_55_cast_fp16 = mul(x = view_81_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_55_cast_fp16")];
            tensor<int32, [4]> slice_39_begin_0 = const()[name = tensor<string, []>("slice_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_39_end_0 = const()[name = tensor<string, []>("slice_39_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_39_end_mask_0 = const()[name = tensor<string, []>("slice_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_39_cast_fp16 = slice_by_index(begin = slice_39_begin_0, end = slice_39_end_0, end_mask = slice_39_end_mask_0, x = view_81_cast_fp16)[name = tensor<string, []>("slice_39_cast_fp16")];
            tensor<int32, [4]> slice_40_begin_0 = const()[name = tensor<string, []>("slice_40_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_40_end_0 = const()[name = tensor<string, []>("slice_40_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_40_end_mask_0 = const()[name = tensor<string, []>("slice_40_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_40_cast_fp16 = slice_by_index(begin = slice_40_begin_0, end = slice_40_end_0, end_mask = slice_40_end_mask_0, x = view_81_cast_fp16)[name = tensor<string, []>("slice_40_cast_fp16")];
            tensor<fp16, []> const_748_promoted_to_fp16 = const()[name = tensor<string, []>("const_748_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_18_cast_fp16 = mul(x = slice_40_cast_fp16, y = const_748_promoted_to_fp16)[name = tensor<string, []>("neg_18_cast_fp16")];
            tensor<int32, []> const_749 = const()[name = tensor<string, []>("const_749"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_18_interleave_0 = const()[name = tensor<string, []>("cat_18_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_18_cast_fp16 = concat(axis = const_749, interleave = cat_18_interleave_0, values = (neg_18_cast_fp16, slice_39_cast_fp16))[name = tensor<string, []>("cat_18_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_56_cast_fp16 = mul(x = cat_18_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_56_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_59_cast_fp16 = add(x = mul_55_cast_fp16, y = mul_56_cast_fp16)[name = tensor<string, []>("add_59_cast_fp16")];
            tensor<int32, [3]> const_758 = const()[name = tensor<string, []>("const_758"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_84_cast_fp16 = reshape(shape = const_758, x = add_59_cast_fp16)[name = tensor<string, []>("view_84_cast_fp16")];
            tensor<int32, [3]> transpose_141_perm_0 = const()[name = tensor<string, []>("transpose_141_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_9_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(130841600))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131431488))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_9_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131433088)))];
            tensor<fp16, [1, 750, 768]> transpose_141_cast_fp16 = transpose(perm = transpose_141_perm_0, x = view_84_cast_fp16)[name = tensor<string, []>("transpose_165")];
            tensor<fp16, [1, 750, 768]> linear_74_cast_fp16 = linear(bias = p_encoder_layers_9_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_9_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_141_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
            tensor<int32, [4]> const_767 = const()[name = tensor<string, []>("const_767"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_87_cast_fp16 = reshape(shape = const_767, x = linear_74_cast_fp16)[name = tensor<string, []>("view_87_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_9_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131434688))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132024576))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_9_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132026176)))];
            tensor<fp16, [1, 750, 768]> linear_75_cast_fp16 = linear(bias = p_encoder_layers_9_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_9_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_141_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")];
            tensor<int32, [4]> const_768 = const()[name = tensor<string, []>("const_768"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_88_cast_fp16 = reshape(shape = const_768, x = linear_75_cast_fp16)[name = tensor<string, []>("view_88_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_9_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132027776))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132617664))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_9_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132619264)))];
            tensor<fp16, [1, 750, 768]> linear_76_cast_fp16 = linear(bias = p_encoder_layers_9_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_9_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_55_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")];
            tensor<int32, [4]> const_769 = const()[name = tensor<string, []>("const_769"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_89_cast_fp16 = reshape(shape = const_769, x = linear_76_cast_fp16)[name = tensor<string, []>("view_89_cast_fp16")];
            tensor<int32, [4]> transpose_146_perm_0 = const()[name = tensor<string, []>("transpose_146_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_11_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_11_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_11_cast_fp16 = mul(x = view_88_cast_fp16, y = _inversed_div_11_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_11_cast_fp16")];
            tensor<bool, []> matmul_18_transpose_x_0 = const()[name = tensor<string, []>("matmul_18_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_18_transpose_y_0 = const()[name = tensor<string, []>("matmul_18_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_83 = transpose(perm = transpose_83_perm_0, x = _inversed_div_11_cast_fp16)[name = tensor<string, []>("transpose_163")];
            tensor<fp16, [1, 16, 750, 48]> transpose_82 = transpose(perm = transpose_82_perm_0, x = view_87_cast_fp16)[name = tensor<string, []>("transpose_164")];
            tensor<fp16, [1, 16, 750, 750]> matmul_18_cast_fp16 = matmul(transpose_x = matmul_18_transpose_x_0, transpose_y = matmul_18_transpose_y_0, x = transpose_82, y = transpose_83)[name = tensor<string, []>("matmul_18_cast_fp16")];
            tensor<int32, []> const_779 = const()[name = tensor<string, []>("const_779"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_9_cast_fp16 = softmax(axis = const_779, x = matmul_18_cast_fp16)[name = tensor<string, []>("softmax_9_cast_fp16")];
            tensor<bool, []> matmul_19_transpose_x_0 = const()[name = tensor<string, []>("matmul_19_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_19_transpose_y_0 = const()[name = tensor<string, []>("matmul_19_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_146_cast_fp16 = transpose(perm = transpose_146_perm_0, x = view_89_cast_fp16)[name = tensor<string, []>("transpose_162")];
            tensor<fp16, [1, 16, 750, 48]> matmul_19_cast_fp16 = matmul(transpose_x = matmul_19_transpose_x_0, transpose_y = matmul_19_transpose_y_0, x = softmax_9_cast_fp16, y = transpose_146_cast_fp16)[name = tensor<string, []>("matmul_19_cast_fp16")];
            tensor<int32, [4]> transpose_148_perm_0 = const()[name = tensor<string, []>("transpose_148_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_782 = const()[name = tensor<string, []>("const_782"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_148_cast_fp16 = transpose(perm = transpose_148_perm_0, x = matmul_19_cast_fp16)[name = tensor<string, []>("transpose_161")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_9_cast_fp16 = reshape(shape = const_782, x = transpose_148_cast_fp16)[name = tensor<string, []>("_unsafe_view_9_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_9_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132620864))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133210752))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_9_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133212352)))];
            tensor<fp16, [1, 750, 768]> linear_77_cast_fp16 = linear(bias = p_encoder_layers_9_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_9_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_9_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_61_cast_fp16 = add(x = add_58_cast_fp16, y = linear_77_cast_fp16)[name = tensor<string, []>("add_61_cast_fp16")];
            tensor<int32, [1]> layer_norm_56_axes_0 = const()[name = tensor<string, []>("layer_norm_56_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_9_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133213952)))];
            tensor<fp16, [768]> p_encoder_layers_9_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133215552)))];
            tensor<fp16, []> layer_norm_56_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_56_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_56_cast_fp16 = layer_norm(axes = layer_norm_56_axes_0, beta = p_encoder_layers_9_norm_conv_bias_to_fp16, epsilon = layer_norm_56_epsilon_0_to_fp16, gamma = p_encoder_layers_9_norm_conv_weight_to_fp16, x = add_61_cast_fp16)[name = tensor<string, []>("layer_norm_56_cast_fp16")];
            tensor<int32, [3]> transpose_149_perm_0 = const()[name = tensor<string, []>("transpose_149_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_29_pad_type_0 = const()[name = tensor<string, []>("conv1d_29_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_29_strides_0 = const()[name = tensor<string, []>("conv1d_29_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_29_pad_0 = const()[name = tensor<string, []>("conv1d_29_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_29_dilations_0 = const()[name = tensor<string, []>("conv1d_29_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_29_groups_0 = const()[name = tensor<string, []>("conv1d_29_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133217152))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134396864))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_9_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134400000)))];
            tensor<fp16, [1, 768, 750]> transpose_149_cast_fp16 = transpose(perm = transpose_149_perm_0, x = layer_norm_56_cast_fp16)[name = tensor<string, []>("transpose_160")];
            tensor<fp16, [1, 1536, 750]> conv1d_29_cast_fp16 = conv(bias = p_encoder_layers_9_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_29_dilations_0, groups = conv1d_29_groups_0, pad = conv1d_29_pad_0, pad_type = conv1d_29_pad_type_0, strides = conv1d_29_strides_0, weight = p_encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_149_cast_fp16)[name = tensor<string, []>("conv1d_29_cast_fp16")];
            tensor<int32, []> glu_9_split_num_splits_0 = const()[name = tensor<string, []>("glu_9_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_9_split_axis_0 = const()[name = tensor<string, []>("glu_9_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_9_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_9_split_cast_fp16_1 = split(axis = glu_9_split_axis_0, num_splits = glu_9_split_num_splits_0, x = conv1d_29_cast_fp16)[name = tensor<string, []>("glu_9_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_9_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_9_split_cast_fp16_1)[name = tensor<string, []>("glu_9_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_9_cast_fp16 = mul(x = glu_9_split_cast_fp16_0, y = glu_9_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_9_cast_fp16")];
            tensor<string, []> conv1d_30_pad_type_0 = const()[name = tensor<string, []>("conv1d_30_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_30_pad_0 = const()[name = tensor<string, []>("conv1d_30_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_30_groups_0 = const()[name = tensor<string, []>("conv1d_30_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_30_strides_0 = const()[name = tensor<string, []>("conv1d_30_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_30_dilations_0 = const()[name = tensor<string, []>("conv1d_30_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_9_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134403136))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134407040))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_9_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134408640)))];
            tensor<fp16, [1, 768, 750]> conv1d_30_cast_fp16 = conv(bias = p_encoder_layers_9_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_30_dilations_0, groups = conv1d_30_groups_0, pad = conv1d_30_pad_0, pad_type = conv1d_30_pad_type_0, strides = conv1d_30_strides_0, weight = p_encoder_layers_9_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_9_cast_fp16)[name = tensor<string, []>("conv1d_30_cast_fp16")];
            tensor<int32, [3]> transpose_150_perm_0 = const()[name = tensor<string, []>("transpose_150_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_57_axes_0 = const()[name = tensor<string, []>("layer_norm_57_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_9_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134410240)))];
            tensor<fp16, [768]> p_encoder_layers_9_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134411840)))];
            tensor<fp16, []> layer_norm_57_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_57_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_150_cast_fp16 = transpose(perm = transpose_150_perm_0, x = conv1d_30_cast_fp16)[name = tensor<string, []>("transpose_159")];
            tensor<fp16, [1, 750, 768]> layer_norm_57_cast_fp16 = layer_norm(axes = layer_norm_57_axes_0, beta = p_encoder_layers_9_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_57_epsilon_0_to_fp16, gamma = p_encoder_layers_9_conv_batch_norm_weight_to_fp16, x = transpose_150_cast_fp16)[name = tensor<string, []>("layer_norm_57_cast_fp16")];
            tensor<int32, [3]> transpose_151_perm_0 = const()[name = tensor<string, []>("transpose_151_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_151_cast_fp16 = transpose(perm = transpose_151_perm_0, x = layer_norm_57_cast_fp16)[name = tensor<string, []>("transpose_158")];
            tensor<fp16, [1, 768, 750]> silu_28_cast_fp16 = silu(x = transpose_151_cast_fp16)[name = tensor<string, []>("silu_28_cast_fp16")];
            tensor<string, []> conv1d_31_pad_type_0 = const()[name = tensor<string, []>("conv1d_31_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_31_strides_0 = const()[name = tensor<string, []>("conv1d_31_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_31_pad_0 = const()[name = tensor<string, []>("conv1d_31_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_31_dilations_0 = const()[name = tensor<string, []>("conv1d_31_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_31_groups_0 = const()[name = tensor<string, []>("conv1d_31_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134413440))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135003328))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_9_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135004928)))];
            tensor<fp16, [1, 768, 750]> conv1d_31_cast_fp16 = conv(bias = p_encoder_layers_9_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_31_dilations_0, groups = conv1d_31_groups_0, pad = conv1d_31_pad_0, pad_type = conv1d_31_pad_type_0, strides = conv1d_31_strides_0, weight = p_encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_28_cast_fp16)[name = tensor<string, []>("conv1d_31_cast_fp16")];
            tensor<int32, [3]> transpose_152_perm_0 = const()[name = tensor<string, []>("transpose_152_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_152_cast_fp16 = transpose(perm = transpose_152_perm_0, x = conv1d_31_cast_fp16)[name = tensor<string, []>("transpose_157")];
            tensor<fp16, [1, 750, 768]> add_62_cast_fp16 = add(x = add_61_cast_fp16, y = transpose_152_cast_fp16)[name = tensor<string, []>("add_62_cast_fp16")];
            tensor<int32, [1]> layer_norm_58_axes_0 = const()[name = tensor<string, []>("layer_norm_58_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135006528)))];
            tensor<fp16, [768]> p_encoder_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135008128)))];
            tensor<fp16, []> layer_norm_58_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_58_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_58_cast_fp16 = layer_norm(axes = layer_norm_58_axes_0, beta = p_encoder_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_58_epsilon_0_to_fp16, gamma = p_encoder_layers_9_norm_feed_forward2_weight_to_fp16, x = add_62_cast_fp16)[name = tensor<string, []>("layer_norm_58_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_9_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135009728))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137369088))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_9_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137375296)))];
            tensor<fp16, [1, 750, 3072]> linear_78_cast_fp16 = linear(bias = p_encoder_layers_9_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_9_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_58_cast_fp16)[name = tensor<string, []>("linear_78_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_29_cast_fp16 = silu(x = linear_78_cast_fp16)[name = tensor<string, []>("silu_29_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_9_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_9_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137381504))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139740864))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_9_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139742464)))];
            tensor<fp16, [1, 750, 768]> linear_79_cast_fp16 = linear(bias = p_encoder_layers_9_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_9_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_29_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")];
            tensor<fp16, []> const_801_to_fp16 = const()[name = tensor<string, []>("const_801_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_59_cast_fp16 = mul(x = linear_79_cast_fp16, y = const_801_to_fp16)[name = tensor<string, []>("mul_59_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_63_cast_fp16 = add(x = add_62_cast_fp16, y = mul_59_cast_fp16)[name = tensor<string, []>("add_63_cast_fp16")];
            tensor<int32, [1]> layer_norm_59_axes_0 = const()[name = tensor<string, []>("layer_norm_59_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_9_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139744064)))];
            tensor<fp16, [768]> p_encoder_layers_9_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139745664)))];
            tensor<fp16, []> layer_norm_59_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_59_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_59_cast_fp16 = layer_norm(axes = layer_norm_59_axes_0, beta = p_encoder_layers_9_norm_out_bias_to_fp16, epsilon = layer_norm_59_epsilon_0_to_fp16, gamma = p_encoder_layers_9_norm_out_weight_to_fp16, x = add_63_cast_fp16)[name = tensor<string, []>("layer_norm_59_cast_fp16")];
            tensor<int32, [1]> layer_norm_60_axes_0 = const()[name = tensor<string, []>("layer_norm_60_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139747264)))];
            tensor<fp16, [768]> p_encoder_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139748864)))];
            tensor<fp16, []> layer_norm_60_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_60_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_60_cast_fp16 = layer_norm(axes = layer_norm_60_axes_0, beta = p_encoder_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_60_epsilon_0_to_fp16, gamma = p_encoder_layers_10_norm_feed_forward1_weight_to_fp16, x = layer_norm_59_cast_fp16)[name = tensor<string, []>("layer_norm_60_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_10_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(139750464))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142109824))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_10_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142116032)))];
            tensor<fp16, [1, 750, 3072]> linear_80_cast_fp16 = linear(bias = p_encoder_layers_10_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_10_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_60_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_30_cast_fp16 = silu(x = linear_80_cast_fp16)[name = tensor<string, []>("silu_30_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_10_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142122240))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144481600))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_10_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144483200)))];
            tensor<fp16, [1, 750, 768]> linear_81_cast_fp16 = linear(bias = p_encoder_layers_10_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_10_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_30_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")];
            tensor<fp16, []> const_804_to_fp16 = const()[name = tensor<string, []>("const_804_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_60_cast_fp16 = mul(x = linear_81_cast_fp16, y = const_804_to_fp16)[name = tensor<string, []>("mul_60_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_64_cast_fp16 = add(x = layer_norm_59_cast_fp16, y = mul_60_cast_fp16)[name = tensor<string, []>("add_64_cast_fp16")];
            tensor<int32, [1]> layer_norm_61_axes_0 = const()[name = tensor<string, []>("layer_norm_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_10_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144484800)))];
            tensor<fp16, [768]> p_encoder_layers_10_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144486400)))];
            tensor<fp16, []> layer_norm_61_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_61_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_61_cast_fp16 = layer_norm(axes = layer_norm_61_axes_0, beta = p_encoder_layers_10_norm_self_att_bias_to_fp16, epsilon = layer_norm_61_epsilon_0_to_fp16, gamma = p_encoder_layers_10_norm_self_att_weight_to_fp16, x = add_64_cast_fp16)[name = tensor<string, []>("layer_norm_61_cast_fp16")];
            tensor<int32, [4]> const_808 = const()[name = tensor<string, []>("const_808"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_58_perm_1 = const()[name = tensor<string, []>("transpose_58_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_58 = transpose(perm = transpose_58_perm_1, x = layer_norm_61_cast_fp16)[name = tensor<string, []>("transpose_156")];
            tensor<fp16, [750, 1, 16, 48]> view_90_cast_fp16 = reshape(shape = const_808, x = transpose_58)[name = tensor<string, []>("view_90_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_61_cast_fp16 = mul(x = view_90_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_61_cast_fp16")];
            tensor<int32, [4]> slice_43_begin_0 = const()[name = tensor<string, []>("slice_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_43_end_0 = const()[name = tensor<string, []>("slice_43_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_43_end_mask_0 = const()[name = tensor<string, []>("slice_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_43_cast_fp16 = slice_by_index(begin = slice_43_begin_0, end = slice_43_end_0, end_mask = slice_43_end_mask_0, x = view_90_cast_fp16)[name = tensor<string, []>("slice_43_cast_fp16")];
            tensor<int32, [4]> slice_44_begin_0 = const()[name = tensor<string, []>("slice_44_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_44_end_0 = const()[name = tensor<string, []>("slice_44_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_44_end_mask_0 = const()[name = tensor<string, []>("slice_44_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_44_cast_fp16 = slice_by_index(begin = slice_44_begin_0, end = slice_44_end_0, end_mask = slice_44_end_mask_0, x = view_90_cast_fp16)[name = tensor<string, []>("slice_44_cast_fp16")];
            tensor<fp16, []> const_825_promoted_to_fp16 = const()[name = tensor<string, []>("const_825_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_20_cast_fp16 = mul(x = slice_44_cast_fp16, y = const_825_promoted_to_fp16)[name = tensor<string, []>("neg_20_cast_fp16")];
            tensor<int32, []> const_826 = const()[name = tensor<string, []>("const_826"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_20_interleave_0 = const()[name = tensor<string, []>("cat_20_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_20_cast_fp16 = concat(axis = const_826, interleave = cat_20_interleave_0, values = (neg_20_cast_fp16, slice_43_cast_fp16))[name = tensor<string, []>("cat_20_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_62_cast_fp16 = mul(x = cat_20_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_62_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_65_cast_fp16 = add(x = mul_61_cast_fp16, y = mul_62_cast_fp16)[name = tensor<string, []>("add_65_cast_fp16")];
            tensor<int32, [3]> const_835 = const()[name = tensor<string, []>("const_835"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_93_cast_fp16 = reshape(shape = const_835, x = add_65_cast_fp16)[name = tensor<string, []>("view_93_cast_fp16")];
            tensor<int32, [3]> transpose_156_perm_0 = const()[name = tensor<string, []>("transpose_156_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_10_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144488000))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145077888))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_10_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145079488)))];
            tensor<fp16, [1, 750, 768]> transpose_156_cast_fp16 = transpose(perm = transpose_156_perm_0, x = view_93_cast_fp16)[name = tensor<string, []>("transpose_155")];
            tensor<fp16, [1, 750, 768]> linear_82_cast_fp16 = linear(bias = p_encoder_layers_10_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_10_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_156_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")];
            tensor<int32, [4]> const_844 = const()[name = tensor<string, []>("const_844"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_96_cast_fp16 = reshape(shape = const_844, x = linear_82_cast_fp16)[name = tensor<string, []>("view_96_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_10_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145081088))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145670976))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_10_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145672576)))];
            tensor<fp16, [1, 750, 768]> linear_83_cast_fp16 = linear(bias = p_encoder_layers_10_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_10_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_156_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")];
            tensor<int32, [4]> const_845 = const()[name = tensor<string, []>("const_845"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_97_cast_fp16 = reshape(shape = const_845, x = linear_83_cast_fp16)[name = tensor<string, []>("view_97_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_10_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145674176))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146264064))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_10_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146265664)))];
            tensor<fp16, [1, 750, 768]> linear_84_cast_fp16 = linear(bias = p_encoder_layers_10_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_10_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_61_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")];
            tensor<int32, [4]> const_846 = const()[name = tensor<string, []>("const_846"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_98_cast_fp16 = reshape(shape = const_846, x = linear_84_cast_fp16)[name = tensor<string, []>("view_98_cast_fp16")];
            tensor<int32, [4]> transpose_161_perm_0 = const()[name = tensor<string, []>("transpose_161_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_12_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_12_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_12_cast_fp16 = mul(x = view_97_cast_fp16, y = _inversed_div_12_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_12_cast_fp16")];
            tensor<bool, []> matmul_20_transpose_x_0 = const()[name = tensor<string, []>("matmul_20_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_20_transpose_y_0 = const()[name = tensor<string, []>("matmul_20_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_85 = transpose(perm = transpose_85_perm_0, x = _inversed_div_12_cast_fp16)[name = tensor<string, []>("transpose_153")];
            tensor<fp16, [1, 16, 750, 48]> transpose_84 = transpose(perm = transpose_84_perm_0, x = view_96_cast_fp16)[name = tensor<string, []>("transpose_154")];
            tensor<fp16, [1, 16, 750, 750]> matmul_20_cast_fp16 = matmul(transpose_x = matmul_20_transpose_x_0, transpose_y = matmul_20_transpose_y_0, x = transpose_84, y = transpose_85)[name = tensor<string, []>("matmul_20_cast_fp16")];
            tensor<int32, []> const_856 = const()[name = tensor<string, []>("const_856"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_10_cast_fp16 = softmax(axis = const_856, x = matmul_20_cast_fp16)[name = tensor<string, []>("softmax_10_cast_fp16")];
            tensor<bool, []> matmul_21_transpose_x_0 = const()[name = tensor<string, []>("matmul_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_21_transpose_y_0 = const()[name = tensor<string, []>("matmul_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_161_cast_fp16 = transpose(perm = transpose_161_perm_0, x = view_98_cast_fp16)[name = tensor<string, []>("transpose_152")];
            tensor<fp16, [1, 16, 750, 48]> matmul_21_cast_fp16 = matmul(transpose_x = matmul_21_transpose_x_0, transpose_y = matmul_21_transpose_y_0, x = softmax_10_cast_fp16, y = transpose_161_cast_fp16)[name = tensor<string, []>("matmul_21_cast_fp16")];
            tensor<int32, [4]> transpose_163_perm_0 = const()[name = tensor<string, []>("transpose_163_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_859 = const()[name = tensor<string, []>("const_859"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_163_cast_fp16 = transpose(perm = transpose_163_perm_0, x = matmul_21_cast_fp16)[name = tensor<string, []>("transpose_151")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_10_cast_fp16 = reshape(shape = const_859, x = transpose_163_cast_fp16)[name = tensor<string, []>("_unsafe_view_10_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_10_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146267264))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146857152))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_10_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146858752)))];
            tensor<fp16, [1, 750, 768]> linear_85_cast_fp16 = linear(bias = p_encoder_layers_10_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_10_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_10_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_67_cast_fp16 = add(x = add_64_cast_fp16, y = linear_85_cast_fp16)[name = tensor<string, []>("add_67_cast_fp16")];
            tensor<int32, [1]> layer_norm_62_axes_0 = const()[name = tensor<string, []>("layer_norm_62_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_10_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146860352)))];
            tensor<fp16, [768]> p_encoder_layers_10_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146861952)))];
            tensor<fp16, []> layer_norm_62_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_62_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_62_cast_fp16 = layer_norm(axes = layer_norm_62_axes_0, beta = p_encoder_layers_10_norm_conv_bias_to_fp16, epsilon = layer_norm_62_epsilon_0_to_fp16, gamma = p_encoder_layers_10_norm_conv_weight_to_fp16, x = add_67_cast_fp16)[name = tensor<string, []>("layer_norm_62_cast_fp16")];
            tensor<int32, [3]> transpose_164_perm_0 = const()[name = tensor<string, []>("transpose_164_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_32_pad_type_0 = const()[name = tensor<string, []>("conv1d_32_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_32_strides_0 = const()[name = tensor<string, []>("conv1d_32_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_32_pad_0 = const()[name = tensor<string, []>("conv1d_32_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_32_dilations_0 = const()[name = tensor<string, []>("conv1d_32_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_32_groups_0 = const()[name = tensor<string, []>("conv1d_32_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146863552))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148043264))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_10_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148046400)))];
            tensor<fp16, [1, 768, 750]> transpose_164_cast_fp16 = transpose(perm = transpose_164_perm_0, x = layer_norm_62_cast_fp16)[name = tensor<string, []>("transpose_150")];
            tensor<fp16, [1, 1536, 750]> conv1d_32_cast_fp16 = conv(bias = p_encoder_layers_10_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_32_dilations_0, groups = conv1d_32_groups_0, pad = conv1d_32_pad_0, pad_type = conv1d_32_pad_type_0, strides = conv1d_32_strides_0, weight = p_encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_164_cast_fp16)[name = tensor<string, []>("conv1d_32_cast_fp16")];
            tensor<int32, []> glu_10_split_num_splits_0 = const()[name = tensor<string, []>("glu_10_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_10_split_axis_0 = const()[name = tensor<string, []>("glu_10_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_10_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_10_split_cast_fp16_1 = split(axis = glu_10_split_axis_0, num_splits = glu_10_split_num_splits_0, x = conv1d_32_cast_fp16)[name = tensor<string, []>("glu_10_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_10_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_10_split_cast_fp16_1)[name = tensor<string, []>("glu_10_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_10_cast_fp16 = mul(x = glu_10_split_cast_fp16_0, y = glu_10_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_10_cast_fp16")];
            tensor<string, []> conv1d_33_pad_type_0 = const()[name = tensor<string, []>("conv1d_33_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_33_pad_0 = const()[name = tensor<string, []>("conv1d_33_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_33_groups_0 = const()[name = tensor<string, []>("conv1d_33_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_33_strides_0 = const()[name = tensor<string, []>("conv1d_33_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_33_dilations_0 = const()[name = tensor<string, []>("conv1d_33_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_10_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148049536))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148053440))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_10_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148055040)))];
            tensor<fp16, [1, 768, 750]> conv1d_33_cast_fp16 = conv(bias = p_encoder_layers_10_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_33_dilations_0, groups = conv1d_33_groups_0, pad = conv1d_33_pad_0, pad_type = conv1d_33_pad_type_0, strides = conv1d_33_strides_0, weight = p_encoder_layers_10_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_10_cast_fp16)[name = tensor<string, []>("conv1d_33_cast_fp16")];
            tensor<int32, [3]> transpose_165_perm_0 = const()[name = tensor<string, []>("transpose_165_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_63_axes_0 = const()[name = tensor<string, []>("layer_norm_63_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_10_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148056640)))];
            tensor<fp16, [768]> p_encoder_layers_10_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148058240)))];
            tensor<fp16, []> layer_norm_63_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_63_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_165_cast_fp16 = transpose(perm = transpose_165_perm_0, x = conv1d_33_cast_fp16)[name = tensor<string, []>("transpose_149")];
            tensor<fp16, [1, 750, 768]> layer_norm_63_cast_fp16 = layer_norm(axes = layer_norm_63_axes_0, beta = p_encoder_layers_10_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_63_epsilon_0_to_fp16, gamma = p_encoder_layers_10_conv_batch_norm_weight_to_fp16, x = transpose_165_cast_fp16)[name = tensor<string, []>("layer_norm_63_cast_fp16")];
            tensor<int32, [3]> transpose_166_perm_0 = const()[name = tensor<string, []>("transpose_166_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_166_cast_fp16 = transpose(perm = transpose_166_perm_0, x = layer_norm_63_cast_fp16)[name = tensor<string, []>("transpose_148")];
            tensor<fp16, [1, 768, 750]> silu_31_cast_fp16 = silu(x = transpose_166_cast_fp16)[name = tensor<string, []>("silu_31_cast_fp16")];
            tensor<string, []> conv1d_34_pad_type_0 = const()[name = tensor<string, []>("conv1d_34_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_34_strides_0 = const()[name = tensor<string, []>("conv1d_34_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_34_pad_0 = const()[name = tensor<string, []>("conv1d_34_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_34_dilations_0 = const()[name = tensor<string, []>("conv1d_34_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_34_groups_0 = const()[name = tensor<string, []>("conv1d_34_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148059840))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148649728))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_10_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148651328)))];
            tensor<fp16, [1, 768, 750]> conv1d_34_cast_fp16 = conv(bias = p_encoder_layers_10_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_34_dilations_0, groups = conv1d_34_groups_0, pad = conv1d_34_pad_0, pad_type = conv1d_34_pad_type_0, strides = conv1d_34_strides_0, weight = p_encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_31_cast_fp16)[name = tensor<string, []>("conv1d_34_cast_fp16")];
            tensor<int32, [3]> transpose_167_perm_0 = const()[name = tensor<string, []>("transpose_167_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_167_cast_fp16 = transpose(perm = transpose_167_perm_0, x = conv1d_34_cast_fp16)[name = tensor<string, []>("transpose_147")];
            tensor<fp16, [1, 750, 768]> add_68_cast_fp16 = add(x = add_67_cast_fp16, y = transpose_167_cast_fp16)[name = tensor<string, []>("add_68_cast_fp16")];
            tensor<int32, [1]> layer_norm_64_axes_0 = const()[name = tensor<string, []>("layer_norm_64_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148652928)))];
            tensor<fp16, [768]> p_encoder_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148654528)))];
            tensor<fp16, []> layer_norm_64_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_64_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_64_cast_fp16 = layer_norm(axes = layer_norm_64_axes_0, beta = p_encoder_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_64_epsilon_0_to_fp16, gamma = p_encoder_layers_10_norm_feed_forward2_weight_to_fp16, x = add_68_cast_fp16)[name = tensor<string, []>("layer_norm_64_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_10_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(148656128))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151015488))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_10_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151021696)))];
            tensor<fp16, [1, 750, 3072]> linear_86_cast_fp16 = linear(bias = p_encoder_layers_10_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_10_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_64_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_32_cast_fp16 = silu(x = linear_86_cast_fp16)[name = tensor<string, []>("silu_32_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_10_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_10_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151027904))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153387264))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_10_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153388864)))];
            tensor<fp16, [1, 750, 768]> linear_87_cast_fp16 = linear(bias = p_encoder_layers_10_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_10_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_32_cast_fp16)[name = tensor<string, []>("linear_87_cast_fp16")];
            tensor<fp16, []> const_878_to_fp16 = const()[name = tensor<string, []>("const_878_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_65_cast_fp16 = mul(x = linear_87_cast_fp16, y = const_878_to_fp16)[name = tensor<string, []>("mul_65_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_69_cast_fp16 = add(x = add_68_cast_fp16, y = mul_65_cast_fp16)[name = tensor<string, []>("add_69_cast_fp16")];
            tensor<int32, [1]> layer_norm_65_axes_0 = const()[name = tensor<string, []>("layer_norm_65_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_10_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153390464)))];
            tensor<fp16, [768]> p_encoder_layers_10_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153392064)))];
            tensor<fp16, []> layer_norm_65_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_65_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_65_cast_fp16 = layer_norm(axes = layer_norm_65_axes_0, beta = p_encoder_layers_10_norm_out_bias_to_fp16, epsilon = layer_norm_65_epsilon_0_to_fp16, gamma = p_encoder_layers_10_norm_out_weight_to_fp16, x = add_69_cast_fp16)[name = tensor<string, []>("layer_norm_65_cast_fp16")];
            tensor<int32, [1]> layer_norm_66_axes_0 = const()[name = tensor<string, []>("layer_norm_66_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153393664)))];
            tensor<fp16, [768]> p_encoder_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153395264)))];
            tensor<fp16, []> layer_norm_66_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_66_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_66_cast_fp16 = layer_norm(axes = layer_norm_66_axes_0, beta = p_encoder_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_66_epsilon_0_to_fp16, gamma = p_encoder_layers_11_norm_feed_forward1_weight_to_fp16, x = layer_norm_65_cast_fp16)[name = tensor<string, []>("layer_norm_66_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_11_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(153396864))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155756224))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_11_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155762432)))];
            tensor<fp16, [1, 750, 3072]> linear_88_cast_fp16 = linear(bias = p_encoder_layers_11_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_11_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_66_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_33_cast_fp16 = silu(x = linear_88_cast_fp16)[name = tensor<string, []>("silu_33_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_11_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155768640))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158128000))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_11_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158129600)))];
            tensor<fp16, [1, 750, 768]> linear_89_cast_fp16 = linear(bias = p_encoder_layers_11_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_11_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_33_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")];
            tensor<fp16, []> const_881_to_fp16 = const()[name = tensor<string, []>("const_881_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_66_cast_fp16 = mul(x = linear_89_cast_fp16, y = const_881_to_fp16)[name = tensor<string, []>("mul_66_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_70_cast_fp16 = add(x = layer_norm_65_cast_fp16, y = mul_66_cast_fp16)[name = tensor<string, []>("add_70_cast_fp16")];
            tensor<int32, [1]> layer_norm_67_axes_0 = const()[name = tensor<string, []>("layer_norm_67_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_11_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158131200)))];
            tensor<fp16, [768]> p_encoder_layers_11_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158132800)))];
            tensor<fp16, []> layer_norm_67_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_67_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_67_cast_fp16 = layer_norm(axes = layer_norm_67_axes_0, beta = p_encoder_layers_11_norm_self_att_bias_to_fp16, epsilon = layer_norm_67_epsilon_0_to_fp16, gamma = p_encoder_layers_11_norm_self_att_weight_to_fp16, x = add_70_cast_fp16)[name = tensor<string, []>("layer_norm_67_cast_fp16")];
            tensor<int32, [4]> const_885 = const()[name = tensor<string, []>("const_885"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_59_perm_1 = const()[name = tensor<string, []>("transpose_59_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_59 = transpose(perm = transpose_59_perm_1, x = layer_norm_67_cast_fp16)[name = tensor<string, []>("transpose_146")];
            tensor<fp16, [750, 1, 16, 48]> view_99_cast_fp16 = reshape(shape = const_885, x = transpose_59)[name = tensor<string, []>("view_99_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_67_cast_fp16 = mul(x = view_99_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_67_cast_fp16")];
            tensor<int32, [4]> slice_47_begin_0 = const()[name = tensor<string, []>("slice_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_47_end_0 = const()[name = tensor<string, []>("slice_47_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_47_end_mask_0 = const()[name = tensor<string, []>("slice_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_47_cast_fp16 = slice_by_index(begin = slice_47_begin_0, end = slice_47_end_0, end_mask = slice_47_end_mask_0, x = view_99_cast_fp16)[name = tensor<string, []>("slice_47_cast_fp16")];
            tensor<int32, [4]> slice_48_begin_0 = const()[name = tensor<string, []>("slice_48_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_48_end_0 = const()[name = tensor<string, []>("slice_48_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_48_end_mask_0 = const()[name = tensor<string, []>("slice_48_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_48_cast_fp16 = slice_by_index(begin = slice_48_begin_0, end = slice_48_end_0, end_mask = slice_48_end_mask_0, x = view_99_cast_fp16)[name = tensor<string, []>("slice_48_cast_fp16")];
            tensor<fp16, []> const_902_promoted_to_fp16 = const()[name = tensor<string, []>("const_902_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_22_cast_fp16 = mul(x = slice_48_cast_fp16, y = const_902_promoted_to_fp16)[name = tensor<string, []>("neg_22_cast_fp16")];
            tensor<int32, []> const_903 = const()[name = tensor<string, []>("const_903"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_22_interleave_0 = const()[name = tensor<string, []>("cat_22_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_22_cast_fp16 = concat(axis = const_903, interleave = cat_22_interleave_0, values = (neg_22_cast_fp16, slice_47_cast_fp16))[name = tensor<string, []>("cat_22_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_68_cast_fp16 = mul(x = cat_22_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_68_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_71_cast_fp16 = add(x = mul_67_cast_fp16, y = mul_68_cast_fp16)[name = tensor<string, []>("add_71_cast_fp16")];
            tensor<int32, [3]> const_912 = const()[name = tensor<string, []>("const_912"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_102_cast_fp16 = reshape(shape = const_912, x = add_71_cast_fp16)[name = tensor<string, []>("view_102_cast_fp16")];
            tensor<int32, [3]> transpose_171_perm_0 = const()[name = tensor<string, []>("transpose_171_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_11_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158134400))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158724288))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_11_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158725888)))];
            tensor<fp16, [1, 750, 768]> transpose_171_cast_fp16 = transpose(perm = transpose_171_perm_0, x = view_102_cast_fp16)[name = tensor<string, []>("transpose_145")];
            tensor<fp16, [1, 750, 768]> linear_90_cast_fp16 = linear(bias = p_encoder_layers_11_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_11_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_171_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")];
            tensor<int32, [4]> const_921 = const()[name = tensor<string, []>("const_921"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_105_cast_fp16 = reshape(shape = const_921, x = linear_90_cast_fp16)[name = tensor<string, []>("view_105_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_11_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158727488))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159317376))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_11_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159318976)))];
            tensor<fp16, [1, 750, 768]> linear_91_cast_fp16 = linear(bias = p_encoder_layers_11_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_11_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_171_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")];
            tensor<int32, [4]> const_922 = const()[name = tensor<string, []>("const_922"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_106_cast_fp16 = reshape(shape = const_922, x = linear_91_cast_fp16)[name = tensor<string, []>("view_106_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_11_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159320576))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159910464))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_11_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159912064)))];
            tensor<fp16, [1, 750, 768]> linear_92_cast_fp16 = linear(bias = p_encoder_layers_11_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_11_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_67_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")];
            tensor<int32, [4]> const_923 = const()[name = tensor<string, []>("const_923"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_107_cast_fp16 = reshape(shape = const_923, x = linear_92_cast_fp16)[name = tensor<string, []>("view_107_cast_fp16")];
            tensor<int32, [4]> transpose_176_perm_0 = const()[name = tensor<string, []>("transpose_176_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_13_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_13_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_13_cast_fp16 = mul(x = view_106_cast_fp16, y = _inversed_div_13_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_13_cast_fp16")];
            tensor<bool, []> matmul_22_transpose_x_0 = const()[name = tensor<string, []>("matmul_22_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_22_transpose_y_0 = const()[name = tensor<string, []>("matmul_22_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_86_perm_0_1 = const()[name = tensor<string, []>("transpose_86_perm_0_1"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_87_perm_0 = const()[name = tensor<string, []>("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_87 = transpose(perm = transpose_87_perm_0, x = _inversed_div_13_cast_fp16)[name = tensor<string, []>("transpose_143")];
            tensor<fp16, [1, 16, 750, 48]> transpose_86 = transpose(perm = transpose_86_perm_0_1, x = view_105_cast_fp16)[name = tensor<string, []>("transpose_144")];
            tensor<fp16, [1, 16, 750, 750]> matmul_22_cast_fp16 = matmul(transpose_x = matmul_22_transpose_x_0, transpose_y = matmul_22_transpose_y_0, x = transpose_86, y = transpose_87)[name = tensor<string, []>("matmul_22_cast_fp16")];
            tensor<int32, []> const_933 = const()[name = tensor<string, []>("const_933"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_11_cast_fp16 = softmax(axis = const_933, x = matmul_22_cast_fp16)[name = tensor<string, []>("softmax_11_cast_fp16")];
            tensor<bool, []> matmul_23_transpose_x_0 = const()[name = tensor<string, []>("matmul_23_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_23_transpose_y_0 = const()[name = tensor<string, []>("matmul_23_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_176_cast_fp16 = transpose(perm = transpose_176_perm_0, x = view_107_cast_fp16)[name = tensor<string, []>("transpose_142")];
            tensor<fp16, [1, 16, 750, 48]> matmul_23_cast_fp16 = matmul(transpose_x = matmul_23_transpose_x_0, transpose_y = matmul_23_transpose_y_0, x = softmax_11_cast_fp16, y = transpose_176_cast_fp16)[name = tensor<string, []>("matmul_23_cast_fp16")];
            tensor<int32, [4]> transpose_178_perm_0 = const()[name = tensor<string, []>("transpose_178_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_936 = const()[name = tensor<string, []>("const_936"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_178_cast_fp16 = transpose(perm = transpose_178_perm_0, x = matmul_23_cast_fp16)[name = tensor<string, []>("transpose_141")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_11_cast_fp16 = reshape(shape = const_936, x = transpose_178_cast_fp16)[name = tensor<string, []>("_unsafe_view_11_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_11_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(159913664))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160503552))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_11_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160505152)))];
            tensor<fp16, [1, 750, 768]> linear_93_cast_fp16 = linear(bias = p_encoder_layers_11_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_11_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_11_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_73_cast_fp16 = add(x = add_70_cast_fp16, y = linear_93_cast_fp16)[name = tensor<string, []>("add_73_cast_fp16")];
            tensor<int32, [1]> layer_norm_68_axes_0 = const()[name = tensor<string, []>("layer_norm_68_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_11_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160506752)))];
            tensor<fp16, [768]> p_encoder_layers_11_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160508352)))];
            tensor<fp16, []> layer_norm_68_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_68_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_68_cast_fp16 = layer_norm(axes = layer_norm_68_axes_0, beta = p_encoder_layers_11_norm_conv_bias_to_fp16, epsilon = layer_norm_68_epsilon_0_to_fp16, gamma = p_encoder_layers_11_norm_conv_weight_to_fp16, x = add_73_cast_fp16)[name = tensor<string, []>("layer_norm_68_cast_fp16")];
            tensor<int32, [3]> transpose_179_perm_0 = const()[name = tensor<string, []>("transpose_179_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_35_pad_type_0 = const()[name = tensor<string, []>("conv1d_35_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_35_strides_0 = const()[name = tensor<string, []>("conv1d_35_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_35_pad_0 = const()[name = tensor<string, []>("conv1d_35_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_35_dilations_0 = const()[name = tensor<string, []>("conv1d_35_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_35_groups_0 = const()[name = tensor<string, []>("conv1d_35_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(160509952))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161689664))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_11_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161692800)))];
            tensor<fp16, [1, 768, 750]> transpose_179_cast_fp16 = transpose(perm = transpose_179_perm_0, x = layer_norm_68_cast_fp16)[name = tensor<string, []>("transpose_140")];
            tensor<fp16, [1, 1536, 750]> conv1d_35_cast_fp16 = conv(bias = p_encoder_layers_11_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_35_dilations_0, groups = conv1d_35_groups_0, pad = conv1d_35_pad_0, pad_type = conv1d_35_pad_type_0, strides = conv1d_35_strides_0, weight = p_encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_179_cast_fp16)[name = tensor<string, []>("conv1d_35_cast_fp16")];
            tensor<int32, []> glu_11_split_num_splits_0 = const()[name = tensor<string, []>("glu_11_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_11_split_axis_0 = const()[name = tensor<string, []>("glu_11_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_11_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_11_split_cast_fp16_1 = split(axis = glu_11_split_axis_0, num_splits = glu_11_split_num_splits_0, x = conv1d_35_cast_fp16)[name = tensor<string, []>("glu_11_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_11_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_11_split_cast_fp16_1)[name = tensor<string, []>("glu_11_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_11_cast_fp16 = mul(x = glu_11_split_cast_fp16_0, y = glu_11_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_11_cast_fp16")];
            tensor<string, []> conv1d_36_pad_type_0 = const()[name = tensor<string, []>("conv1d_36_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_36_pad_0 = const()[name = tensor<string, []>("conv1d_36_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_36_groups_0 = const()[name = tensor<string, []>("conv1d_36_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_36_strides_0 = const()[name = tensor<string, []>("conv1d_36_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_36_dilations_0 = const()[name = tensor<string, []>("conv1d_36_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_11_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161695936))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161699840))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_11_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161701440)))];
            tensor<fp16, [1, 768, 750]> conv1d_36_cast_fp16 = conv(bias = p_encoder_layers_11_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_36_dilations_0, groups = conv1d_36_groups_0, pad = conv1d_36_pad_0, pad_type = conv1d_36_pad_type_0, strides = conv1d_36_strides_0, weight = p_encoder_layers_11_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_11_cast_fp16)[name = tensor<string, []>("conv1d_36_cast_fp16")];
            tensor<int32, [3]> transpose_180_perm_0 = const()[name = tensor<string, []>("transpose_180_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_69_axes_0 = const()[name = tensor<string, []>("layer_norm_69_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_11_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161703040)))];
            tensor<fp16, [768]> p_encoder_layers_11_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161704640)))];
            tensor<fp16, []> layer_norm_69_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_69_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_180_cast_fp16 = transpose(perm = transpose_180_perm_0, x = conv1d_36_cast_fp16)[name = tensor<string, []>("transpose_139")];
            tensor<fp16, [1, 750, 768]> layer_norm_69_cast_fp16 = layer_norm(axes = layer_norm_69_axes_0, beta = p_encoder_layers_11_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_69_epsilon_0_to_fp16, gamma = p_encoder_layers_11_conv_batch_norm_weight_to_fp16, x = transpose_180_cast_fp16)[name = tensor<string, []>("layer_norm_69_cast_fp16")];
            tensor<int32, [3]> transpose_181_perm_0 = const()[name = tensor<string, []>("transpose_181_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_181_cast_fp16 = transpose(perm = transpose_181_perm_0, x = layer_norm_69_cast_fp16)[name = tensor<string, []>("transpose_138")];
            tensor<fp16, [1, 768, 750]> silu_34_cast_fp16 = silu(x = transpose_181_cast_fp16)[name = tensor<string, []>("silu_34_cast_fp16")];
            tensor<string, []> conv1d_37_pad_type_0 = const()[name = tensor<string, []>("conv1d_37_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_37_strides_0 = const()[name = tensor<string, []>("conv1d_37_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_37_pad_0 = const()[name = tensor<string, []>("conv1d_37_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_37_dilations_0 = const()[name = tensor<string, []>("conv1d_37_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_37_groups_0 = const()[name = tensor<string, []>("conv1d_37_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(161706240))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162296128))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_11_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162297728)))];
            tensor<fp16, [1, 768, 750]> conv1d_37_cast_fp16 = conv(bias = p_encoder_layers_11_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_37_dilations_0, groups = conv1d_37_groups_0, pad = conv1d_37_pad_0, pad_type = conv1d_37_pad_type_0, strides = conv1d_37_strides_0, weight = p_encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_34_cast_fp16)[name = tensor<string, []>("conv1d_37_cast_fp16")];
            tensor<int32, [3]> transpose_182_perm_0 = const()[name = tensor<string, []>("transpose_182_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_182_cast_fp16 = transpose(perm = transpose_182_perm_0, x = conv1d_37_cast_fp16)[name = tensor<string, []>("transpose_137")];
            tensor<fp16, [1, 750, 768]> add_74_cast_fp16 = add(x = add_73_cast_fp16, y = transpose_182_cast_fp16)[name = tensor<string, []>("add_74_cast_fp16")];
            tensor<int32, [1]> layer_norm_70_axes_0 = const()[name = tensor<string, []>("layer_norm_70_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162299328)))];
            tensor<fp16, [768]> p_encoder_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162300928)))];
            tensor<fp16, []> layer_norm_70_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_70_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_70_cast_fp16 = layer_norm(axes = layer_norm_70_axes_0, beta = p_encoder_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_70_epsilon_0_to_fp16, gamma = p_encoder_layers_11_norm_feed_forward2_weight_to_fp16, x = add_74_cast_fp16)[name = tensor<string, []>("layer_norm_70_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_11_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162302528))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164661888))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_11_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164668096)))];
            tensor<fp16, [1, 750, 3072]> linear_94_cast_fp16 = linear(bias = p_encoder_layers_11_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_11_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_70_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_35_cast_fp16 = silu(x = linear_94_cast_fp16)[name = tensor<string, []>("silu_35_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_11_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_11_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164674304))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167033664))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_11_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167035264)))];
            tensor<fp16, [1, 750, 768]> linear_95_cast_fp16 = linear(bias = p_encoder_layers_11_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_11_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_35_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")];
            tensor<fp16, []> const_955_to_fp16 = const()[name = tensor<string, []>("const_955_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_71_cast_fp16 = mul(x = linear_95_cast_fp16, y = const_955_to_fp16)[name = tensor<string, []>("mul_71_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_75_cast_fp16 = add(x = add_74_cast_fp16, y = mul_71_cast_fp16)[name = tensor<string, []>("add_75_cast_fp16")];
            tensor<int32, [1]> layer_norm_71_axes_0 = const()[name = tensor<string, []>("layer_norm_71_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_11_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167036864)))];
            tensor<fp16, [768]> p_encoder_layers_11_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167038464)))];
            tensor<fp16, []> layer_norm_71_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_71_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_71_cast_fp16 = layer_norm(axes = layer_norm_71_axes_0, beta = p_encoder_layers_11_norm_out_bias_to_fp16, epsilon = layer_norm_71_epsilon_0_to_fp16, gamma = p_encoder_layers_11_norm_out_weight_to_fp16, x = add_75_cast_fp16)[name = tensor<string, []>("layer_norm_71_cast_fp16")];
            tensor<int32, [1]> layer_norm_72_axes_0 = const()[name = tensor<string, []>("layer_norm_72_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167040064)))];
            tensor<fp16, [768]> p_encoder_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167041664)))];
            tensor<fp16, []> layer_norm_72_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_72_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_72_cast_fp16 = layer_norm(axes = layer_norm_72_axes_0, beta = p_encoder_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_72_epsilon_0_to_fp16, gamma = p_encoder_layers_12_norm_feed_forward1_weight_to_fp16, x = layer_norm_71_cast_fp16)[name = tensor<string, []>("layer_norm_72_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_12_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(167043264))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169402624))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_12_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169408832)))];
            tensor<fp16, [1, 750, 3072]> linear_96_cast_fp16 = linear(bias = p_encoder_layers_12_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_12_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_72_cast_fp16)[name = tensor<string, []>("linear_96_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_36_cast_fp16 = silu(x = linear_96_cast_fp16)[name = tensor<string, []>("silu_36_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_12_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(169415040))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171774400))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_12_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171776000)))];
            tensor<fp16, [1, 750, 768]> linear_97_cast_fp16 = linear(bias = p_encoder_layers_12_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_12_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_36_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")];
            tensor<fp16, []> const_958_to_fp16 = const()[name = tensor<string, []>("const_958_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_72_cast_fp16 = mul(x = linear_97_cast_fp16, y = const_958_to_fp16)[name = tensor<string, []>("mul_72_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_76_cast_fp16 = add(x = layer_norm_71_cast_fp16, y = mul_72_cast_fp16)[name = tensor<string, []>("add_76_cast_fp16")];
            tensor<int32, [1]> layer_norm_73_axes_0 = const()[name = tensor<string, []>("layer_norm_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_12_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171777600)))];
            tensor<fp16, [768]> p_encoder_layers_12_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171779200)))];
            tensor<fp16, []> layer_norm_73_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_73_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_73_cast_fp16 = layer_norm(axes = layer_norm_73_axes_0, beta = p_encoder_layers_12_norm_self_att_bias_to_fp16, epsilon = layer_norm_73_epsilon_0_to_fp16, gamma = p_encoder_layers_12_norm_self_att_weight_to_fp16, x = add_76_cast_fp16)[name = tensor<string, []>("layer_norm_73_cast_fp16")];
            tensor<int32, [4]> const_962 = const()[name = tensor<string, []>("const_962"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_60_perm_1 = const()[name = tensor<string, []>("transpose_60_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_60 = transpose(perm = transpose_60_perm_1, x = layer_norm_73_cast_fp16)[name = tensor<string, []>("transpose_136")];
            tensor<fp16, [750, 1, 16, 48]> view_108_cast_fp16 = reshape(shape = const_962, x = transpose_60)[name = tensor<string, []>("view_108_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_73_cast_fp16 = mul(x = view_108_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_73_cast_fp16")];
            tensor<int32, [4]> slice_51_begin_0 = const()[name = tensor<string, []>("slice_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_51_end_0 = const()[name = tensor<string, []>("slice_51_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_51_end_mask_0 = const()[name = tensor<string, []>("slice_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_51_cast_fp16 = slice_by_index(begin = slice_51_begin_0, end = slice_51_end_0, end_mask = slice_51_end_mask_0, x = view_108_cast_fp16)[name = tensor<string, []>("slice_51_cast_fp16")];
            tensor<int32, [4]> slice_52_begin_0 = const()[name = tensor<string, []>("slice_52_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_52_end_0 = const()[name = tensor<string, []>("slice_52_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_52_end_mask_0 = const()[name = tensor<string, []>("slice_52_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_52_cast_fp16 = slice_by_index(begin = slice_52_begin_0, end = slice_52_end_0, end_mask = slice_52_end_mask_0, x = view_108_cast_fp16)[name = tensor<string, []>("slice_52_cast_fp16")];
            tensor<fp16, []> const_979_promoted_to_fp16 = const()[name = tensor<string, []>("const_979_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_24_cast_fp16 = mul(x = slice_52_cast_fp16, y = const_979_promoted_to_fp16)[name = tensor<string, []>("neg_24_cast_fp16")];
            tensor<int32, []> const_980 = const()[name = tensor<string, []>("const_980"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_24_interleave_0 = const()[name = tensor<string, []>("cat_24_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_24_cast_fp16 = concat(axis = const_980, interleave = cat_24_interleave_0, values = (neg_24_cast_fp16, slice_51_cast_fp16))[name = tensor<string, []>("cat_24_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_74_cast_fp16 = mul(x = cat_24_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_74_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_77_cast_fp16 = add(x = mul_73_cast_fp16, y = mul_74_cast_fp16)[name = tensor<string, []>("add_77_cast_fp16")];
            tensor<int32, [3]> const_989 = const()[name = tensor<string, []>("const_989"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_111_cast_fp16 = reshape(shape = const_989, x = add_77_cast_fp16)[name = tensor<string, []>("view_111_cast_fp16")];
            tensor<int32, [3]> transpose_186_perm_0 = const()[name = tensor<string, []>("transpose_186_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_12_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171780800))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172370688))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_12_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172372288)))];
            tensor<fp16, [1, 750, 768]> transpose_186_cast_fp16 = transpose(perm = transpose_186_perm_0, x = view_111_cast_fp16)[name = tensor<string, []>("transpose_135")];
            tensor<fp16, [1, 750, 768]> linear_98_cast_fp16 = linear(bias = p_encoder_layers_12_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_12_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_186_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")];
            tensor<int32, [4]> const_998 = const()[name = tensor<string, []>("const_998"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_114_cast_fp16 = reshape(shape = const_998, x = linear_98_cast_fp16)[name = tensor<string, []>("view_114_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_12_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172373888))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172963776))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_12_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172965376)))];
            tensor<fp16, [1, 750, 768]> linear_99_cast_fp16 = linear(bias = p_encoder_layers_12_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_12_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_186_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")];
            tensor<int32, [4]> const_999 = const()[name = tensor<string, []>("const_999"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_115_cast_fp16 = reshape(shape = const_999, x = linear_99_cast_fp16)[name = tensor<string, []>("view_115_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_12_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(172966976))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173556864))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_12_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173558464)))];
            tensor<fp16, [1, 750, 768]> linear_100_cast_fp16 = linear(bias = p_encoder_layers_12_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_12_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_73_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")];
            tensor<int32, [4]> const_1000 = const()[name = tensor<string, []>("const_1000"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_116_cast_fp16 = reshape(shape = const_1000, x = linear_100_cast_fp16)[name = tensor<string, []>("view_116_cast_fp16")];
            tensor<int32, [4]> transpose_191_perm_0 = const()[name = tensor<string, []>("transpose_191_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_14_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_14_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_14_cast_fp16 = mul(x = view_115_cast_fp16, y = _inversed_div_14_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_14_cast_fp16")];
            tensor<bool, []> matmul_24_transpose_x_0 = const()[name = tensor<string, []>("matmul_24_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_24_transpose_y_0 = const()[name = tensor<string, []>("matmul_24_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_88_perm_0_1 = const()[name = tensor<string, []>("transpose_88_perm_0_1"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_89_perm_0_1 = const()[name = tensor<string, []>("transpose_89_perm_0_1"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_89 = transpose(perm = transpose_89_perm_0_1, x = _inversed_div_14_cast_fp16)[name = tensor<string, []>("transpose_133")];
            tensor<fp16, [1, 16, 750, 48]> transpose_88 = transpose(perm = transpose_88_perm_0_1, x = view_114_cast_fp16)[name = tensor<string, []>("transpose_134")];
            tensor<fp16, [1, 16, 750, 750]> matmul_24_cast_fp16 = matmul(transpose_x = matmul_24_transpose_x_0, transpose_y = matmul_24_transpose_y_0, x = transpose_88, y = transpose_89)[name = tensor<string, []>("matmul_24_cast_fp16")];
            tensor<int32, []> const_1010 = const()[name = tensor<string, []>("const_1010"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_12_cast_fp16 = softmax(axis = const_1010, x = matmul_24_cast_fp16)[name = tensor<string, []>("softmax_12_cast_fp16")];
            tensor<bool, []> matmul_25_transpose_x_0 = const()[name = tensor<string, []>("matmul_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_25_transpose_y_0 = const()[name = tensor<string, []>("matmul_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_191_cast_fp16 = transpose(perm = transpose_191_perm_0, x = view_116_cast_fp16)[name = tensor<string, []>("transpose_132")];
            tensor<fp16, [1, 16, 750, 48]> matmul_25_cast_fp16 = matmul(transpose_x = matmul_25_transpose_x_0, transpose_y = matmul_25_transpose_y_0, x = softmax_12_cast_fp16, y = transpose_191_cast_fp16)[name = tensor<string, []>("matmul_25_cast_fp16")];
            tensor<int32, [4]> transpose_193_perm_0 = const()[name = tensor<string, []>("transpose_193_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1013 = const()[name = tensor<string, []>("const_1013"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_193_cast_fp16 = transpose(perm = transpose_193_perm_0, x = matmul_25_cast_fp16)[name = tensor<string, []>("transpose_131")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_12_cast_fp16 = reshape(shape = const_1013, x = transpose_193_cast_fp16)[name = tensor<string, []>("_unsafe_view_12_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_12_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173560064))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174149952))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_12_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174151552)))];
            tensor<fp16, [1, 750, 768]> linear_101_cast_fp16 = linear(bias = p_encoder_layers_12_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_12_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_12_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_79_cast_fp16 = add(x = add_76_cast_fp16, y = linear_101_cast_fp16)[name = tensor<string, []>("add_79_cast_fp16")];
            tensor<int32, [1]> layer_norm_74_axes_0 = const()[name = tensor<string, []>("layer_norm_74_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_12_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174153152)))];
            tensor<fp16, [768]> p_encoder_layers_12_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174154752)))];
            tensor<fp16, []> layer_norm_74_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_74_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_74_cast_fp16 = layer_norm(axes = layer_norm_74_axes_0, beta = p_encoder_layers_12_norm_conv_bias_to_fp16, epsilon = layer_norm_74_epsilon_0_to_fp16, gamma = p_encoder_layers_12_norm_conv_weight_to_fp16, x = add_79_cast_fp16)[name = tensor<string, []>("layer_norm_74_cast_fp16")];
            tensor<int32, [3]> transpose_194_perm_0 = const()[name = tensor<string, []>("transpose_194_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_38_pad_type_0 = const()[name = tensor<string, []>("conv1d_38_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_38_strides_0 = const()[name = tensor<string, []>("conv1d_38_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_38_pad_0 = const()[name = tensor<string, []>("conv1d_38_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_38_dilations_0 = const()[name = tensor<string, []>("conv1d_38_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_38_groups_0 = const()[name = tensor<string, []>("conv1d_38_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174156352))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175336064))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_12_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175339200)))];
            tensor<fp16, [1, 768, 750]> transpose_194_cast_fp16 = transpose(perm = transpose_194_perm_0, x = layer_norm_74_cast_fp16)[name = tensor<string, []>("transpose_130")];
            tensor<fp16, [1, 1536, 750]> conv1d_38_cast_fp16 = conv(bias = p_encoder_layers_12_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_38_dilations_0, groups = conv1d_38_groups_0, pad = conv1d_38_pad_0, pad_type = conv1d_38_pad_type_0, strides = conv1d_38_strides_0, weight = p_encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_194_cast_fp16)[name = tensor<string, []>("conv1d_38_cast_fp16")];
            tensor<int32, []> glu_12_split_num_splits_0 = const()[name = tensor<string, []>("glu_12_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_12_split_axis_0 = const()[name = tensor<string, []>("glu_12_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_12_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_12_split_cast_fp16_1 = split(axis = glu_12_split_axis_0, num_splits = glu_12_split_num_splits_0, x = conv1d_38_cast_fp16)[name = tensor<string, []>("glu_12_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_12_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_12_split_cast_fp16_1)[name = tensor<string, []>("glu_12_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_12_cast_fp16 = mul(x = glu_12_split_cast_fp16_0, y = glu_12_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_12_cast_fp16")];
            tensor<string, []> conv1d_39_pad_type_0 = const()[name = tensor<string, []>("conv1d_39_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_39_pad_0 = const()[name = tensor<string, []>("conv1d_39_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_39_groups_0 = const()[name = tensor<string, []>("conv1d_39_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_39_strides_0 = const()[name = tensor<string, []>("conv1d_39_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_39_dilations_0 = const()[name = tensor<string, []>("conv1d_39_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_12_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175342336))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175346240))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_12_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175347840)))];
            tensor<fp16, [1, 768, 750]> conv1d_39_cast_fp16 = conv(bias = p_encoder_layers_12_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_39_dilations_0, groups = conv1d_39_groups_0, pad = conv1d_39_pad_0, pad_type = conv1d_39_pad_type_0, strides = conv1d_39_strides_0, weight = p_encoder_layers_12_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_12_cast_fp16)[name = tensor<string, []>("conv1d_39_cast_fp16")];
            tensor<int32, [3]> transpose_195_perm_0 = const()[name = tensor<string, []>("transpose_195_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_75_axes_0 = const()[name = tensor<string, []>("layer_norm_75_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_12_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175349440)))];
            tensor<fp16, [768]> p_encoder_layers_12_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175351040)))];
            tensor<fp16, []> layer_norm_75_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_75_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_195_cast_fp16 = transpose(perm = transpose_195_perm_0, x = conv1d_39_cast_fp16)[name = tensor<string, []>("transpose_129")];
            tensor<fp16, [1, 750, 768]> layer_norm_75_cast_fp16 = layer_norm(axes = layer_norm_75_axes_0, beta = p_encoder_layers_12_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_75_epsilon_0_to_fp16, gamma = p_encoder_layers_12_conv_batch_norm_weight_to_fp16, x = transpose_195_cast_fp16)[name = tensor<string, []>("layer_norm_75_cast_fp16")];
            tensor<int32, [3]> transpose_196_perm_0 = const()[name = tensor<string, []>("transpose_196_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_196_cast_fp16 = transpose(perm = transpose_196_perm_0, x = layer_norm_75_cast_fp16)[name = tensor<string, []>("transpose_128")];
            tensor<fp16, [1, 768, 750]> silu_37_cast_fp16 = silu(x = transpose_196_cast_fp16)[name = tensor<string, []>("silu_37_cast_fp16")];
            tensor<string, []> conv1d_40_pad_type_0 = const()[name = tensor<string, []>("conv1d_40_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_40_strides_0 = const()[name = tensor<string, []>("conv1d_40_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_40_pad_0 = const()[name = tensor<string, []>("conv1d_40_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_40_dilations_0 = const()[name = tensor<string, []>("conv1d_40_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_40_groups_0 = const()[name = tensor<string, []>("conv1d_40_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175352640))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175942528))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_12_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175944128)))];
            tensor<fp16, [1, 768, 750]> conv1d_40_cast_fp16 = conv(bias = p_encoder_layers_12_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_40_dilations_0, groups = conv1d_40_groups_0, pad = conv1d_40_pad_0, pad_type = conv1d_40_pad_type_0, strides = conv1d_40_strides_0, weight = p_encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_37_cast_fp16)[name = tensor<string, []>("conv1d_40_cast_fp16")];
            tensor<int32, [3]> transpose_197_perm_0 = const()[name = tensor<string, []>("transpose_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_197_cast_fp16 = transpose(perm = transpose_197_perm_0, x = conv1d_40_cast_fp16)[name = tensor<string, []>("transpose_127")];
            tensor<fp16, [1, 750, 768]> add_80_cast_fp16 = add(x = add_79_cast_fp16, y = transpose_197_cast_fp16)[name = tensor<string, []>("add_80_cast_fp16")];
            tensor<int32, [1]> layer_norm_76_axes_0 = const()[name = tensor<string, []>("layer_norm_76_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175945728)))];
            tensor<fp16, [768]> p_encoder_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175947328)))];
            tensor<fp16, []> layer_norm_76_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_76_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_76_cast_fp16 = layer_norm(axes = layer_norm_76_axes_0, beta = p_encoder_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_76_epsilon_0_to_fp16, gamma = p_encoder_layers_12_norm_feed_forward2_weight_to_fp16, x = add_80_cast_fp16)[name = tensor<string, []>("layer_norm_76_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_12_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(175948928))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178308288))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_12_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178314496)))];
            tensor<fp16, [1, 750, 3072]> linear_102_cast_fp16 = linear(bias = p_encoder_layers_12_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_12_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_76_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_38_cast_fp16 = silu(x = linear_102_cast_fp16)[name = tensor<string, []>("silu_38_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_12_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_12_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178320704))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180680064))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_12_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180681664)))];
            tensor<fp16, [1, 750, 768]> linear_103_cast_fp16 = linear(bias = p_encoder_layers_12_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_12_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_38_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")];
            tensor<fp16, []> const_1032_to_fp16 = const()[name = tensor<string, []>("const_1032_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_77_cast_fp16 = mul(x = linear_103_cast_fp16, y = const_1032_to_fp16)[name = tensor<string, []>("mul_77_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_81_cast_fp16 = add(x = add_80_cast_fp16, y = mul_77_cast_fp16)[name = tensor<string, []>("add_81_cast_fp16")];
            tensor<int32, [1]> layer_norm_77_axes_0 = const()[name = tensor<string, []>("layer_norm_77_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_12_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180683264)))];
            tensor<fp16, [768]> p_encoder_layers_12_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180684864)))];
            tensor<fp16, []> layer_norm_77_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_77_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_77_cast_fp16 = layer_norm(axes = layer_norm_77_axes_0, beta = p_encoder_layers_12_norm_out_bias_to_fp16, epsilon = layer_norm_77_epsilon_0_to_fp16, gamma = p_encoder_layers_12_norm_out_weight_to_fp16, x = add_81_cast_fp16)[name = tensor<string, []>("layer_norm_77_cast_fp16")];
            tensor<int32, [1]> layer_norm_78_axes_0 = const()[name = tensor<string, []>("layer_norm_78_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180686464)))];
            tensor<fp16, [768]> p_encoder_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180688064)))];
            tensor<fp16, []> layer_norm_78_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_78_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_78_cast_fp16 = layer_norm(axes = layer_norm_78_axes_0, beta = p_encoder_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_78_epsilon_0_to_fp16, gamma = p_encoder_layers_13_norm_feed_forward1_weight_to_fp16, x = layer_norm_77_cast_fp16)[name = tensor<string, []>("layer_norm_78_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_13_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180689664))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183049024))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_13_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183055232)))];
            tensor<fp16, [1, 750, 3072]> linear_104_cast_fp16 = linear(bias = p_encoder_layers_13_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_13_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_78_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_39_cast_fp16 = silu(x = linear_104_cast_fp16)[name = tensor<string, []>("silu_39_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_13_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(183061440))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185420800))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_13_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185422400)))];
            tensor<fp16, [1, 750, 768]> linear_105_cast_fp16 = linear(bias = p_encoder_layers_13_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_13_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_39_cast_fp16)[name = tensor<string, []>("linear_105_cast_fp16")];
            tensor<fp16, []> const_1035_to_fp16 = const()[name = tensor<string, []>("const_1035_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_78_cast_fp16 = mul(x = linear_105_cast_fp16, y = const_1035_to_fp16)[name = tensor<string, []>("mul_78_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_82_cast_fp16 = add(x = layer_norm_77_cast_fp16, y = mul_78_cast_fp16)[name = tensor<string, []>("add_82_cast_fp16")];
            tensor<int32, [1]> layer_norm_79_axes_0 = const()[name = tensor<string, []>("layer_norm_79_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_13_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185424000)))];
            tensor<fp16, [768]> p_encoder_layers_13_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185425600)))];
            tensor<fp16, []> layer_norm_79_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_79_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_79_cast_fp16 = layer_norm(axes = layer_norm_79_axes_0, beta = p_encoder_layers_13_norm_self_att_bias_to_fp16, epsilon = layer_norm_79_epsilon_0_to_fp16, gamma = p_encoder_layers_13_norm_self_att_weight_to_fp16, x = add_82_cast_fp16)[name = tensor<string, []>("layer_norm_79_cast_fp16")];
            tensor<int32, [4]> const_1039 = const()[name = tensor<string, []>("const_1039"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_61_perm_1 = const()[name = tensor<string, []>("transpose_61_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_61 = transpose(perm = transpose_61_perm_1, x = layer_norm_79_cast_fp16)[name = tensor<string, []>("transpose_126")];
            tensor<fp16, [750, 1, 16, 48]> view_117_cast_fp16 = reshape(shape = const_1039, x = transpose_61)[name = tensor<string, []>("view_117_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_79_cast_fp16 = mul(x = view_117_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_79_cast_fp16")];
            tensor<int32, [4]> slice_55_begin_0 = const()[name = tensor<string, []>("slice_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_55_end_0 = const()[name = tensor<string, []>("slice_55_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_55_end_mask_0 = const()[name = tensor<string, []>("slice_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_55_cast_fp16 = slice_by_index(begin = slice_55_begin_0, end = slice_55_end_0, end_mask = slice_55_end_mask_0, x = view_117_cast_fp16)[name = tensor<string, []>("slice_55_cast_fp16")];
            tensor<int32, [4]> slice_56_begin_0 = const()[name = tensor<string, []>("slice_56_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_56_end_0 = const()[name = tensor<string, []>("slice_56_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_56_end_mask_0 = const()[name = tensor<string, []>("slice_56_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_56_cast_fp16 = slice_by_index(begin = slice_56_begin_0, end = slice_56_end_0, end_mask = slice_56_end_mask_0, x = view_117_cast_fp16)[name = tensor<string, []>("slice_56_cast_fp16")];
            tensor<fp16, []> const_1056_promoted_to_fp16 = const()[name = tensor<string, []>("const_1056_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_26_cast_fp16 = mul(x = slice_56_cast_fp16, y = const_1056_promoted_to_fp16)[name = tensor<string, []>("neg_26_cast_fp16")];
            tensor<int32, []> const_1057 = const()[name = tensor<string, []>("const_1057"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_26_interleave_0 = const()[name = tensor<string, []>("cat_26_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_26_cast_fp16 = concat(axis = const_1057, interleave = cat_26_interleave_0, values = (neg_26_cast_fp16, slice_55_cast_fp16))[name = tensor<string, []>("cat_26_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_80_cast_fp16 = mul(x = cat_26_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_80_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_83_cast_fp16 = add(x = mul_79_cast_fp16, y = mul_80_cast_fp16)[name = tensor<string, []>("add_83_cast_fp16")];
            tensor<int32, [3]> const_1066 = const()[name = tensor<string, []>("const_1066"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_120_cast_fp16 = reshape(shape = const_1066, x = add_83_cast_fp16)[name = tensor<string, []>("view_120_cast_fp16")];
            tensor<int32, [3]> transpose_201_perm_0 = const()[name = tensor<string, []>("transpose_201_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_13_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185427200))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186017088))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_13_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186018688)))];
            tensor<fp16, [1, 750, 768]> transpose_201_cast_fp16 = transpose(perm = transpose_201_perm_0, x = view_120_cast_fp16)[name = tensor<string, []>("transpose_125")];
            tensor<fp16, [1, 750, 768]> linear_106_cast_fp16 = linear(bias = p_encoder_layers_13_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_13_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_201_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")];
            tensor<int32, [4]> const_1075 = const()[name = tensor<string, []>("const_1075"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_123_cast_fp16 = reshape(shape = const_1075, x = linear_106_cast_fp16)[name = tensor<string, []>("view_123_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_13_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186020288))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186610176))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_13_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186611776)))];
            tensor<fp16, [1, 750, 768]> linear_107_cast_fp16 = linear(bias = p_encoder_layers_13_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_13_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_201_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")];
            tensor<int32, [4]> const_1076 = const()[name = tensor<string, []>("const_1076"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_124_cast_fp16 = reshape(shape = const_1076, x = linear_107_cast_fp16)[name = tensor<string, []>("view_124_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_13_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(186613376))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187203264))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_13_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187204864)))];
            tensor<fp16, [1, 750, 768]> linear_108_cast_fp16 = linear(bias = p_encoder_layers_13_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_13_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_79_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")];
            tensor<int32, [4]> const_1077 = const()[name = tensor<string, []>("const_1077"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_125_cast_fp16 = reshape(shape = const_1077, x = linear_108_cast_fp16)[name = tensor<string, []>("view_125_cast_fp16")];
            tensor<int32, [4]> transpose_206_perm_0 = const()[name = tensor<string, []>("transpose_206_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_15_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_15_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_15_cast_fp16 = mul(x = view_124_cast_fp16, y = _inversed_div_15_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_15_cast_fp16")];
            tensor<bool, []> matmul_26_transpose_x_0 = const()[name = tensor<string, []>("matmul_26_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_26_transpose_y_0 = const()[name = tensor<string, []>("matmul_26_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_90_perm_0_1 = const()[name = tensor<string, []>("transpose_90_perm_0_1"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_91_perm_0_1 = const()[name = tensor<string, []>("transpose_91_perm_0_1"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_91 = transpose(perm = transpose_91_perm_0_1, x = _inversed_div_15_cast_fp16)[name = tensor<string, []>("transpose_123")];
            tensor<fp16, [1, 16, 750, 48]> transpose_90 = transpose(perm = transpose_90_perm_0_1, x = view_123_cast_fp16)[name = tensor<string, []>("transpose_124")];
            tensor<fp16, [1, 16, 750, 750]> matmul_26_cast_fp16 = matmul(transpose_x = matmul_26_transpose_x_0, transpose_y = matmul_26_transpose_y_0, x = transpose_90, y = transpose_91)[name = tensor<string, []>("matmul_26_cast_fp16")];
            tensor<int32, []> const_1087 = const()[name = tensor<string, []>("const_1087"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_13_cast_fp16 = softmax(axis = const_1087, x = matmul_26_cast_fp16)[name = tensor<string, []>("softmax_13_cast_fp16")];
            tensor<bool, []> matmul_27_transpose_x_0 = const()[name = tensor<string, []>("matmul_27_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_27_transpose_y_0 = const()[name = tensor<string, []>("matmul_27_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_206_cast_fp16 = transpose(perm = transpose_206_perm_0, x = view_125_cast_fp16)[name = tensor<string, []>("transpose_122")];
            tensor<fp16, [1, 16, 750, 48]> matmul_27_cast_fp16 = matmul(transpose_x = matmul_27_transpose_x_0, transpose_y = matmul_27_transpose_y_0, x = softmax_13_cast_fp16, y = transpose_206_cast_fp16)[name = tensor<string, []>("matmul_27_cast_fp16")];
            tensor<int32, [4]> transpose_208_perm_0 = const()[name = tensor<string, []>("transpose_208_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1090 = const()[name = tensor<string, []>("const_1090"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_208_cast_fp16 = transpose(perm = transpose_208_perm_0, x = matmul_27_cast_fp16)[name = tensor<string, []>("transpose_121")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_13_cast_fp16 = reshape(shape = const_1090, x = transpose_208_cast_fp16)[name = tensor<string, []>("_unsafe_view_13_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_13_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187206464))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187796352))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_13_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187797952)))];
            tensor<fp16, [1, 750, 768]> linear_109_cast_fp16 = linear(bias = p_encoder_layers_13_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_13_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_13_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_85_cast_fp16 = add(x = add_82_cast_fp16, y = linear_109_cast_fp16)[name = tensor<string, []>("add_85_cast_fp16")];
            tensor<int32, [1]> layer_norm_80_axes_0 = const()[name = tensor<string, []>("layer_norm_80_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_13_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187799552)))];
            tensor<fp16, [768]> p_encoder_layers_13_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187801152)))];
            tensor<fp16, []> layer_norm_80_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_80_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_80_cast_fp16 = layer_norm(axes = layer_norm_80_axes_0, beta = p_encoder_layers_13_norm_conv_bias_to_fp16, epsilon = layer_norm_80_epsilon_0_to_fp16, gamma = p_encoder_layers_13_norm_conv_weight_to_fp16, x = add_85_cast_fp16)[name = tensor<string, []>("layer_norm_80_cast_fp16")];
            tensor<int32, [3]> transpose_209_perm_0 = const()[name = tensor<string, []>("transpose_209_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_41_pad_type_0 = const()[name = tensor<string, []>("conv1d_41_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_41_strides_0 = const()[name = tensor<string, []>("conv1d_41_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_41_pad_0 = const()[name = tensor<string, []>("conv1d_41_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_41_dilations_0 = const()[name = tensor<string, []>("conv1d_41_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_41_groups_0 = const()[name = tensor<string, []>("conv1d_41_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187802752))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188982464))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_13_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188985600)))];
            tensor<fp16, [1, 768, 750]> transpose_209_cast_fp16 = transpose(perm = transpose_209_perm_0, x = layer_norm_80_cast_fp16)[name = tensor<string, []>("transpose_120")];
            tensor<fp16, [1, 1536, 750]> conv1d_41_cast_fp16 = conv(bias = p_encoder_layers_13_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_41_dilations_0, groups = conv1d_41_groups_0, pad = conv1d_41_pad_0, pad_type = conv1d_41_pad_type_0, strides = conv1d_41_strides_0, weight = p_encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_209_cast_fp16)[name = tensor<string, []>("conv1d_41_cast_fp16")];
            tensor<int32, []> glu_13_split_num_splits_0 = const()[name = tensor<string, []>("glu_13_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_13_split_axis_0 = const()[name = tensor<string, []>("glu_13_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_13_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_13_split_cast_fp16_1 = split(axis = glu_13_split_axis_0, num_splits = glu_13_split_num_splits_0, x = conv1d_41_cast_fp16)[name = tensor<string, []>("glu_13_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_13_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_13_split_cast_fp16_1)[name = tensor<string, []>("glu_13_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_13_cast_fp16 = mul(x = glu_13_split_cast_fp16_0, y = glu_13_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_13_cast_fp16")];
            tensor<string, []> conv1d_42_pad_type_0 = const()[name = tensor<string, []>("conv1d_42_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_42_pad_0 = const()[name = tensor<string, []>("conv1d_42_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_42_groups_0 = const()[name = tensor<string, []>("conv1d_42_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_42_strides_0 = const()[name = tensor<string, []>("conv1d_42_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_42_dilations_0 = const()[name = tensor<string, []>("conv1d_42_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_13_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188988736))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188992640))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_13_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188994240)))];
            tensor<fp16, [1, 768, 750]> conv1d_42_cast_fp16 = conv(bias = p_encoder_layers_13_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_42_dilations_0, groups = conv1d_42_groups_0, pad = conv1d_42_pad_0, pad_type = conv1d_42_pad_type_0, strides = conv1d_42_strides_0, weight = p_encoder_layers_13_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_13_cast_fp16)[name = tensor<string, []>("conv1d_42_cast_fp16")];
            tensor<int32, [3]> transpose_210_perm_0 = const()[name = tensor<string, []>("transpose_210_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_81_axes_0 = const()[name = tensor<string, []>("layer_norm_81_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_13_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188995840)))];
            tensor<fp16, [768]> p_encoder_layers_13_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188997440)))];
            tensor<fp16, []> layer_norm_81_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_81_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_210_cast_fp16 = transpose(perm = transpose_210_perm_0, x = conv1d_42_cast_fp16)[name = tensor<string, []>("transpose_119")];
            tensor<fp16, [1, 750, 768]> layer_norm_81_cast_fp16 = layer_norm(axes = layer_norm_81_axes_0, beta = p_encoder_layers_13_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_81_epsilon_0_to_fp16, gamma = p_encoder_layers_13_conv_batch_norm_weight_to_fp16, x = transpose_210_cast_fp16)[name = tensor<string, []>("layer_norm_81_cast_fp16")];
            tensor<int32, [3]> transpose_211_perm_0 = const()[name = tensor<string, []>("transpose_211_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_211_cast_fp16 = transpose(perm = transpose_211_perm_0, x = layer_norm_81_cast_fp16)[name = tensor<string, []>("transpose_118")];
            tensor<fp16, [1, 768, 750]> silu_40_cast_fp16 = silu(x = transpose_211_cast_fp16)[name = tensor<string, []>("silu_40_cast_fp16")];
            tensor<string, []> conv1d_43_pad_type_0 = const()[name = tensor<string, []>("conv1d_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_43_strides_0 = const()[name = tensor<string, []>("conv1d_43_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_43_pad_0 = const()[name = tensor<string, []>("conv1d_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_43_dilations_0 = const()[name = tensor<string, []>("conv1d_43_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_43_groups_0 = const()[name = tensor<string, []>("conv1d_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(188999040))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189588928))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_13_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189590528)))];
            tensor<fp16, [1, 768, 750]> conv1d_43_cast_fp16 = conv(bias = p_encoder_layers_13_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_43_dilations_0, groups = conv1d_43_groups_0, pad = conv1d_43_pad_0, pad_type = conv1d_43_pad_type_0, strides = conv1d_43_strides_0, weight = p_encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_40_cast_fp16)[name = tensor<string, []>("conv1d_43_cast_fp16")];
            tensor<int32, [3]> transpose_212_perm_0 = const()[name = tensor<string, []>("transpose_212_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_212_cast_fp16 = transpose(perm = transpose_212_perm_0, x = conv1d_43_cast_fp16)[name = tensor<string, []>("transpose_117")];
            tensor<fp16, [1, 750, 768]> add_86_cast_fp16 = add(x = add_85_cast_fp16, y = transpose_212_cast_fp16)[name = tensor<string, []>("add_86_cast_fp16")];
            tensor<int32, [1]> layer_norm_82_axes_0 = const()[name = tensor<string, []>("layer_norm_82_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189592128)))];
            tensor<fp16, [768]> p_encoder_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189593728)))];
            tensor<fp16, []> layer_norm_82_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_82_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_82_cast_fp16 = layer_norm(axes = layer_norm_82_axes_0, beta = p_encoder_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_82_epsilon_0_to_fp16, gamma = p_encoder_layers_13_norm_feed_forward2_weight_to_fp16, x = add_86_cast_fp16)[name = tensor<string, []>("layer_norm_82_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_13_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(189595328))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191954688))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_13_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191960896)))];
            tensor<fp16, [1, 750, 3072]> linear_110_cast_fp16 = linear(bias = p_encoder_layers_13_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_13_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_82_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_41_cast_fp16 = silu(x = linear_110_cast_fp16)[name = tensor<string, []>("silu_41_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_13_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_13_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(191967104))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194326464))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_13_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194328064)))];
            tensor<fp16, [1, 750, 768]> linear_111_cast_fp16 = linear(bias = p_encoder_layers_13_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_13_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_41_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")];
            tensor<fp16, []> const_1109_to_fp16 = const()[name = tensor<string, []>("const_1109_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_83_cast_fp16 = mul(x = linear_111_cast_fp16, y = const_1109_to_fp16)[name = tensor<string, []>("mul_83_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_87_cast_fp16 = add(x = add_86_cast_fp16, y = mul_83_cast_fp16)[name = tensor<string, []>("add_87_cast_fp16")];
            tensor<int32, [1]> layer_norm_83_axes_0 = const()[name = tensor<string, []>("layer_norm_83_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_13_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194329664)))];
            tensor<fp16, [768]> p_encoder_layers_13_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194331264)))];
            tensor<fp16, []> layer_norm_83_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_83_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_83_cast_fp16 = layer_norm(axes = layer_norm_83_axes_0, beta = p_encoder_layers_13_norm_out_bias_to_fp16, epsilon = layer_norm_83_epsilon_0_to_fp16, gamma = p_encoder_layers_13_norm_out_weight_to_fp16, x = add_87_cast_fp16)[name = tensor<string, []>("layer_norm_83_cast_fp16")];
            tensor<int32, [1]> layer_norm_84_axes_0 = const()[name = tensor<string, []>("layer_norm_84_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194332864)))];
            tensor<fp16, [768]> p_encoder_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194334464)))];
            tensor<fp16, []> layer_norm_84_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_84_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_84_cast_fp16 = layer_norm(axes = layer_norm_84_axes_0, beta = p_encoder_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_84_epsilon_0_to_fp16, gamma = p_encoder_layers_14_norm_feed_forward1_weight_to_fp16, x = layer_norm_83_cast_fp16)[name = tensor<string, []>("layer_norm_84_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_14_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(194336064))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196695424))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_14_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196701632)))];
            tensor<fp16, [1, 750, 3072]> linear_112_cast_fp16 = linear(bias = p_encoder_layers_14_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_14_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_84_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_42_cast_fp16 = silu(x = linear_112_cast_fp16)[name = tensor<string, []>("silu_42_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_14_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(196707840))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199067200))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_14_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199068800)))];
            tensor<fp16, [1, 750, 768]> linear_113_cast_fp16 = linear(bias = p_encoder_layers_14_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_14_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_42_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")];
            tensor<fp16, []> const_1112_to_fp16 = const()[name = tensor<string, []>("const_1112_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_84_cast_fp16 = mul(x = linear_113_cast_fp16, y = const_1112_to_fp16)[name = tensor<string, []>("mul_84_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_88_cast_fp16 = add(x = layer_norm_83_cast_fp16, y = mul_84_cast_fp16)[name = tensor<string, []>("add_88_cast_fp16")];
            tensor<int32, [1]> layer_norm_85_axes_0 = const()[name = tensor<string, []>("layer_norm_85_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_14_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199070400)))];
            tensor<fp16, [768]> p_encoder_layers_14_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199072000)))];
            tensor<fp16, []> layer_norm_85_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_85_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_85_cast_fp16 = layer_norm(axes = layer_norm_85_axes_0, beta = p_encoder_layers_14_norm_self_att_bias_to_fp16, epsilon = layer_norm_85_epsilon_0_to_fp16, gamma = p_encoder_layers_14_norm_self_att_weight_to_fp16, x = add_88_cast_fp16)[name = tensor<string, []>("layer_norm_85_cast_fp16")];
            tensor<int32, [4]> const_1116 = const()[name = tensor<string, []>("const_1116"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_62_perm_1 = const()[name = tensor<string, []>("transpose_62_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_62 = transpose(perm = transpose_62_perm_1, x = layer_norm_85_cast_fp16)[name = tensor<string, []>("transpose_116")];
            tensor<fp16, [750, 1, 16, 48]> view_126_cast_fp16 = reshape(shape = const_1116, x = transpose_62)[name = tensor<string, []>("view_126_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_85_cast_fp16 = mul(x = view_126_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_85_cast_fp16")];
            tensor<int32, [4]> slice_59_begin_0 = const()[name = tensor<string, []>("slice_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_59_end_0 = const()[name = tensor<string, []>("slice_59_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_59_end_mask_0 = const()[name = tensor<string, []>("slice_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_59_cast_fp16 = slice_by_index(begin = slice_59_begin_0, end = slice_59_end_0, end_mask = slice_59_end_mask_0, x = view_126_cast_fp16)[name = tensor<string, []>("slice_59_cast_fp16")];
            tensor<int32, [4]> slice_60_begin_0 = const()[name = tensor<string, []>("slice_60_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_60_end_0 = const()[name = tensor<string, []>("slice_60_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_60_end_mask_0 = const()[name = tensor<string, []>("slice_60_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_60_cast_fp16 = slice_by_index(begin = slice_60_begin_0, end = slice_60_end_0, end_mask = slice_60_end_mask_0, x = view_126_cast_fp16)[name = tensor<string, []>("slice_60_cast_fp16")];
            tensor<fp16, []> const_1133_promoted_to_fp16 = const()[name = tensor<string, []>("const_1133_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_28_cast_fp16 = mul(x = slice_60_cast_fp16, y = const_1133_promoted_to_fp16)[name = tensor<string, []>("neg_28_cast_fp16")];
            tensor<int32, []> const_1134 = const()[name = tensor<string, []>("const_1134"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_28_interleave_0 = const()[name = tensor<string, []>("cat_28_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_28_cast_fp16 = concat(axis = const_1134, interleave = cat_28_interleave_0, values = (neg_28_cast_fp16, slice_59_cast_fp16))[name = tensor<string, []>("cat_28_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_86_cast_fp16 = mul(x = cat_28_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_86_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_89_cast_fp16 = add(x = mul_85_cast_fp16, y = mul_86_cast_fp16)[name = tensor<string, []>("add_89_cast_fp16")];
            tensor<int32, [3]> const_1143 = const()[name = tensor<string, []>("const_1143"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_129_cast_fp16 = reshape(shape = const_1143, x = add_89_cast_fp16)[name = tensor<string, []>("view_129_cast_fp16")];
            tensor<int32, [3]> transpose_216_perm_0 = const()[name = tensor<string, []>("transpose_216_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_14_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199073600))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199663488))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_14_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199665088)))];
            tensor<fp16, [1, 750, 768]> transpose_216_cast_fp16 = transpose(perm = transpose_216_perm_0, x = view_129_cast_fp16)[name = tensor<string, []>("transpose_115")];
            tensor<fp16, [1, 750, 768]> linear_114_cast_fp16 = linear(bias = p_encoder_layers_14_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_14_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_216_cast_fp16)[name = tensor<string, []>("linear_114_cast_fp16")];
            tensor<int32, [4]> const_1152 = const()[name = tensor<string, []>("const_1152"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_132_cast_fp16 = reshape(shape = const_1152, x = linear_114_cast_fp16)[name = tensor<string, []>("view_132_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_14_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(199666688))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200256576))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_14_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200258176)))];
            tensor<fp16, [1, 750, 768]> linear_115_cast_fp16 = linear(bias = p_encoder_layers_14_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_14_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_216_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")];
            tensor<int32, [4]> const_1153 = const()[name = tensor<string, []>("const_1153"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_133_cast_fp16 = reshape(shape = const_1153, x = linear_115_cast_fp16)[name = tensor<string, []>("view_133_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_14_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200259776))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200849664))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_14_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200851264)))];
            tensor<fp16, [1, 750, 768]> linear_116_cast_fp16 = linear(bias = p_encoder_layers_14_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_14_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_85_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")];
            tensor<int32, [4]> const_1154 = const()[name = tensor<string, []>("const_1154"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_134_cast_fp16 = reshape(shape = const_1154, x = linear_116_cast_fp16)[name = tensor<string, []>("view_134_cast_fp16")];
            tensor<int32, [4]> transpose_221_perm_0 = const()[name = tensor<string, []>("transpose_221_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_16_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_16_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_16_cast_fp16 = mul(x = view_133_cast_fp16, y = _inversed_div_16_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_16_cast_fp16")];
            tensor<bool, []> matmul_28_transpose_x_0 = const()[name = tensor<string, []>("matmul_28_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_28_transpose_y_0 = const()[name = tensor<string, []>("matmul_28_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_92_perm_0_1 = const()[name = tensor<string, []>("transpose_92_perm_0_1"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_93_perm_0 = const()[name = tensor<string, []>("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_93 = transpose(perm = transpose_93_perm_0, x = _inversed_div_16_cast_fp16)[name = tensor<string, []>("transpose_113")];
            tensor<fp16, [1, 16, 750, 48]> transpose_92 = transpose(perm = transpose_92_perm_0_1, x = view_132_cast_fp16)[name = tensor<string, []>("transpose_114")];
            tensor<fp16, [1, 16, 750, 750]> matmul_28_cast_fp16 = matmul(transpose_x = matmul_28_transpose_x_0, transpose_y = matmul_28_transpose_y_0, x = transpose_92, y = transpose_93)[name = tensor<string, []>("matmul_28_cast_fp16")];
            tensor<int32, []> const_1164 = const()[name = tensor<string, []>("const_1164"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_14_cast_fp16 = softmax(axis = const_1164, x = matmul_28_cast_fp16)[name = tensor<string, []>("softmax_14_cast_fp16")];
            tensor<bool, []> matmul_29_transpose_x_0 = const()[name = tensor<string, []>("matmul_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_29_transpose_y_0 = const()[name = tensor<string, []>("matmul_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_221_cast_fp16 = transpose(perm = transpose_221_perm_0, x = view_134_cast_fp16)[name = tensor<string, []>("transpose_112")];
            tensor<fp16, [1, 16, 750, 48]> matmul_29_cast_fp16 = matmul(transpose_x = matmul_29_transpose_x_0, transpose_y = matmul_29_transpose_y_0, x = softmax_14_cast_fp16, y = transpose_221_cast_fp16)[name = tensor<string, []>("matmul_29_cast_fp16")];
            tensor<int32, [4]> transpose_223_perm_0 = const()[name = tensor<string, []>("transpose_223_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1167 = const()[name = tensor<string, []>("const_1167"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_223_cast_fp16 = transpose(perm = transpose_223_perm_0, x = matmul_29_cast_fp16)[name = tensor<string, []>("transpose_111")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_14_cast_fp16 = reshape(shape = const_1167, x = transpose_223_cast_fp16)[name = tensor<string, []>("_unsafe_view_14_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_14_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(200852864))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201442752))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_14_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201444352)))];
            tensor<fp16, [1, 750, 768]> linear_117_cast_fp16 = linear(bias = p_encoder_layers_14_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_14_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_14_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_91_cast_fp16 = add(x = add_88_cast_fp16, y = linear_117_cast_fp16)[name = tensor<string, []>("add_91_cast_fp16")];
            tensor<int32, [1]> layer_norm_86_axes_0 = const()[name = tensor<string, []>("layer_norm_86_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_14_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201445952)))];
            tensor<fp16, [768]> p_encoder_layers_14_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201447552)))];
            tensor<fp16, []> layer_norm_86_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_86_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_86_cast_fp16 = layer_norm(axes = layer_norm_86_axes_0, beta = p_encoder_layers_14_norm_conv_bias_to_fp16, epsilon = layer_norm_86_epsilon_0_to_fp16, gamma = p_encoder_layers_14_norm_conv_weight_to_fp16, x = add_91_cast_fp16)[name = tensor<string, []>("layer_norm_86_cast_fp16")];
            tensor<int32, [3]> transpose_224_perm_0 = const()[name = tensor<string, []>("transpose_224_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_44_pad_type_0 = const()[name = tensor<string, []>("conv1d_44_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_44_strides_0 = const()[name = tensor<string, []>("conv1d_44_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_44_pad_0 = const()[name = tensor<string, []>("conv1d_44_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_44_dilations_0 = const()[name = tensor<string, []>("conv1d_44_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_44_groups_0 = const()[name = tensor<string, []>("conv1d_44_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(201449152))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202628864))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_14_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202632000)))];
            tensor<fp16, [1, 768, 750]> transpose_224_cast_fp16 = transpose(perm = transpose_224_perm_0, x = layer_norm_86_cast_fp16)[name = tensor<string, []>("transpose_110")];
            tensor<fp16, [1, 1536, 750]> conv1d_44_cast_fp16 = conv(bias = p_encoder_layers_14_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_44_dilations_0, groups = conv1d_44_groups_0, pad = conv1d_44_pad_0, pad_type = conv1d_44_pad_type_0, strides = conv1d_44_strides_0, weight = p_encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_224_cast_fp16)[name = tensor<string, []>("conv1d_44_cast_fp16")];
            tensor<int32, []> glu_14_split_num_splits_0 = const()[name = tensor<string, []>("glu_14_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_14_split_axis_0 = const()[name = tensor<string, []>("glu_14_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_14_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_14_split_cast_fp16_1 = split(axis = glu_14_split_axis_0, num_splits = glu_14_split_num_splits_0, x = conv1d_44_cast_fp16)[name = tensor<string, []>("glu_14_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_14_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_14_split_cast_fp16_1)[name = tensor<string, []>("glu_14_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_14_cast_fp16 = mul(x = glu_14_split_cast_fp16_0, y = glu_14_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_14_cast_fp16")];
            tensor<string, []> conv1d_45_pad_type_0 = const()[name = tensor<string, []>("conv1d_45_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_45_pad_0 = const()[name = tensor<string, []>("conv1d_45_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_45_groups_0 = const()[name = tensor<string, []>("conv1d_45_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_45_strides_0 = const()[name = tensor<string, []>("conv1d_45_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_45_dilations_0 = const()[name = tensor<string, []>("conv1d_45_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_14_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202635136))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202639040))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_14_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202640640)))];
            tensor<fp16, [1, 768, 750]> conv1d_45_cast_fp16 = conv(bias = p_encoder_layers_14_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_45_dilations_0, groups = conv1d_45_groups_0, pad = conv1d_45_pad_0, pad_type = conv1d_45_pad_type_0, strides = conv1d_45_strides_0, weight = p_encoder_layers_14_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_14_cast_fp16)[name = tensor<string, []>("conv1d_45_cast_fp16")];
            tensor<int32, [3]> transpose_225_perm_0 = const()[name = tensor<string, []>("transpose_225_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_87_axes_0 = const()[name = tensor<string, []>("layer_norm_87_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_14_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202642240)))];
            tensor<fp16, [768]> p_encoder_layers_14_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202643840)))];
            tensor<fp16, []> layer_norm_87_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_87_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_225_cast_fp16 = transpose(perm = transpose_225_perm_0, x = conv1d_45_cast_fp16)[name = tensor<string, []>("transpose_109")];
            tensor<fp16, [1, 750, 768]> layer_norm_87_cast_fp16 = layer_norm(axes = layer_norm_87_axes_0, beta = p_encoder_layers_14_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_87_epsilon_0_to_fp16, gamma = p_encoder_layers_14_conv_batch_norm_weight_to_fp16, x = transpose_225_cast_fp16)[name = tensor<string, []>("layer_norm_87_cast_fp16")];
            tensor<int32, [3]> transpose_226_perm_0 = const()[name = tensor<string, []>("transpose_226_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_226_cast_fp16 = transpose(perm = transpose_226_perm_0, x = layer_norm_87_cast_fp16)[name = tensor<string, []>("transpose_108")];
            tensor<fp16, [1, 768, 750]> silu_43_cast_fp16 = silu(x = transpose_226_cast_fp16)[name = tensor<string, []>("silu_43_cast_fp16")];
            tensor<string, []> conv1d_46_pad_type_0 = const()[name = tensor<string, []>("conv1d_46_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_46_strides_0 = const()[name = tensor<string, []>("conv1d_46_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_46_pad_0 = const()[name = tensor<string, []>("conv1d_46_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_46_dilations_0 = const()[name = tensor<string, []>("conv1d_46_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_46_groups_0 = const()[name = tensor<string, []>("conv1d_46_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(202645440))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203235328))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_14_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203236928)))];
            tensor<fp16, [1, 768, 750]> conv1d_46_cast_fp16 = conv(bias = p_encoder_layers_14_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_46_dilations_0, groups = conv1d_46_groups_0, pad = conv1d_46_pad_0, pad_type = conv1d_46_pad_type_0, strides = conv1d_46_strides_0, weight = p_encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_43_cast_fp16)[name = tensor<string, []>("conv1d_46_cast_fp16")];
            tensor<int32, [3]> transpose_227_perm_0 = const()[name = tensor<string, []>("transpose_227_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_227_cast_fp16 = transpose(perm = transpose_227_perm_0, x = conv1d_46_cast_fp16)[name = tensor<string, []>("transpose_107")];
            tensor<fp16, [1, 750, 768]> add_92_cast_fp16 = add(x = add_91_cast_fp16, y = transpose_227_cast_fp16)[name = tensor<string, []>("add_92_cast_fp16")];
            tensor<int32, [1]> layer_norm_88_axes_0 = const()[name = tensor<string, []>("layer_norm_88_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203238528)))];
            tensor<fp16, [768]> p_encoder_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203240128)))];
            tensor<fp16, []> layer_norm_88_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_88_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_88_cast_fp16 = layer_norm(axes = layer_norm_88_axes_0, beta = p_encoder_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_88_epsilon_0_to_fp16, gamma = p_encoder_layers_14_norm_feed_forward2_weight_to_fp16, x = add_92_cast_fp16)[name = tensor<string, []>("layer_norm_88_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_14_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203241728))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205601088))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_14_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205607296)))];
            tensor<fp16, [1, 750, 3072]> linear_118_cast_fp16 = linear(bias = p_encoder_layers_14_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_14_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_88_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_44_cast_fp16 = silu(x = linear_118_cast_fp16)[name = tensor<string, []>("silu_44_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_14_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_14_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(205613504))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207972864))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_14_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207974464)))];
            tensor<fp16, [1, 750, 768]> linear_119_cast_fp16 = linear(bias = p_encoder_layers_14_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_14_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_44_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")];
            tensor<fp16, []> const_1186_to_fp16 = const()[name = tensor<string, []>("const_1186_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_89_cast_fp16 = mul(x = linear_119_cast_fp16, y = const_1186_to_fp16)[name = tensor<string, []>("mul_89_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_93_cast_fp16 = add(x = add_92_cast_fp16, y = mul_89_cast_fp16)[name = tensor<string, []>("add_93_cast_fp16")];
            tensor<int32, [1]> layer_norm_89_axes_0 = const()[name = tensor<string, []>("layer_norm_89_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_14_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207976064)))];
            tensor<fp16, [768]> p_encoder_layers_14_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207977664)))];
            tensor<fp16, []> layer_norm_89_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_89_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_89_cast_fp16 = layer_norm(axes = layer_norm_89_axes_0, beta = p_encoder_layers_14_norm_out_bias_to_fp16, epsilon = layer_norm_89_epsilon_0_to_fp16, gamma = p_encoder_layers_14_norm_out_weight_to_fp16, x = add_93_cast_fp16)[name = tensor<string, []>("layer_norm_89_cast_fp16")];
            tensor<int32, [1]> layer_norm_90_axes_0 = const()[name = tensor<string, []>("layer_norm_90_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207979264)))];
            tensor<fp16, [768]> p_encoder_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207980864)))];
            tensor<fp16, []> layer_norm_90_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_90_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_90_cast_fp16 = layer_norm(axes = layer_norm_90_axes_0, beta = p_encoder_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = layer_norm_90_epsilon_0_to_fp16, gamma = p_encoder_layers_15_norm_feed_forward1_weight_to_fp16, x = layer_norm_89_cast_fp16)[name = tensor<string, []>("layer_norm_90_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_15_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(207982464))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210341824))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_15_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210348032)))];
            tensor<fp16, [1, 750, 3072]> linear_120_cast_fp16 = linear(bias = p_encoder_layers_15_feed_forward1_linear1_bias_to_fp16, weight = p_encoder_layers_15_feed_forward1_linear1_weight_to_fp16_quantized, x = layer_norm_90_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_45_cast_fp16 = silu(x = linear_120_cast_fp16)[name = tensor<string, []>("silu_45_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_15_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(210354240))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212713600))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_15_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212715200)))];
            tensor<fp16, [1, 750, 768]> linear_121_cast_fp16 = linear(bias = p_encoder_layers_15_feed_forward1_linear2_bias_to_fp16, weight = p_encoder_layers_15_feed_forward1_linear2_weight_to_fp16_quantized, x = silu_45_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")];
            tensor<fp16, []> const_1189_to_fp16 = const()[name = tensor<string, []>("const_1189_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_90_cast_fp16 = mul(x = linear_121_cast_fp16, y = const_1189_to_fp16)[name = tensor<string, []>("mul_90_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_94_cast_fp16 = add(x = layer_norm_89_cast_fp16, y = mul_90_cast_fp16)[name = tensor<string, []>("add_94_cast_fp16")];
            tensor<int32, [1]> layer_norm_91_axes_0 = const()[name = tensor<string, []>("layer_norm_91_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_15_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212716800)))];
            tensor<fp16, [768]> p_encoder_layers_15_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212718400)))];
            tensor<fp16, []> layer_norm_91_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_91_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_91_cast_fp16 = layer_norm(axes = layer_norm_91_axes_0, beta = p_encoder_layers_15_norm_self_att_bias_to_fp16, epsilon = layer_norm_91_epsilon_0_to_fp16, gamma = p_encoder_layers_15_norm_self_att_weight_to_fp16, x = add_94_cast_fp16)[name = tensor<string, []>("layer_norm_91_cast_fp16")];
            tensor<int32, [4]> const_1193 = const()[name = tensor<string, []>("const_1193"), val = tensor<int32, [4]>([750, 1, 16, 48])];
            tensor<int32, [3]> transpose_63_perm_1 = const()[name = tensor<string, []>("transpose_63_perm_1"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [750, 1, 768]> transpose_63 = transpose(perm = transpose_63_perm_1, x = layer_norm_91_cast_fp16)[name = tensor<string, []>("transpose_106")];
            tensor<fp16, [750, 1, 16, 48]> view_135_cast_fp16 = reshape(shape = const_1193, x = transpose_63)[name = tensor<string, []>("view_135_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_91_cast_fp16 = mul(x = view_135_cast_fp16, y = const_46_to_fp16_quantized)[name = tensor<string, []>("mul_91_cast_fp16")];
            tensor<int32, [4]> slice_63_begin_0 = const()[name = tensor<string, []>("slice_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> slice_63_end_0 = const()[name = tensor<string, []>("slice_63_end_0"), val = tensor<int32, [4]>([750, 1, 16, 24])];
            tensor<bool, [4]> slice_63_end_mask_0 = const()[name = tensor<string, []>("slice_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [750, 1, 16, 24]> slice_63_cast_fp16 = slice_by_index(begin = slice_63_begin_0, end = slice_63_end_0, end_mask = slice_63_end_mask_0, x = view_135_cast_fp16)[name = tensor<string, []>("slice_63_cast_fp16")];
            tensor<int32, [4]> slice_64_begin_0 = const()[name = tensor<string, []>("slice_64_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 24])];
            tensor<int32, [4]> slice_64_end_0 = const()[name = tensor<string, []>("slice_64_end_0"), val = tensor<int32, [4]>([750, 1, 16, 1])];
            tensor<bool, [4]> slice_64_end_mask_0 = const()[name = tensor<string, []>("slice_64_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [750, 1, 16, 24]> slice_64_cast_fp16 = slice_by_index(begin = slice_64_begin_0, end = slice_64_end_0, end_mask = slice_64_end_mask_0, x = view_135_cast_fp16)[name = tensor<string, []>("slice_64_cast_fp16")];
            tensor<fp16, []> const_1210_promoted_to_fp16 = const()[name = tensor<string, []>("const_1210_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
            tensor<fp16, [750, 1, 16, 24]> neg_30_cast_fp16 = mul(x = slice_64_cast_fp16, y = const_1210_promoted_to_fp16)[name = tensor<string, []>("neg_30_cast_fp16")];
            tensor<int32, []> const_1211 = const()[name = tensor<string, []>("const_1211"), val = tensor<int32, []>(3)];
            tensor<bool, []> cat_30_interleave_0 = const()[name = tensor<string, []>("cat_30_interleave_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [750, 1, 16, 48]> cat_30_cast_fp16 = concat(axis = const_1211, interleave = cat_30_interleave_0, values = (neg_30_cast_fp16, slice_63_cast_fp16))[name = tensor<string, []>("cat_30_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> mul_92_cast_fp16 = mul(x = cat_30_cast_fp16, y = const_48_to_fp16_quantized)[name = tensor<string, []>("mul_92_cast_fp16")];
            tensor<fp16, [750, 1, 16, 48]> add_95_cast_fp16 = add(x = mul_91_cast_fp16, y = mul_92_cast_fp16)[name = tensor<string, []>("add_95_cast_fp16")];
            tensor<int32, [3]> const_1220 = const()[name = tensor<string, []>("const_1220"), val = tensor<int32, [3]>([750, 1, 768])];
            tensor<fp16, [750, 1, 768]> view_138_cast_fp16 = reshape(shape = const_1220, x = add_95_cast_fp16)[name = tensor<string, []>("view_138_cast_fp16")];
            tensor<int32, [3]> transpose_231_perm_0 = const()[name = tensor<string, []>("transpose_231_perm_0"), val = tensor<int32, [3]>([1, 0, 2])];
            tensor<fp16, [768, 768]> p_encoder_layers_15_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212720000))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213309888))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_15_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213311488)))];
            tensor<fp16, [1, 750, 768]> transpose_231_cast_fp16 = transpose(perm = transpose_231_perm_0, x = view_138_cast_fp16)[name = tensor<string, []>("transpose_105")];
            tensor<fp16, [1, 750, 768]> linear_122_cast_fp16 = linear(bias = p_encoder_layers_15_self_attn_linear_q_bias_to_fp16, weight = p_encoder_layers_15_self_attn_linear_q_weight_to_fp16_quantized, x = transpose_231_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")];
            tensor<int32, [4]> const_1229 = const()[name = tensor<string, []>("const_1229"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_141_cast_fp16 = reshape(shape = const_1229, x = linear_122_cast_fp16)[name = tensor<string, []>("view_141_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_15_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213313088))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213902976))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_15_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213904576)))];
            tensor<fp16, [1, 750, 768]> linear_123_cast_fp16 = linear(bias = p_encoder_layers_15_self_attn_linear_k_bias_to_fp16, weight = p_encoder_layers_15_self_attn_linear_k_weight_to_fp16_quantized, x = transpose_231_cast_fp16)[name = tensor<string, []>("linear_123_cast_fp16")];
            tensor<int32, [4]> const_1230 = const()[name = tensor<string, []>("const_1230"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_142_cast_fp16 = reshape(shape = const_1230, x = linear_123_cast_fp16)[name = tensor<string, []>("view_142_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_15_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(213906176))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214496064))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_15_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214497664)))];
            tensor<fp16, [1, 750, 768]> linear_124_cast_fp16 = linear(bias = p_encoder_layers_15_self_attn_linear_v_bias_to_fp16, weight = p_encoder_layers_15_self_attn_linear_v_weight_to_fp16_quantized, x = layer_norm_91_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")];
            tensor<int32, [4]> const_1231 = const()[name = tensor<string, []>("const_1231"), val = tensor<int32, [4]>([1, -1, 16, 48])];
            tensor<fp16, [1, 750, 16, 48]> view_143_cast_fp16 = reshape(shape = const_1231, x = linear_124_cast_fp16)[name = tensor<string, []>("view_143_cast_fp16")];
            tensor<int32, [4]> transpose_236_perm_0 = const()[name = tensor<string, []>("transpose_236_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, []> _inversed_div_17_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_div_17_y_0_to_fp16"), val = tensor<fp16, []>(0x1.278p-3)];
            tensor<fp16, [1, 750, 16, 48]> _inversed_div_17_cast_fp16 = mul(x = view_142_cast_fp16, y = _inversed_div_17_y_0_to_fp16)[name = tensor<string, []>("_inversed_div_17_cast_fp16")];
            tensor<bool, []> matmul_30_transpose_x_0 = const()[name = tensor<string, []>("matmul_30_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_30_transpose_y_0 = const()[name = tensor<string, []>("matmul_30_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_94_perm_0 = const()[name = tensor<string, []>("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_95_perm_0 = const()[name = tensor<string, []>("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 16, 48, 750]> transpose_95 = transpose(perm = transpose_95_perm_0, x = _inversed_div_17_cast_fp16)[name = tensor<string, []>("transpose_103")];
            tensor<fp16, [1, 16, 750, 48]> transpose_94 = transpose(perm = transpose_94_perm_0, x = view_141_cast_fp16)[name = tensor<string, []>("transpose_104")];
            tensor<fp16, [1, 16, 750, 750]> matmul_30_cast_fp16 = matmul(transpose_x = matmul_30_transpose_x_0, transpose_y = matmul_30_transpose_y_0, x = transpose_94, y = transpose_95)[name = tensor<string, []>("matmul_30_cast_fp16")];
            tensor<int32, []> const_1241 = const()[name = tensor<string, []>("const_1241"), val = tensor<int32, []>(-1)];
            tensor<fp16, [1, 16, 750, 750]> softmax_15_cast_fp16 = softmax(axis = const_1241, x = matmul_30_cast_fp16)[name = tensor<string, []>("softmax_15_cast_fp16")];
            tensor<bool, []> matmul_31_transpose_x_0 = const()[name = tensor<string, []>("matmul_31_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matmul_31_transpose_y_0 = const()[name = tensor<string, []>("matmul_31_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 16, 750, 48]> transpose_236_cast_fp16 = transpose(perm = transpose_236_perm_0, x = view_143_cast_fp16)[name = tensor<string, []>("transpose_102")];
            tensor<fp16, [1, 16, 750, 48]> matmul_31_cast_fp16 = matmul(transpose_x = matmul_31_transpose_x_0, transpose_y = matmul_31_transpose_y_0, x = softmax_15_cast_fp16, y = transpose_236_cast_fp16)[name = tensor<string, []>("matmul_31_cast_fp16")];
            tensor<int32, [4]> transpose_238_perm_0 = const()[name = tensor<string, []>("transpose_238_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> const_1244 = const()[name = tensor<string, []>("const_1244"), val = tensor<int32, [3]>([1, 750, 768])];
            tensor<fp16, [1, 750, 16, 48]> transpose_238_cast_fp16 = transpose(perm = transpose_238_perm_0, x = matmul_31_cast_fp16)[name = tensor<string, []>("transpose_101")];
            tensor<fp16, [1, 750, 768]> _unsafe_view_15_cast_fp16 = reshape(shape = const_1244, x = transpose_238_cast_fp16)[name = tensor<string, []>("_unsafe_view_15_cast_fp16")];
            tensor<fp16, [768, 768]> p_encoder_layers_15_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214499264))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215089152))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_15_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215090752)))];
            tensor<fp16, [1, 750, 768]> linear_125_cast_fp16 = linear(bias = p_encoder_layers_15_self_attn_linear_out_bias_to_fp16, weight = p_encoder_layers_15_self_attn_linear_out_weight_to_fp16_quantized, x = _unsafe_view_15_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_97_cast_fp16 = add(x = add_94_cast_fp16, y = linear_125_cast_fp16)[name = tensor<string, []>("add_97_cast_fp16")];
            tensor<int32, [1]> layer_norm_92_axes_0 = const()[name = tensor<string, []>("layer_norm_92_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_15_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215092352)))];
            tensor<fp16, [768]> p_encoder_layers_15_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215093952)))];
            tensor<fp16, []> layer_norm_92_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_92_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_92_cast_fp16 = layer_norm(axes = layer_norm_92_axes_0, beta = p_encoder_layers_15_norm_conv_bias_to_fp16, epsilon = layer_norm_92_epsilon_0_to_fp16, gamma = p_encoder_layers_15_norm_conv_weight_to_fp16, x = add_97_cast_fp16)[name = tensor<string, []>("layer_norm_92_cast_fp16")];
            tensor<int32, [3]> transpose_239_perm_0 = const()[name = tensor<string, []>("transpose_239_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> conv1d_47_pad_type_0 = const()[name = tensor<string, []>("conv1d_47_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_47_strides_0 = const()[name = tensor<string, []>("conv1d_47_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_47_pad_0 = const()[name = tensor<string, []>("conv1d_47_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_47_dilations_0 = const()[name = tensor<string, []>("conv1d_47_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_47_groups_0 = const()[name = tensor<string, []>("conv1d_47_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1536, 768, 1]> p_encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1536, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(215095552))), scale = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216275264))), zero_point = tensor<int8, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11577664)))];
            tensor<fp16, [1536]> p_encoder_layers_15_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216278400)))];
            tensor<fp16, [1, 768, 750]> transpose_239_cast_fp16 = transpose(perm = transpose_239_perm_0, x = layer_norm_92_cast_fp16)[name = tensor<string, []>("transpose_100")];
            tensor<fp16, [1, 1536, 750]> conv1d_47_cast_fp16 = conv(bias = p_encoder_layers_15_conv_pointwise_conv1_bias_to_fp16, dilations = conv1d_47_dilations_0, groups = conv1d_47_groups_0, pad = conv1d_47_pad_0, pad_type = conv1d_47_pad_type_0, strides = conv1d_47_strides_0, weight = p_encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized, x = transpose_239_cast_fp16)[name = tensor<string, []>("conv1d_47_cast_fp16")];
            tensor<int32, []> glu_15_split_num_splits_0 = const()[name = tensor<string, []>("glu_15_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> glu_15_split_axis_0 = const()[name = tensor<string, []>("glu_15_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 768, 750]> glu_15_split_cast_fp16_0, tensor<fp16, [1, 768, 750]> glu_15_split_cast_fp16_1 = split(axis = glu_15_split_axis_0, num_splits = glu_15_split_num_splits_0, x = conv1d_47_cast_fp16)[name = tensor<string, []>("glu_15_split_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_15_split_1_sigmoid_cast_fp16 = sigmoid(x = glu_15_split_cast_fp16_1)[name = tensor<string, []>("glu_15_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 768, 750]> glu_15_cast_fp16 = mul(x = glu_15_split_cast_fp16_0, y = glu_15_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("glu_15_cast_fp16")];
            tensor<string, []> conv1d_48_pad_type_0 = const()[name = tensor<string, []>("conv1d_48_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [2]> conv1d_48_pad_0 = const()[name = tensor<string, []>("conv1d_48_pad_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> conv1d_48_groups_0 = const()[name = tensor<string, []>("conv1d_48_groups_0"), val = tensor<int32, []>(768)];
            tensor<int32, [1]> conv1d_48_strides_0 = const()[name = tensor<string, []>("conv1d_48_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> conv1d_48_dilations_0 = const()[name = tensor<string, []>("conv1d_48_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [768, 1, 5]> p_encoder_layers_15_conv_depthwise_conv_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_conv_depthwise_conv_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 1, 5]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216281536))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216285440))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_15_conv_depthwise_conv_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_conv_depthwise_conv_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216287040)))];
            tensor<fp16, [1, 768, 750]> conv1d_48_cast_fp16 = conv(bias = p_encoder_layers_15_conv_depthwise_conv_bias_to_fp16, dilations = conv1d_48_dilations_0, groups = conv1d_48_groups_0, pad = conv1d_48_pad_0, pad_type = conv1d_48_pad_type_0, strides = conv1d_48_strides_0, weight = p_encoder_layers_15_conv_depthwise_conv_weight_to_fp16_quantized, x = glu_15_cast_fp16)[name = tensor<string, []>("conv1d_48_cast_fp16")];
            tensor<int32, [3]> transpose_240_perm_0 = const()[name = tensor<string, []>("transpose_240_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<int32, [1]> layer_norm_93_axes_0 = const()[name = tensor<string, []>("layer_norm_93_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_15_conv_batch_norm_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_conv_batch_norm_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216288640)))];
            tensor<fp16, [768]> p_encoder_layers_15_conv_batch_norm_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_conv_batch_norm_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216290240)))];
            tensor<fp16, []> layer_norm_93_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_93_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> transpose_240_cast_fp16 = transpose(perm = transpose_240_perm_0, x = conv1d_48_cast_fp16)[name = tensor<string, []>("transpose_99")];
            tensor<fp16, [1, 750, 768]> layer_norm_93_cast_fp16 = layer_norm(axes = layer_norm_93_axes_0, beta = p_encoder_layers_15_conv_batch_norm_bias_to_fp16, epsilon = layer_norm_93_epsilon_0_to_fp16, gamma = p_encoder_layers_15_conv_batch_norm_weight_to_fp16, x = transpose_240_cast_fp16)[name = tensor<string, []>("layer_norm_93_cast_fp16")];
            tensor<int32, [3]> transpose_241_perm_0 = const()[name = tensor<string, []>("transpose_241_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> transpose_241_cast_fp16 = transpose(perm = transpose_241_perm_0, x = layer_norm_93_cast_fp16)[name = tensor<string, []>("transpose_98")];
            tensor<fp16, [1, 768, 750]> silu_46_cast_fp16 = silu(x = transpose_241_cast_fp16)[name = tensor<string, []>("silu_46_cast_fp16")];
            tensor<string, []> conv1d_49_pad_type_0 = const()[name = tensor<string, []>("conv1d_49_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> conv1d_49_strides_0 = const()[name = tensor<string, []>("conv1d_49_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> conv1d_49_pad_0 = const()[name = tensor<string, []>("conv1d_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> conv1d_49_dilations_0 = const()[name = tensor<string, []>("conv1d_49_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> conv1d_49_groups_0 = const()[name = tensor<string, []>("conv1d_49_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [768, 768, 1]> p_encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 768, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216291840))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216881728))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_15_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216883328)))];
            tensor<fp16, [1, 768, 750]> conv1d_49_cast_fp16 = conv(bias = p_encoder_layers_15_conv_pointwise_conv2_bias_to_fp16, dilations = conv1d_49_dilations_0, groups = conv1d_49_groups_0, pad = conv1d_49_pad_0, pad_type = conv1d_49_pad_type_0, strides = conv1d_49_strides_0, weight = p_encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized, x = silu_46_cast_fp16)[name = tensor<string, []>("conv1d_49_cast_fp16")];
            tensor<int32, [3]> transpose_242_perm_0 = const()[name = tensor<string, []>("transpose_242_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 750, 768]> transpose_242_cast_fp16 = transpose(perm = transpose_242_perm_0, x = conv1d_49_cast_fp16)[name = tensor<string, []>("transpose_97")];
            tensor<fp16, [1, 750, 768]> add_98_cast_fp16 = add(x = add_97_cast_fp16, y = transpose_242_cast_fp16)[name = tensor<string, []>("add_98_cast_fp16")];
            tensor<int32, [1]> layer_norm_94_axes_0 = const()[name = tensor<string, []>("layer_norm_94_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216884928)))];
            tensor<fp16, [768]> p_encoder_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216886528)))];
            tensor<fp16, []> layer_norm_94_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_94_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_94_cast_fp16 = layer_norm(axes = layer_norm_94_axes_0, beta = p_encoder_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = layer_norm_94_epsilon_0_to_fp16, gamma = p_encoder_layers_15_norm_feed_forward2_weight_to_fp16, x = add_98_cast_fp16)[name = tensor<string, []>("layer_norm_94_cast_fp16")];
            tensor<fp16, [3072, 768]> p_encoder_layers_15_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(216888128))), scale = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219247488))), zero_point = tensor<int8, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5564864)))];
            tensor<fp16, [3072]> p_encoder_layers_15_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219253696)))];
            tensor<fp16, [1, 750, 3072]> linear_126_cast_fp16 = linear(bias = p_encoder_layers_15_feed_forward2_linear1_bias_to_fp16, weight = p_encoder_layers_15_feed_forward2_linear1_weight_to_fp16_quantized, x = layer_norm_94_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")];
            tensor<fp16, [1, 750, 3072]> silu_47_cast_fp16 = silu(x = linear_126_cast_fp16)[name = tensor<string, []>("silu_47_cast_fp16")];
            tensor<fp16, [768, 3072]> p_encoder_layers_15_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("p_encoder_layers_15_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(219259904))), scale = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221619264))), zero_point = tensor<int8, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(245888)))];
            tensor<fp16, [768]> p_encoder_layers_15_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221620864)))];
            tensor<fp16, [1, 750, 768]> linear_127_cast_fp16 = linear(bias = p_encoder_layers_15_feed_forward2_linear2_bias_to_fp16, weight = p_encoder_layers_15_feed_forward2_linear2_weight_to_fp16_quantized, x = silu_47_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")];
            tensor<fp16, []> const_1263_to_fp16 = const()[name = tensor<string, []>("const_1263_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 750, 768]> mul_95_cast_fp16 = mul(x = linear_127_cast_fp16, y = const_1263_to_fp16)[name = tensor<string, []>("mul_95_cast_fp16")];
            tensor<fp16, [1, 750, 768]> add_99_cast_fp16 = add(x = add_98_cast_fp16, y = mul_95_cast_fp16)[name = tensor<string, []>("add_99_cast_fp16")];
            tensor<int32, [1]> layer_norm_95_axes_0 = const()[name = tensor<string, []>("layer_norm_95_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [768]> p_encoder_layers_15_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221622464)))];
            tensor<fp16, [768]> p_encoder_layers_15_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("p_encoder_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(221624064)))];
            tensor<fp16, []> layer_norm_95_epsilon_0_to_fp16 = const()[name = tensor<string, []>("layer_norm_95_epsilon_0_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 750, 768]> layer_norm_95_cast_fp16 = layer_norm(axes = layer_norm_95_axes_0, beta = p_encoder_layers_15_norm_out_bias_to_fp16, epsilon = layer_norm_95_epsilon_0_to_fp16, gamma = p_encoder_layers_15_norm_out_weight_to_fp16, x = add_99_cast_fp16)[name = tensor<string, []>("layer_norm_95_cast_fp16")];
            tensor<int32, [3]> transpose_243_perm_0 = const()[name = tensor<string, []>("transpose_243_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 768, 750]> encoded = transpose(perm = transpose_243_perm_0, x = layer_norm_95_cast_fp16)[name = tensor<string, []>("transpose_96")];
        } -> (encoded);
}