program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})]
{
    func main<ios18>(tensor<fp16, [1, 80, 1, 3000]> melspectrogram_features) {
            string var_58_pad_type_0 = const()[name = string("op_58_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> var_58_pad_0 = const()[name = string("op_58_pad_0"), val = tensor<int32, [4]>([0, 0, 1, 1])];
            tensor<int32, [2]> var_58_strides_0 = const()[name = string("op_58_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [2]> var_58_dilations_0 = const()[name = string("op_58_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_58_groups_0 = const()[name = string("op_58_groups_0"), val = int32(1)];
            tensor<fp16, [384, 80, 1, 3]> var_33_to_fp16 = const()[name = string("op_33_to_fp16"), val = tensor<fp16, [384, 80, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            tensor<fp16, [384]> var_39_to_fp16 = const()[name = string("op_39_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(184448)))];
            tensor<fp16, [1, 384, 1, 3000]> var_58_cast_fp16 = conv(bias = var_39_to_fp16, dilations = var_58_dilations_0, groups = var_58_groups_0, pad = var_58_pad_0, pad_type = var_58_pad_type_0, strides = var_58_strides_0, weight = var_33_to_fp16, x = melspectrogram_features)[name = string("op_58_cast_fp16")];
            string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 384, 1, 3000]> hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_58_cast_fp16)[name = string("hidden_states_1_cast_fp16")];
            string var_98_pad_type_0 = const()[name = string("op_98_pad_type_0"), val = string("custom")];
            tensor<int32, [4]> var_98_pad_0 = const()[name = string("op_98_pad_0"), val = tensor<int32, [4]>([0, 0, 1, 1])];
            tensor<int32, [2]> var_98_strides_0 = const()[name = string("op_98_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> var_98_dilations_0 = const()[name = string("op_98_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_98_groups_0 = const()[name = string("op_98_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 3]> var_73_to_fp16 = const()[name = string("op_73_to_fp16"), val = tensor<fp16, [384, 384, 1, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(185280)))];
            tensor<fp16, [384]> var_79_to_fp16 = const()[name = string("op_79_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070080)))];
            tensor<fp16, [1, 384, 1, 1500]> var_98_cast_fp16 = conv(bias = var_79_to_fp16, dilations = var_98_dilations_0, groups = var_98_groups_0, pad = var_98_pad_0, pad_type = var_98_pad_type_0, strides = var_98_strides_0, weight = var_73_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_98_cast_fp16")];
            string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 384, 1, 1500]> hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_98_cast_fp16)[name = string("hidden_states_3_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> var_116_to_fp16 = const()[name = string("op_116_to_fp16"), val = tensor<fp16, [1, 384, 1, 1500]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1070912)))];
            tensor<fp16, [1, 384, 1, 1500]> inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_116_to_fp16)[name = string("inputs_1_cast_fp16")];
            int32 var_126 = const()[name = string("op_126"), val = int32(3)];
            int32 var_137 = const()[name = string("op_137"), val = int32(1)];
            tensor<int32, [1]> out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_154_to_fp16 = const()[name = string("op_154_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_154_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")];
            tensor<fp16, [384]> obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2222976)))];
            tensor<fp16, [384]> obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2223808)))];
            tensor<fp16, [384]> obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2224640)))];
            tensor<fp16, [384]> obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2225472)))];
            fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")];
            string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2226304)))];
            tensor<fp16, [384]> layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2521280)))];
            tensor<fp16, [1, 384, 1, 1500]> query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")];
            string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2522112)))];
            tensor<fp16, [1, 384, 1, 1500]> key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")];
            string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2817088)))];
            tensor<fp16, [384]> layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3112064)))];
            tensor<fp16, [1, 384, 1, 1500]> value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")];
            tensor<int32, [4]> var_192_begin_0 = const()[name = string("op_192_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_192_end_0 = const()[name = string("op_192_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_192_end_mask_0 = const()[name = string("op_192_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_192_cast_fp16 = slice_by_index(begin = var_192_begin_0, end = var_192_end_0, end_mask = var_192_end_mask_0, x = query_1_cast_fp16)[name = string("op_192_cast_fp16")];
            tensor<int32, [4]> var_196_begin_0 = const()[name = string("op_196_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_196_end_0 = const()[name = string("op_196_end_0"), val = tensor<int32, [4]>([1, 128, 1, 1500])];
            tensor<bool, [4]> var_196_end_mask_0 = const()[name = string("op_196_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_196_cast_fp16 = slice_by_index(begin = var_196_begin_0, end = var_196_end_0, end_mask = var_196_end_mask_0, x = query_1_cast_fp16)[name = string("op_196_cast_fp16")];
            tensor<int32, [4]> var_200_begin_0 = const()[name = string("op_200_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_200_end_0 = const()[name = string("op_200_end_0"), val = tensor<int32, [4]>([1, 192, 1, 1500])];
            tensor<bool, [4]> var_200_end_mask_0 = const()[name = string("op_200_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_200_cast_fp16 = slice_by_index(begin = var_200_begin_0, end = var_200_end_0, end_mask = var_200_end_mask_0, x = query_1_cast_fp16)[name = string("op_200_cast_fp16")];
            tensor<int32, [4]> var_204_begin_0 = const()[name = string("op_204_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_204_end_0 = const()[name = string("op_204_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1500])];
            tensor<bool, [4]> var_204_end_mask_0 = const()[name = string("op_204_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_204_cast_fp16 = slice_by_index(begin = var_204_begin_0, end = var_204_end_0, end_mask = var_204_end_mask_0, x = query_1_cast_fp16)[name = string("op_204_cast_fp16")];
            tensor<int32, [4]> var_208_begin_0 = const()[name = string("op_208_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_208_end_0 = const()[name = string("op_208_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1500])];
            tensor<bool, [4]> var_208_end_mask_0 = const()[name = string("op_208_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_208_cast_fp16 = slice_by_index(begin = var_208_begin_0, end = var_208_end_0, end_mask = var_208_end_mask_0, x = query_1_cast_fp16)[name = string("op_208_cast_fp16")];
            tensor<int32, [4]> var_212_begin_0 = const()[name = string("op_212_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_212_end_0 = const()[name = string("op_212_end_0"), val = tensor<int32, [4]>([1, 384, 1, 1500])];
            tensor<bool, [4]> var_212_end_mask_0 = const()[name = string("op_212_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_212_cast_fp16 = slice_by_index(begin = var_212_begin_0, end = var_212_end_0, end_mask = var_212_end_mask_0, x = query_1_cast_fp16)[name = string("op_212_cast_fp16")];
            tensor<int32, [4]> var_221_begin_0 = const()[name = string("op_221_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_221_end_0 = const()[name = string("op_221_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_221_end_mask_0 = const()[name = string("op_221_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_221_cast_fp16 = slice_by_index(begin = var_221_begin_0, end = var_221_end_0, end_mask = var_221_end_mask_0, x = var_192_cast_fp16)[name = string("op_221_cast_fp16")];
            tensor<int32, [4]> var_228_begin_0 = const()[name = string("op_228_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_228_end_0 = const()[name = string("op_228_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_228_end_mask_0 = const()[name = string("op_228_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_228_cast_fp16 = slice_by_index(begin = var_228_begin_0, end = var_228_end_0, end_mask = var_228_end_mask_0, x = var_192_cast_fp16)[name = string("op_228_cast_fp16")];
            tensor<int32, [4]> var_235_begin_0 = const()[name = string("op_235_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_235_end_0 = const()[name = string("op_235_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_235_end_mask_0 = const()[name = string("op_235_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_235_cast_fp16 = slice_by_index(begin = var_235_begin_0, end = var_235_end_0, end_mask = var_235_end_mask_0, x = var_192_cast_fp16)[name = string("op_235_cast_fp16")];
            tensor<int32, [4]> var_242_begin_0 = const()[name = string("op_242_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_242_end_0 = const()[name = string("op_242_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_242_end_mask_0 = const()[name = string("op_242_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_242_cast_fp16 = slice_by_index(begin = var_242_begin_0, end = var_242_end_0, end_mask = var_242_end_mask_0, x = var_192_cast_fp16)[name = string("op_242_cast_fp16")];
            tensor<int32, [4]> var_249_begin_0 = const()[name = string("op_249_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_249_end_0 = const()[name = string("op_249_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_249_end_mask_0 = const()[name = string("op_249_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_249_cast_fp16 = slice_by_index(begin = var_249_begin_0, end = var_249_end_0, end_mask = var_249_end_mask_0, x = var_196_cast_fp16)[name = string("op_249_cast_fp16")];
            tensor<int32, [4]> var_256_begin_0 = const()[name = string("op_256_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_256_end_0 = const()[name = string("op_256_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_256_end_mask_0 = const()[name = string("op_256_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_256_cast_fp16 = slice_by_index(begin = var_256_begin_0, end = var_256_end_0, end_mask = var_256_end_mask_0, x = var_196_cast_fp16)[name = string("op_256_cast_fp16")];
            tensor<int32, [4]> var_263_begin_0 = const()[name = string("op_263_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_263_end_0 = const()[name = string("op_263_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_263_end_mask_0 = const()[name = string("op_263_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_263_cast_fp16 = slice_by_index(begin = var_263_begin_0, end = var_263_end_0, end_mask = var_263_end_mask_0, x = var_196_cast_fp16)[name = string("op_263_cast_fp16")];
            tensor<int32, [4]> var_270_begin_0 = const()[name = string("op_270_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_270_end_0 = const()[name = string("op_270_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_270_end_mask_0 = const()[name = string("op_270_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_270_cast_fp16 = slice_by_index(begin = var_270_begin_0, end = var_270_end_0, end_mask = var_270_end_mask_0, x = var_196_cast_fp16)[name = string("op_270_cast_fp16")];
            tensor<int32, [4]> var_277_begin_0 = const()[name = string("op_277_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_277_end_0 = const()[name = string("op_277_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_277_end_mask_0 = const()[name = string("op_277_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_277_cast_fp16 = slice_by_index(begin = var_277_begin_0, end = var_277_end_0, end_mask = var_277_end_mask_0, x = var_200_cast_fp16)[name = string("op_277_cast_fp16")];
            tensor<int32, [4]> var_284_begin_0 = const()[name = string("op_284_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_284_end_0 = const()[name = string("op_284_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_284_end_mask_0 = const()[name = string("op_284_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_284_cast_fp16 = slice_by_index(begin = var_284_begin_0, end = var_284_end_0, end_mask = var_284_end_mask_0, x = var_200_cast_fp16)[name = string("op_284_cast_fp16")];
            tensor<int32, [4]> var_291_begin_0 = const()[name = string("op_291_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_291_end_0 = const()[name = string("op_291_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_291_end_mask_0 = const()[name = string("op_291_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_291_cast_fp16 = slice_by_index(begin = var_291_begin_0, end = var_291_end_0, end_mask = var_291_end_mask_0, x = var_200_cast_fp16)[name = string("op_291_cast_fp16")];
            tensor<int32, [4]> var_298_begin_0 = const()[name = string("op_298_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_298_end_0 = const()[name = string("op_298_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_298_end_mask_0 = const()[name = string("op_298_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_298_cast_fp16 = slice_by_index(begin = var_298_begin_0, end = var_298_end_0, end_mask = var_298_end_mask_0, x = var_200_cast_fp16)[name = string("op_298_cast_fp16")];
            tensor<int32, [4]> var_305_begin_0 = const()[name = string("op_305_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_305_end_0 = const()[name = string("op_305_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_305_end_mask_0 = const()[name = string("op_305_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_305_cast_fp16 = slice_by_index(begin = var_305_begin_0, end = var_305_end_0, end_mask = var_305_end_mask_0, x = var_204_cast_fp16)[name = string("op_305_cast_fp16")];
            tensor<int32, [4]> var_312_begin_0 = const()[name = string("op_312_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_312_end_0 = const()[name = string("op_312_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_312_end_mask_0 = const()[name = string("op_312_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_312_cast_fp16 = slice_by_index(begin = var_312_begin_0, end = var_312_end_0, end_mask = var_312_end_mask_0, x = var_204_cast_fp16)[name = string("op_312_cast_fp16")];
            tensor<int32, [4]> var_319_begin_0 = const()[name = string("op_319_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_319_end_0 = const()[name = string("op_319_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_319_end_mask_0 = const()[name = string("op_319_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_319_cast_fp16 = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = var_204_cast_fp16)[name = string("op_319_cast_fp16")];
            tensor<int32, [4]> var_326_begin_0 = const()[name = string("op_326_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_326_end_0 = const()[name = string("op_326_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_326_end_mask_0 = const()[name = string("op_326_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_326_cast_fp16 = slice_by_index(begin = var_326_begin_0, end = var_326_end_0, end_mask = var_326_end_mask_0, x = var_204_cast_fp16)[name = string("op_326_cast_fp16")];
            tensor<int32, [4]> var_333_begin_0 = const()[name = string("op_333_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_333_end_0 = const()[name = string("op_333_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_333_end_mask_0 = const()[name = string("op_333_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_333_cast_fp16 = slice_by_index(begin = var_333_begin_0, end = var_333_end_0, end_mask = var_333_end_mask_0, x = var_208_cast_fp16)[name = string("op_333_cast_fp16")];
            tensor<int32, [4]> var_340_begin_0 = const()[name = string("op_340_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_340_end_0 = const()[name = string("op_340_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_340_end_mask_0 = const()[name = string("op_340_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_340_cast_fp16 = slice_by_index(begin = var_340_begin_0, end = var_340_end_0, end_mask = var_340_end_mask_0, x = var_208_cast_fp16)[name = string("op_340_cast_fp16")];
            tensor<int32, [4]> var_347_begin_0 = const()[name = string("op_347_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_347_end_0 = const()[name = string("op_347_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_347_end_mask_0 = const()[name = string("op_347_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_347_cast_fp16 = slice_by_index(begin = var_347_begin_0, end = var_347_end_0, end_mask = var_347_end_mask_0, x = var_208_cast_fp16)[name = string("op_347_cast_fp16")];
            tensor<int32, [4]> var_354_begin_0 = const()[name = string("op_354_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_354_end_0 = const()[name = string("op_354_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_354_end_mask_0 = const()[name = string("op_354_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_354_cast_fp16 = slice_by_index(begin = var_354_begin_0, end = var_354_end_0, end_mask = var_354_end_mask_0, x = var_208_cast_fp16)[name = string("op_354_cast_fp16")];
            tensor<int32, [4]> var_361_begin_0 = const()[name = string("op_361_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_361_end_0 = const()[name = string("op_361_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_361_end_mask_0 = const()[name = string("op_361_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_361_cast_fp16 = slice_by_index(begin = var_361_begin_0, end = var_361_end_0, end_mask = var_361_end_mask_0, x = var_212_cast_fp16)[name = string("op_361_cast_fp16")];
            tensor<int32, [4]> var_368_begin_0 = const()[name = string("op_368_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_368_end_0 = const()[name = string("op_368_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_368_end_mask_0 = const()[name = string("op_368_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_368_cast_fp16 = slice_by_index(begin = var_368_begin_0, end = var_368_end_0, end_mask = var_368_end_mask_0, x = var_212_cast_fp16)[name = string("op_368_cast_fp16")];
            tensor<int32, [4]> var_375_begin_0 = const()[name = string("op_375_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_375_end_0 = const()[name = string("op_375_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_375_end_mask_0 = const()[name = string("op_375_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_375_cast_fp16 = slice_by_index(begin = var_375_begin_0, end = var_375_end_0, end_mask = var_375_end_mask_0, x = var_212_cast_fp16)[name = string("op_375_cast_fp16")];
            tensor<int32, [4]> var_382_begin_0 = const()[name = string("op_382_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_382_end_0 = const()[name = string("op_382_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_382_end_mask_0 = const()[name = string("op_382_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_382_cast_fp16 = slice_by_index(begin = var_382_begin_0, end = var_382_end_0, end_mask = var_382_end_mask_0, x = var_212_cast_fp16)[name = string("op_382_cast_fp16")];
            tensor<int32, [4]> k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_387_begin_0 = const()[name = string("op_387_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_387_end_0 = const()[name = string("op_387_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 64])];
            tensor<bool, [4]> var_387_end_mask_0 = const()[name = string("op_387_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 384]> k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [1, 1500, 1, 64]> var_387_cast_fp16 = slice_by_index(begin = var_387_begin_0, end = var_387_end_0, end_mask = var_387_end_mask_0, x = k_1_cast_fp16)[name = string("op_387_cast_fp16")];
            tensor<int32, [4]> var_391_begin_0 = const()[name = string("op_391_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_391_end_0 = const()[name = string("op_391_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 128])];
            tensor<bool, [4]> var_391_end_mask_0 = const()[name = string("op_391_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_391_cast_fp16 = slice_by_index(begin = var_391_begin_0, end = var_391_end_0, end_mask = var_391_end_mask_0, x = k_1_cast_fp16)[name = string("op_391_cast_fp16")];
            tensor<int32, [4]> var_395_begin_0 = const()[name = string("op_395_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_395_end_0 = const()[name = string("op_395_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 192])];
            tensor<bool, [4]> var_395_end_mask_0 = const()[name = string("op_395_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_395_cast_fp16 = slice_by_index(begin = var_395_begin_0, end = var_395_end_0, end_mask = var_395_end_mask_0, x = k_1_cast_fp16)[name = string("op_395_cast_fp16")];
            tensor<int32, [4]> var_399_begin_0 = const()[name = string("op_399_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_399_end_0 = const()[name = string("op_399_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 256])];
            tensor<bool, [4]> var_399_end_mask_0 = const()[name = string("op_399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_399_cast_fp16 = slice_by_index(begin = var_399_begin_0, end = var_399_end_0, end_mask = var_399_end_mask_0, x = k_1_cast_fp16)[name = string("op_399_cast_fp16")];
            tensor<int32, [4]> var_403_begin_0 = const()[name = string("op_403_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_403_end_0 = const()[name = string("op_403_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 320])];
            tensor<bool, [4]> var_403_end_mask_0 = const()[name = string("op_403_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_403_cast_fp16 = slice_by_index(begin = var_403_begin_0, end = var_403_end_0, end_mask = var_403_end_mask_0, x = k_1_cast_fp16)[name = string("op_403_cast_fp16")];
            tensor<int32, [4]> var_407_begin_0 = const()[name = string("op_407_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_407_end_0 = const()[name = string("op_407_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 384])];
            tensor<bool, [4]> var_407_end_mask_0 = const()[name = string("op_407_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_407_cast_fp16 = slice_by_index(begin = var_407_begin_0, end = var_407_end_0, end_mask = var_407_end_mask_0, x = k_1_cast_fp16)[name = string("op_407_cast_fp16")];
            tensor<int32, [4]> var_409_begin_0 = const()[name = string("op_409_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_409_end_0 = const()[name = string("op_409_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_409_end_mask_0 = const()[name = string("op_409_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_409_cast_fp16 = slice_by_index(begin = var_409_begin_0, end = var_409_end_0, end_mask = var_409_end_mask_0, x = value_1_cast_fp16)[name = string("op_409_cast_fp16")];
            tensor<int32, [4]> var_413_begin_0 = const()[name = string("op_413_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_413_end_0 = const()[name = string("op_413_end_0"), val = tensor<int32, [4]>([1, 128, 1, 1500])];
            tensor<bool, [4]> var_413_end_mask_0 = const()[name = string("op_413_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_413_cast_fp16 = slice_by_index(begin = var_413_begin_0, end = var_413_end_0, end_mask = var_413_end_mask_0, x = value_1_cast_fp16)[name = string("op_413_cast_fp16")];
            tensor<int32, [4]> var_417_begin_0 = const()[name = string("op_417_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_417_end_0 = const()[name = string("op_417_end_0"), val = tensor<int32, [4]>([1, 192, 1, 1500])];
            tensor<bool, [4]> var_417_end_mask_0 = const()[name = string("op_417_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_417_cast_fp16 = slice_by_index(begin = var_417_begin_0, end = var_417_end_0, end_mask = var_417_end_mask_0, x = value_1_cast_fp16)[name = string("op_417_cast_fp16")];
            tensor<int32, [4]> var_421_begin_0 = const()[name = string("op_421_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_421_end_0 = const()[name = string("op_421_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1500])];
            tensor<bool, [4]> var_421_end_mask_0 = const()[name = string("op_421_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_421_cast_fp16 = slice_by_index(begin = var_421_begin_0, end = var_421_end_0, end_mask = var_421_end_mask_0, x = value_1_cast_fp16)[name = string("op_421_cast_fp16")];
            tensor<int32, [4]> var_425_begin_0 = const()[name = string("op_425_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_425_end_0 = const()[name = string("op_425_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1500])];
            tensor<bool, [4]> var_425_end_mask_0 = const()[name = string("op_425_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_425_cast_fp16 = slice_by_index(begin = var_425_begin_0, end = var_425_end_0, end_mask = var_425_end_mask_0, x = value_1_cast_fp16)[name = string("op_425_cast_fp16")];
            tensor<int32, [4]> var_429_begin_0 = const()[name = string("op_429_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_429_end_0 = const()[name = string("op_429_end_0"), val = tensor<int32, [4]>([1, 384, 1, 1500])];
            tensor<bool, [4]> var_429_end_mask_0 = const()[name = string("op_429_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_429_cast_fp16 = slice_by_index(begin = var_429_begin_0, end = var_429_end_0, end_mask = var_429_end_mask_0, x = value_1_cast_fp16)[name = string("op_429_cast_fp16")];
            string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_387_cast_fp16, var_221_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")];
            string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_387_cast_fp16, var_228_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")];
            string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_387_cast_fp16, var_235_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")];
            string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_387_cast_fp16, var_242_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")];
            string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_391_cast_fp16, var_249_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")];
            string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_391_cast_fp16, var_256_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")];
            string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_391_cast_fp16, var_263_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")];
            string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_391_cast_fp16, var_270_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")];
            string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_395_cast_fp16, var_277_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")];
            string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_395_cast_fp16, var_284_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")];
            string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_395_cast_fp16, var_291_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")];
            string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_395_cast_fp16, var_298_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")];
            string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_399_cast_fp16, var_305_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")];
            string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_399_cast_fp16, var_312_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")];
            string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_399_cast_fp16, var_319_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")];
            string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_399_cast_fp16, var_326_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")];
            string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_403_cast_fp16, var_333_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")];
            string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_403_cast_fp16, var_340_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")];
            string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_403_cast_fp16, var_347_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")];
            string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_403_cast_fp16, var_354_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")];
            string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_407_cast_fp16, var_361_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")];
            string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_407_cast_fp16, var_368_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")];
            string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_407_cast_fp16, var_375_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")];
            string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_407_cast_fp16, var_382_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")];
            fp16 var_480_to_fp16 = const()[name = string("op_480_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_480_to_fp16)[name = string("aw_chunk_1_cast_fp16")];
            fp16 var_482_to_fp16 = const()[name = string("op_482_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_482_to_fp16)[name = string("aw_chunk_3_cast_fp16")];
            fp16 var_484_to_fp16 = const()[name = string("op_484_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_484_to_fp16)[name = string("aw_chunk_5_cast_fp16")];
            fp16 var_486_to_fp16 = const()[name = string("op_486_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_486_to_fp16)[name = string("aw_chunk_7_cast_fp16")];
            fp16 var_488_to_fp16 = const()[name = string("op_488_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_488_to_fp16)[name = string("aw_chunk_9_cast_fp16")];
            fp16 var_490_to_fp16 = const()[name = string("op_490_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_490_to_fp16)[name = string("aw_chunk_11_cast_fp16")];
            fp16 var_492_to_fp16 = const()[name = string("op_492_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_492_to_fp16)[name = string("aw_chunk_13_cast_fp16")];
            fp16 var_494_to_fp16 = const()[name = string("op_494_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_494_to_fp16)[name = string("aw_chunk_15_cast_fp16")];
            fp16 var_496_to_fp16 = const()[name = string("op_496_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_496_to_fp16)[name = string("aw_chunk_17_cast_fp16")];
            fp16 var_498_to_fp16 = const()[name = string("op_498_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_498_to_fp16)[name = string("aw_chunk_19_cast_fp16")];
            fp16 var_500_to_fp16 = const()[name = string("op_500_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_500_to_fp16)[name = string("aw_chunk_21_cast_fp16")];
            fp16 var_502_to_fp16 = const()[name = string("op_502_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_502_to_fp16)[name = string("aw_chunk_23_cast_fp16")];
            fp16 var_504_to_fp16 = const()[name = string("op_504_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_504_to_fp16)[name = string("aw_chunk_25_cast_fp16")];
            fp16 var_506_to_fp16 = const()[name = string("op_506_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_506_to_fp16)[name = string("aw_chunk_27_cast_fp16")];
            fp16 var_508_to_fp16 = const()[name = string("op_508_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_508_to_fp16)[name = string("aw_chunk_29_cast_fp16")];
            fp16 var_510_to_fp16 = const()[name = string("op_510_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_510_to_fp16)[name = string("aw_chunk_31_cast_fp16")];
            fp16 var_512_to_fp16 = const()[name = string("op_512_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_512_to_fp16)[name = string("aw_chunk_33_cast_fp16")];
            fp16 var_514_to_fp16 = const()[name = string("op_514_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_514_to_fp16)[name = string("aw_chunk_35_cast_fp16")];
            fp16 var_516_to_fp16 = const()[name = string("op_516_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_516_to_fp16)[name = string("aw_chunk_37_cast_fp16")];
            fp16 var_518_to_fp16 = const()[name = string("op_518_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_518_to_fp16)[name = string("aw_chunk_39_cast_fp16")];
            fp16 var_520_to_fp16 = const()[name = string("op_520_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_520_to_fp16)[name = string("aw_chunk_41_cast_fp16")];
            fp16 var_522_to_fp16 = const()[name = string("op_522_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_522_to_fp16)[name = string("aw_chunk_43_cast_fp16")];
            fp16 var_524_to_fp16 = const()[name = string("op_524_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_524_to_fp16)[name = string("aw_chunk_45_cast_fp16")];
            fp16 var_526_to_fp16 = const()[name = string("op_526_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_526_to_fp16)[name = string("aw_chunk_47_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_528_cast_fp16 = softmax(axis = var_137, x = aw_chunk_1_cast_fp16)[name = string("op_528_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_529_cast_fp16 = softmax(axis = var_137, x = aw_chunk_3_cast_fp16)[name = string("op_529_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_530_cast_fp16 = softmax(axis = var_137, x = aw_chunk_5_cast_fp16)[name = string("op_530_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_531_cast_fp16 = softmax(axis = var_137, x = aw_chunk_7_cast_fp16)[name = string("op_531_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_532_cast_fp16 = softmax(axis = var_137, x = aw_chunk_9_cast_fp16)[name = string("op_532_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_533_cast_fp16 = softmax(axis = var_137, x = aw_chunk_11_cast_fp16)[name = string("op_533_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_534_cast_fp16 = softmax(axis = var_137, x = aw_chunk_13_cast_fp16)[name = string("op_534_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_535_cast_fp16 = softmax(axis = var_137, x = aw_chunk_15_cast_fp16)[name = string("op_535_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_536_cast_fp16 = softmax(axis = var_137, x = aw_chunk_17_cast_fp16)[name = string("op_536_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_537_cast_fp16 = softmax(axis = var_137, x = aw_chunk_19_cast_fp16)[name = string("op_537_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_538_cast_fp16 = softmax(axis = var_137, x = aw_chunk_21_cast_fp16)[name = string("op_538_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_539_cast_fp16 = softmax(axis = var_137, x = aw_chunk_23_cast_fp16)[name = string("op_539_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_540_cast_fp16 = softmax(axis = var_137, x = aw_chunk_25_cast_fp16)[name = string("op_540_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_541_cast_fp16 = softmax(axis = var_137, x = aw_chunk_27_cast_fp16)[name = string("op_541_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_542_cast_fp16 = softmax(axis = var_137, x = aw_chunk_29_cast_fp16)[name = string("op_542_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_543_cast_fp16 = softmax(axis = var_137, x = aw_chunk_31_cast_fp16)[name = string("op_543_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_544_cast_fp16 = softmax(axis = var_137, x = aw_chunk_33_cast_fp16)[name = string("op_544_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_545_cast_fp16 = softmax(axis = var_137, x = aw_chunk_35_cast_fp16)[name = string("op_545_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_546_cast_fp16 = softmax(axis = var_137, x = aw_chunk_37_cast_fp16)[name = string("op_546_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_547_cast_fp16 = softmax(axis = var_137, x = aw_chunk_39_cast_fp16)[name = string("op_547_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_548_cast_fp16 = softmax(axis = var_137, x = aw_chunk_41_cast_fp16)[name = string("op_548_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_549_cast_fp16 = softmax(axis = var_137, x = aw_chunk_43_cast_fp16)[name = string("op_549_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_550_cast_fp16 = softmax(axis = var_137, x = aw_chunk_45_cast_fp16)[name = string("op_550_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_551_cast_fp16 = softmax(axis = var_137, x = aw_chunk_47_cast_fp16)[name = string("op_551_cast_fp16")];
            string var_553_equation_0 = const()[name = string("op_553_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_553_cast_fp16 = einsum(equation = var_553_equation_0, values = (var_409_cast_fp16, var_528_cast_fp16))[name = string("op_553_cast_fp16")];
            string var_555_equation_0 = const()[name = string("op_555_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_555_cast_fp16 = einsum(equation = var_555_equation_0, values = (var_409_cast_fp16, var_529_cast_fp16))[name = string("op_555_cast_fp16")];
            string var_557_equation_0 = const()[name = string("op_557_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_557_cast_fp16 = einsum(equation = var_557_equation_0, values = (var_409_cast_fp16, var_530_cast_fp16))[name = string("op_557_cast_fp16")];
            string var_559_equation_0 = const()[name = string("op_559_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_559_cast_fp16 = einsum(equation = var_559_equation_0, values = (var_409_cast_fp16, var_531_cast_fp16))[name = string("op_559_cast_fp16")];
            string var_561_equation_0 = const()[name = string("op_561_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_561_cast_fp16 = einsum(equation = var_561_equation_0, values = (var_413_cast_fp16, var_532_cast_fp16))[name = string("op_561_cast_fp16")];
            string var_563_equation_0 = const()[name = string("op_563_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_563_cast_fp16 = einsum(equation = var_563_equation_0, values = (var_413_cast_fp16, var_533_cast_fp16))[name = string("op_563_cast_fp16")];
            string var_565_equation_0 = const()[name = string("op_565_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_565_cast_fp16 = einsum(equation = var_565_equation_0, values = (var_413_cast_fp16, var_534_cast_fp16))[name = string("op_565_cast_fp16")];
            string var_567_equation_0 = const()[name = string("op_567_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_567_cast_fp16 = einsum(equation = var_567_equation_0, values = (var_413_cast_fp16, var_535_cast_fp16))[name = string("op_567_cast_fp16")];
            string var_569_equation_0 = const()[name = string("op_569_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_569_cast_fp16 = einsum(equation = var_569_equation_0, values = (var_417_cast_fp16, var_536_cast_fp16))[name = string("op_569_cast_fp16")];
            string var_571_equation_0 = const()[name = string("op_571_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_571_cast_fp16 = einsum(equation = var_571_equation_0, values = (var_417_cast_fp16, var_537_cast_fp16))[name = string("op_571_cast_fp16")];
            string var_573_equation_0 = const()[name = string("op_573_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_573_cast_fp16 = einsum(equation = var_573_equation_0, values = (var_417_cast_fp16, var_538_cast_fp16))[name = string("op_573_cast_fp16")];
            string var_575_equation_0 = const()[name = string("op_575_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_575_cast_fp16 = einsum(equation = var_575_equation_0, values = (var_417_cast_fp16, var_539_cast_fp16))[name = string("op_575_cast_fp16")];
            string var_577_equation_0 = const()[name = string("op_577_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_577_cast_fp16 = einsum(equation = var_577_equation_0, values = (var_421_cast_fp16, var_540_cast_fp16))[name = string("op_577_cast_fp16")];
            string var_579_equation_0 = const()[name = string("op_579_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_579_cast_fp16 = einsum(equation = var_579_equation_0, values = (var_421_cast_fp16, var_541_cast_fp16))[name = string("op_579_cast_fp16")];
            string var_581_equation_0 = const()[name = string("op_581_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_581_cast_fp16 = einsum(equation = var_581_equation_0, values = (var_421_cast_fp16, var_542_cast_fp16))[name = string("op_581_cast_fp16")];
            string var_583_equation_0 = const()[name = string("op_583_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_583_cast_fp16 = einsum(equation = var_583_equation_0, values = (var_421_cast_fp16, var_543_cast_fp16))[name = string("op_583_cast_fp16")];
            string var_585_equation_0 = const()[name = string("op_585_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_585_cast_fp16 = einsum(equation = var_585_equation_0, values = (var_425_cast_fp16, var_544_cast_fp16))[name = string("op_585_cast_fp16")];
            string var_587_equation_0 = const()[name = string("op_587_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_587_cast_fp16 = einsum(equation = var_587_equation_0, values = (var_425_cast_fp16, var_545_cast_fp16))[name = string("op_587_cast_fp16")];
            string var_589_equation_0 = const()[name = string("op_589_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_589_cast_fp16 = einsum(equation = var_589_equation_0, values = (var_425_cast_fp16, var_546_cast_fp16))[name = string("op_589_cast_fp16")];
            string var_591_equation_0 = const()[name = string("op_591_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_591_cast_fp16 = einsum(equation = var_591_equation_0, values = (var_425_cast_fp16, var_547_cast_fp16))[name = string("op_591_cast_fp16")];
            string var_593_equation_0 = const()[name = string("op_593_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_593_cast_fp16 = einsum(equation = var_593_equation_0, values = (var_429_cast_fp16, var_548_cast_fp16))[name = string("op_593_cast_fp16")];
            string var_595_equation_0 = const()[name = string("op_595_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_595_cast_fp16 = einsum(equation = var_595_equation_0, values = (var_429_cast_fp16, var_549_cast_fp16))[name = string("op_595_cast_fp16")];
            string var_597_equation_0 = const()[name = string("op_597_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_597_cast_fp16 = einsum(equation = var_597_equation_0, values = (var_429_cast_fp16, var_550_cast_fp16))[name = string("op_597_cast_fp16")];
            string var_599_equation_0 = const()[name = string("op_599_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_599_cast_fp16 = einsum(equation = var_599_equation_0, values = (var_429_cast_fp16, var_551_cast_fp16))[name = string("op_599_cast_fp16")];
            bool var_601_interleave_0 = const()[name = string("op_601_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_601_cast_fp16 = concat(axis = var_126, interleave = var_601_interleave_0, values = (var_553_cast_fp16, var_555_cast_fp16, var_557_cast_fp16, var_559_cast_fp16))[name = string("op_601_cast_fp16")];
            bool var_603_interleave_0 = const()[name = string("op_603_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_603_cast_fp16 = concat(axis = var_126, interleave = var_603_interleave_0, values = (var_561_cast_fp16, var_563_cast_fp16, var_565_cast_fp16, var_567_cast_fp16))[name = string("op_603_cast_fp16")];
            bool var_605_interleave_0 = const()[name = string("op_605_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_605_cast_fp16 = concat(axis = var_126, interleave = var_605_interleave_0, values = (var_569_cast_fp16, var_571_cast_fp16, var_573_cast_fp16, var_575_cast_fp16))[name = string("op_605_cast_fp16")];
            bool var_607_interleave_0 = const()[name = string("op_607_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_607_cast_fp16 = concat(axis = var_126, interleave = var_607_interleave_0, values = (var_577_cast_fp16, var_579_cast_fp16, var_581_cast_fp16, var_583_cast_fp16))[name = string("op_607_cast_fp16")];
            bool var_609_interleave_0 = const()[name = string("op_609_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_609_cast_fp16 = concat(axis = var_126, interleave = var_609_interleave_0, values = (var_585_cast_fp16, var_587_cast_fp16, var_589_cast_fp16, var_591_cast_fp16))[name = string("op_609_cast_fp16")];
            bool var_611_interleave_0 = const()[name = string("op_611_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_611_cast_fp16 = concat(axis = var_126, interleave = var_611_interleave_0, values = (var_593_cast_fp16, var_595_cast_fp16, var_597_cast_fp16, var_599_cast_fp16))[name = string("op_611_cast_fp16")];
            bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 384, 1, 1500]> input_1_cast_fp16 = concat(axis = var_137, interleave = input_1_interleave_0, values = (var_601_cast_fp16, var_603_cast_fp16, var_605_cast_fp16, var_607_cast_fp16, var_609_cast_fp16, var_611_cast_fp16))[name = string("input_1_cast_fp16")];
            string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3112896)))];
            tensor<fp16, [384]> layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3407872)))];
            tensor<fp16, [1, 384, 1, 1500]> obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")];
            tensor<int32, [1]> out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_630_to_fp16 = const()[name = string("op_630_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_630_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")];
            tensor<fp16, [384]> input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3408704)))];
            tensor<fp16, [384]> input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3409536)))];
            fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")];
            string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)];
            tensor<fp16, [1536, 384, 1, 1]> layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [1536, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3410368)))];
            tensor<fp16, [1536]> layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4590080)))];
            tensor<fp16, [1, 1536, 1, 1500]> input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")];
            string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1536, 1, 1500]> input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")];
            string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)];
            tensor<fp16, [384, 1536, 1, 1]> layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [384, 1536, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4593216)))];
            tensor<fp16, [384]> layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5772928)))];
            tensor<fp16, [1, 384, 1, 1500]> hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")];
            int32 var_659 = const()[name = string("op_659"), val = int32(3)];
            int32 var_670 = const()[name = string("op_670"), val = int32(1)];
            tensor<int32, [1]> out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_687_to_fp16 = const()[name = string("op_687_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_687_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")];
            tensor<fp16, [384]> obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5773760)))];
            tensor<fp16, [384]> obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5774592)))];
            fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")];
            string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5775424)))];
            tensor<fp16, [384]> layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6070400)))];
            tensor<fp16, [1, 384, 1, 1500]> query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")];
            string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6071232)))];
            tensor<fp16, [1, 384, 1, 1500]> key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")];
            string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6366208)))];
            tensor<fp16, [384]> layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6661184)))];
            tensor<fp16, [1, 384, 1, 1500]> value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")];
            tensor<int32, [4]> var_725_begin_0 = const()[name = string("op_725_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_725_end_0 = const()[name = string("op_725_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_725_end_mask_0 = const()[name = string("op_725_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_725_cast_fp16 = slice_by_index(begin = var_725_begin_0, end = var_725_end_0, end_mask = var_725_end_mask_0, x = query_3_cast_fp16)[name = string("op_725_cast_fp16")];
            tensor<int32, [4]> var_729_begin_0 = const()[name = string("op_729_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_729_end_0 = const()[name = string("op_729_end_0"), val = tensor<int32, [4]>([1, 128, 1, 1500])];
            tensor<bool, [4]> var_729_end_mask_0 = const()[name = string("op_729_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_729_cast_fp16 = slice_by_index(begin = var_729_begin_0, end = var_729_end_0, end_mask = var_729_end_mask_0, x = query_3_cast_fp16)[name = string("op_729_cast_fp16")];
            tensor<int32, [4]> var_733_begin_0 = const()[name = string("op_733_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_733_end_0 = const()[name = string("op_733_end_0"), val = tensor<int32, [4]>([1, 192, 1, 1500])];
            tensor<bool, [4]> var_733_end_mask_0 = const()[name = string("op_733_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_733_cast_fp16 = slice_by_index(begin = var_733_begin_0, end = var_733_end_0, end_mask = var_733_end_mask_0, x = query_3_cast_fp16)[name = string("op_733_cast_fp16")];
            tensor<int32, [4]> var_737_begin_0 = const()[name = string("op_737_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_737_end_0 = const()[name = string("op_737_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1500])];
            tensor<bool, [4]> var_737_end_mask_0 = const()[name = string("op_737_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_737_cast_fp16 = slice_by_index(begin = var_737_begin_0, end = var_737_end_0, end_mask = var_737_end_mask_0, x = query_3_cast_fp16)[name = string("op_737_cast_fp16")];
            tensor<int32, [4]> var_741_begin_0 = const()[name = string("op_741_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_741_end_0 = const()[name = string("op_741_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1500])];
            tensor<bool, [4]> var_741_end_mask_0 = const()[name = string("op_741_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_741_cast_fp16 = slice_by_index(begin = var_741_begin_0, end = var_741_end_0, end_mask = var_741_end_mask_0, x = query_3_cast_fp16)[name = string("op_741_cast_fp16")];
            tensor<int32, [4]> var_745_begin_0 = const()[name = string("op_745_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_745_end_0 = const()[name = string("op_745_end_0"), val = tensor<int32, [4]>([1, 384, 1, 1500])];
            tensor<bool, [4]> var_745_end_mask_0 = const()[name = string("op_745_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_745_cast_fp16 = slice_by_index(begin = var_745_begin_0, end = var_745_end_0, end_mask = var_745_end_mask_0, x = query_3_cast_fp16)[name = string("op_745_cast_fp16")];
            tensor<int32, [4]> var_754_begin_0 = const()[name = string("op_754_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_754_end_0 = const()[name = string("op_754_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_754_end_mask_0 = const()[name = string("op_754_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_754_cast_fp16 = slice_by_index(begin = var_754_begin_0, end = var_754_end_0, end_mask = var_754_end_mask_0, x = var_725_cast_fp16)[name = string("op_754_cast_fp16")];
            tensor<int32, [4]> var_761_begin_0 = const()[name = string("op_761_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_761_end_0 = const()[name = string("op_761_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_761_end_mask_0 = const()[name = string("op_761_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_761_cast_fp16 = slice_by_index(begin = var_761_begin_0, end = var_761_end_0, end_mask = var_761_end_mask_0, x = var_725_cast_fp16)[name = string("op_761_cast_fp16")];
            tensor<int32, [4]> var_768_begin_0 = const()[name = string("op_768_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_768_end_0 = const()[name = string("op_768_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_768_end_mask_0 = const()[name = string("op_768_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_768_cast_fp16 = slice_by_index(begin = var_768_begin_0, end = var_768_end_0, end_mask = var_768_end_mask_0, x = var_725_cast_fp16)[name = string("op_768_cast_fp16")];
            tensor<int32, [4]> var_775_begin_0 = const()[name = string("op_775_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_775_end_0 = const()[name = string("op_775_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_775_end_mask_0 = const()[name = string("op_775_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_775_cast_fp16 = slice_by_index(begin = var_775_begin_0, end = var_775_end_0, end_mask = var_775_end_mask_0, x = var_725_cast_fp16)[name = string("op_775_cast_fp16")];
            tensor<int32, [4]> var_782_begin_0 = const()[name = string("op_782_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_782_end_0 = const()[name = string("op_782_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_782_end_mask_0 = const()[name = string("op_782_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_782_cast_fp16 = slice_by_index(begin = var_782_begin_0, end = var_782_end_0, end_mask = var_782_end_mask_0, x = var_729_cast_fp16)[name = string("op_782_cast_fp16")];
            tensor<int32, [4]> var_789_begin_0 = const()[name = string("op_789_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_789_end_0 = const()[name = string("op_789_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_789_end_mask_0 = const()[name = string("op_789_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_789_cast_fp16 = slice_by_index(begin = var_789_begin_0, end = var_789_end_0, end_mask = var_789_end_mask_0, x = var_729_cast_fp16)[name = string("op_789_cast_fp16")];
            tensor<int32, [4]> var_796_begin_0 = const()[name = string("op_796_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_796_end_0 = const()[name = string("op_796_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_796_end_mask_0 = const()[name = string("op_796_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_796_cast_fp16 = slice_by_index(begin = var_796_begin_0, end = var_796_end_0, end_mask = var_796_end_mask_0, x = var_729_cast_fp16)[name = string("op_796_cast_fp16")];
            tensor<int32, [4]> var_803_begin_0 = const()[name = string("op_803_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_803_end_0 = const()[name = string("op_803_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_803_end_mask_0 = const()[name = string("op_803_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_803_cast_fp16 = slice_by_index(begin = var_803_begin_0, end = var_803_end_0, end_mask = var_803_end_mask_0, x = var_729_cast_fp16)[name = string("op_803_cast_fp16")];
            tensor<int32, [4]> var_810_begin_0 = const()[name = string("op_810_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_810_end_0 = const()[name = string("op_810_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_810_end_mask_0 = const()[name = string("op_810_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_810_cast_fp16 = slice_by_index(begin = var_810_begin_0, end = var_810_end_0, end_mask = var_810_end_mask_0, x = var_733_cast_fp16)[name = string("op_810_cast_fp16")];
            tensor<int32, [4]> var_817_begin_0 = const()[name = string("op_817_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_817_end_0 = const()[name = string("op_817_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_817_end_mask_0 = const()[name = string("op_817_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_817_cast_fp16 = slice_by_index(begin = var_817_begin_0, end = var_817_end_0, end_mask = var_817_end_mask_0, x = var_733_cast_fp16)[name = string("op_817_cast_fp16")];
            tensor<int32, [4]> var_824_begin_0 = const()[name = string("op_824_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_824_end_0 = const()[name = string("op_824_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_824_end_mask_0 = const()[name = string("op_824_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_824_cast_fp16 = slice_by_index(begin = var_824_begin_0, end = var_824_end_0, end_mask = var_824_end_mask_0, x = var_733_cast_fp16)[name = string("op_824_cast_fp16")];
            tensor<int32, [4]> var_831_begin_0 = const()[name = string("op_831_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_831_end_0 = const()[name = string("op_831_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_831_end_mask_0 = const()[name = string("op_831_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_831_cast_fp16 = slice_by_index(begin = var_831_begin_0, end = var_831_end_0, end_mask = var_831_end_mask_0, x = var_733_cast_fp16)[name = string("op_831_cast_fp16")];
            tensor<int32, [4]> var_838_begin_0 = const()[name = string("op_838_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_838_end_0 = const()[name = string("op_838_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_838_end_mask_0 = const()[name = string("op_838_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_838_cast_fp16 = slice_by_index(begin = var_838_begin_0, end = var_838_end_0, end_mask = var_838_end_mask_0, x = var_737_cast_fp16)[name = string("op_838_cast_fp16")];
            tensor<int32, [4]> var_845_begin_0 = const()[name = string("op_845_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_845_end_0 = const()[name = string("op_845_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_845_end_mask_0 = const()[name = string("op_845_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_845_cast_fp16 = slice_by_index(begin = var_845_begin_0, end = var_845_end_0, end_mask = var_845_end_mask_0, x = var_737_cast_fp16)[name = string("op_845_cast_fp16")];
            tensor<int32, [4]> var_852_begin_0 = const()[name = string("op_852_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_852_end_0 = const()[name = string("op_852_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_852_end_mask_0 = const()[name = string("op_852_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_852_cast_fp16 = slice_by_index(begin = var_852_begin_0, end = var_852_end_0, end_mask = var_852_end_mask_0, x = var_737_cast_fp16)[name = string("op_852_cast_fp16")];
            tensor<int32, [4]> var_859_begin_0 = const()[name = string("op_859_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_859_end_0 = const()[name = string("op_859_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_859_end_mask_0 = const()[name = string("op_859_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_859_cast_fp16 = slice_by_index(begin = var_859_begin_0, end = var_859_end_0, end_mask = var_859_end_mask_0, x = var_737_cast_fp16)[name = string("op_859_cast_fp16")];
            tensor<int32, [4]> var_866_begin_0 = const()[name = string("op_866_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_866_end_0 = const()[name = string("op_866_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_866_end_mask_0 = const()[name = string("op_866_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_866_cast_fp16 = slice_by_index(begin = var_866_begin_0, end = var_866_end_0, end_mask = var_866_end_mask_0, x = var_741_cast_fp16)[name = string("op_866_cast_fp16")];
            tensor<int32, [4]> var_873_begin_0 = const()[name = string("op_873_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_873_end_0 = const()[name = string("op_873_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_873_end_mask_0 = const()[name = string("op_873_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_873_cast_fp16 = slice_by_index(begin = var_873_begin_0, end = var_873_end_0, end_mask = var_873_end_mask_0, x = var_741_cast_fp16)[name = string("op_873_cast_fp16")];
            tensor<int32, [4]> var_880_begin_0 = const()[name = string("op_880_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_880_end_0 = const()[name = string("op_880_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_880_end_mask_0 = const()[name = string("op_880_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_880_cast_fp16 = slice_by_index(begin = var_880_begin_0, end = var_880_end_0, end_mask = var_880_end_mask_0, x = var_741_cast_fp16)[name = string("op_880_cast_fp16")];
            tensor<int32, [4]> var_887_begin_0 = const()[name = string("op_887_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_887_end_0 = const()[name = string("op_887_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_887_end_mask_0 = const()[name = string("op_887_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_887_cast_fp16 = slice_by_index(begin = var_887_begin_0, end = var_887_end_0, end_mask = var_887_end_mask_0, x = var_741_cast_fp16)[name = string("op_887_cast_fp16")];
            tensor<int32, [4]> var_894_begin_0 = const()[name = string("op_894_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_894_end_0 = const()[name = string("op_894_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_894_end_mask_0 = const()[name = string("op_894_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_894_cast_fp16 = slice_by_index(begin = var_894_begin_0, end = var_894_end_0, end_mask = var_894_end_mask_0, x = var_745_cast_fp16)[name = string("op_894_cast_fp16")];
            tensor<int32, [4]> var_901_begin_0 = const()[name = string("op_901_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_901_end_0 = const()[name = string("op_901_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_901_end_mask_0 = const()[name = string("op_901_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_901_cast_fp16 = slice_by_index(begin = var_901_begin_0, end = var_901_end_0, end_mask = var_901_end_mask_0, x = var_745_cast_fp16)[name = string("op_901_cast_fp16")];
            tensor<int32, [4]> var_908_begin_0 = const()[name = string("op_908_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_908_end_0 = const()[name = string("op_908_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_908_end_mask_0 = const()[name = string("op_908_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_908_cast_fp16 = slice_by_index(begin = var_908_begin_0, end = var_908_end_0, end_mask = var_908_end_mask_0, x = var_745_cast_fp16)[name = string("op_908_cast_fp16")];
            tensor<int32, [4]> var_915_begin_0 = const()[name = string("op_915_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_915_end_0 = const()[name = string("op_915_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_915_end_mask_0 = const()[name = string("op_915_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_915_cast_fp16 = slice_by_index(begin = var_915_begin_0, end = var_915_end_0, end_mask = var_915_end_mask_0, x = var_745_cast_fp16)[name = string("op_915_cast_fp16")];
            tensor<int32, [4]> k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_920_begin_0 = const()[name = string("op_920_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_920_end_0 = const()[name = string("op_920_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 64])];
            tensor<bool, [4]> var_920_end_mask_0 = const()[name = string("op_920_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 384]> k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 1500, 1, 64]> var_920_cast_fp16 = slice_by_index(begin = var_920_begin_0, end = var_920_end_0, end_mask = var_920_end_mask_0, x = k_3_cast_fp16)[name = string("op_920_cast_fp16")];
            tensor<int32, [4]> var_924_begin_0 = const()[name = string("op_924_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_924_end_0 = const()[name = string("op_924_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 128])];
            tensor<bool, [4]> var_924_end_mask_0 = const()[name = string("op_924_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_924_cast_fp16 = slice_by_index(begin = var_924_begin_0, end = var_924_end_0, end_mask = var_924_end_mask_0, x = k_3_cast_fp16)[name = string("op_924_cast_fp16")];
            tensor<int32, [4]> var_928_begin_0 = const()[name = string("op_928_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_928_end_0 = const()[name = string("op_928_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 192])];
            tensor<bool, [4]> var_928_end_mask_0 = const()[name = string("op_928_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_928_cast_fp16 = slice_by_index(begin = var_928_begin_0, end = var_928_end_0, end_mask = var_928_end_mask_0, x = k_3_cast_fp16)[name = string("op_928_cast_fp16")];
            tensor<int32, [4]> var_932_begin_0 = const()[name = string("op_932_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_932_end_0 = const()[name = string("op_932_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 256])];
            tensor<bool, [4]> var_932_end_mask_0 = const()[name = string("op_932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_932_cast_fp16 = slice_by_index(begin = var_932_begin_0, end = var_932_end_0, end_mask = var_932_end_mask_0, x = k_3_cast_fp16)[name = string("op_932_cast_fp16")];
            tensor<int32, [4]> var_936_begin_0 = const()[name = string("op_936_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_936_end_0 = const()[name = string("op_936_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 320])];
            tensor<bool, [4]> var_936_end_mask_0 = const()[name = string("op_936_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_936_cast_fp16 = slice_by_index(begin = var_936_begin_0, end = var_936_end_0, end_mask = var_936_end_mask_0, x = k_3_cast_fp16)[name = string("op_936_cast_fp16")];
            tensor<int32, [4]> var_940_begin_0 = const()[name = string("op_940_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_940_end_0 = const()[name = string("op_940_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 384])];
            tensor<bool, [4]> var_940_end_mask_0 = const()[name = string("op_940_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_940_cast_fp16 = slice_by_index(begin = var_940_begin_0, end = var_940_end_0, end_mask = var_940_end_mask_0, x = k_3_cast_fp16)[name = string("op_940_cast_fp16")];
            tensor<int32, [4]> var_942_begin_0 = const()[name = string("op_942_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_942_end_0 = const()[name = string("op_942_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_942_end_mask_0 = const()[name = string("op_942_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_942_cast_fp16 = slice_by_index(begin = var_942_begin_0, end = var_942_end_0, end_mask = var_942_end_mask_0, x = value_3_cast_fp16)[name = string("op_942_cast_fp16")];
            tensor<int32, [4]> var_946_begin_0 = const()[name = string("op_946_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_946_end_0 = const()[name = string("op_946_end_0"), val = tensor<int32, [4]>([1, 128, 1, 1500])];
            tensor<bool, [4]> var_946_end_mask_0 = const()[name = string("op_946_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_946_cast_fp16 = slice_by_index(begin = var_946_begin_0, end = var_946_end_0, end_mask = var_946_end_mask_0, x = value_3_cast_fp16)[name = string("op_946_cast_fp16")];
            tensor<int32, [4]> var_950_begin_0 = const()[name = string("op_950_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_950_end_0 = const()[name = string("op_950_end_0"), val = tensor<int32, [4]>([1, 192, 1, 1500])];
            tensor<bool, [4]> var_950_end_mask_0 = const()[name = string("op_950_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_950_cast_fp16 = slice_by_index(begin = var_950_begin_0, end = var_950_end_0, end_mask = var_950_end_mask_0, x = value_3_cast_fp16)[name = string("op_950_cast_fp16")];
            tensor<int32, [4]> var_954_begin_0 = const()[name = string("op_954_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_954_end_0 = const()[name = string("op_954_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1500])];
            tensor<bool, [4]> var_954_end_mask_0 = const()[name = string("op_954_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_954_cast_fp16 = slice_by_index(begin = var_954_begin_0, end = var_954_end_0, end_mask = var_954_end_mask_0, x = value_3_cast_fp16)[name = string("op_954_cast_fp16")];
            tensor<int32, [4]> var_958_begin_0 = const()[name = string("op_958_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_958_end_0 = const()[name = string("op_958_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1500])];
            tensor<bool, [4]> var_958_end_mask_0 = const()[name = string("op_958_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_958_cast_fp16 = slice_by_index(begin = var_958_begin_0, end = var_958_end_0, end_mask = var_958_end_mask_0, x = value_3_cast_fp16)[name = string("op_958_cast_fp16")];
            tensor<int32, [4]> var_962_begin_0 = const()[name = string("op_962_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_962_end_0 = const()[name = string("op_962_end_0"), val = tensor<int32, [4]>([1, 384, 1, 1500])];
            tensor<bool, [4]> var_962_end_mask_0 = const()[name = string("op_962_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_962_cast_fp16 = slice_by_index(begin = var_962_begin_0, end = var_962_end_0, end_mask = var_962_end_mask_0, x = value_3_cast_fp16)[name = string("op_962_cast_fp16")];
            string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_920_cast_fp16, var_754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")];
            string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_920_cast_fp16, var_761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")];
            string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_920_cast_fp16, var_768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")];
            string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_920_cast_fp16, var_775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")];
            string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_924_cast_fp16, var_782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")];
            string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_924_cast_fp16, var_789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")];
            string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_924_cast_fp16, var_796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")];
            string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_924_cast_fp16, var_803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")];
            string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_928_cast_fp16, var_810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")];
            string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_928_cast_fp16, var_817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")];
            string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_928_cast_fp16, var_824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")];
            string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_928_cast_fp16, var_831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")];
            string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_932_cast_fp16, var_838_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")];
            string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_932_cast_fp16, var_845_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")];
            string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_932_cast_fp16, var_852_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")];
            string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_932_cast_fp16, var_859_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")];
            string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_936_cast_fp16, var_866_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")];
            string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_936_cast_fp16, var_873_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")];
            string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_936_cast_fp16, var_880_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")];
            string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_936_cast_fp16, var_887_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")];
            string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_940_cast_fp16, var_894_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")];
            string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_940_cast_fp16, var_901_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")];
            string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_940_cast_fp16, var_908_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")];
            string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_940_cast_fp16, var_915_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")];
            fp16 var_1013_to_fp16 = const()[name = string("op_1013_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_1013_to_fp16)[name = string("aw_chunk_49_cast_fp16")];
            fp16 var_1015_to_fp16 = const()[name = string("op_1015_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_1015_to_fp16)[name = string("aw_chunk_51_cast_fp16")];
            fp16 var_1017_to_fp16 = const()[name = string("op_1017_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_1017_to_fp16)[name = string("aw_chunk_53_cast_fp16")];
            fp16 var_1019_to_fp16 = const()[name = string("op_1019_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_1019_to_fp16)[name = string("aw_chunk_55_cast_fp16")];
            fp16 var_1021_to_fp16 = const()[name = string("op_1021_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_1021_to_fp16)[name = string("aw_chunk_57_cast_fp16")];
            fp16 var_1023_to_fp16 = const()[name = string("op_1023_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_1023_to_fp16)[name = string("aw_chunk_59_cast_fp16")];
            fp16 var_1025_to_fp16 = const()[name = string("op_1025_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_1025_to_fp16)[name = string("aw_chunk_61_cast_fp16")];
            fp16 var_1027_to_fp16 = const()[name = string("op_1027_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_1027_to_fp16)[name = string("aw_chunk_63_cast_fp16")];
            fp16 var_1029_to_fp16 = const()[name = string("op_1029_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_1029_to_fp16)[name = string("aw_chunk_65_cast_fp16")];
            fp16 var_1031_to_fp16 = const()[name = string("op_1031_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_1031_to_fp16)[name = string("aw_chunk_67_cast_fp16")];
            fp16 var_1033_to_fp16 = const()[name = string("op_1033_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_1033_to_fp16)[name = string("aw_chunk_69_cast_fp16")];
            fp16 var_1035_to_fp16 = const()[name = string("op_1035_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_1035_to_fp16)[name = string("aw_chunk_71_cast_fp16")];
            fp16 var_1037_to_fp16 = const()[name = string("op_1037_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_1037_to_fp16)[name = string("aw_chunk_73_cast_fp16")];
            fp16 var_1039_to_fp16 = const()[name = string("op_1039_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_1039_to_fp16)[name = string("aw_chunk_75_cast_fp16")];
            fp16 var_1041_to_fp16 = const()[name = string("op_1041_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_1041_to_fp16)[name = string("aw_chunk_77_cast_fp16")];
            fp16 var_1043_to_fp16 = const()[name = string("op_1043_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_1043_to_fp16)[name = string("aw_chunk_79_cast_fp16")];
            fp16 var_1045_to_fp16 = const()[name = string("op_1045_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_1045_to_fp16)[name = string("aw_chunk_81_cast_fp16")];
            fp16 var_1047_to_fp16 = const()[name = string("op_1047_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_1047_to_fp16)[name = string("aw_chunk_83_cast_fp16")];
            fp16 var_1049_to_fp16 = const()[name = string("op_1049_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_1049_to_fp16)[name = string("aw_chunk_85_cast_fp16")];
            fp16 var_1051_to_fp16 = const()[name = string("op_1051_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_1051_to_fp16)[name = string("aw_chunk_87_cast_fp16")];
            fp16 var_1053_to_fp16 = const()[name = string("op_1053_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_1053_to_fp16)[name = string("aw_chunk_89_cast_fp16")];
            fp16 var_1055_to_fp16 = const()[name = string("op_1055_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_1055_to_fp16)[name = string("aw_chunk_91_cast_fp16")];
            fp16 var_1057_to_fp16 = const()[name = string("op_1057_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_1057_to_fp16)[name = string("aw_chunk_93_cast_fp16")];
            fp16 var_1059_to_fp16 = const()[name = string("op_1059_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_1059_to_fp16)[name = string("aw_chunk_95_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1061_cast_fp16 = softmax(axis = var_670, x = aw_chunk_49_cast_fp16)[name = string("op_1061_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1062_cast_fp16 = softmax(axis = var_670, x = aw_chunk_51_cast_fp16)[name = string("op_1062_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1063_cast_fp16 = softmax(axis = var_670, x = aw_chunk_53_cast_fp16)[name = string("op_1063_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1064_cast_fp16 = softmax(axis = var_670, x = aw_chunk_55_cast_fp16)[name = string("op_1064_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1065_cast_fp16 = softmax(axis = var_670, x = aw_chunk_57_cast_fp16)[name = string("op_1065_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1066_cast_fp16 = softmax(axis = var_670, x = aw_chunk_59_cast_fp16)[name = string("op_1066_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1067_cast_fp16 = softmax(axis = var_670, x = aw_chunk_61_cast_fp16)[name = string("op_1067_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1068_cast_fp16 = softmax(axis = var_670, x = aw_chunk_63_cast_fp16)[name = string("op_1068_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1069_cast_fp16 = softmax(axis = var_670, x = aw_chunk_65_cast_fp16)[name = string("op_1069_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1070_cast_fp16 = softmax(axis = var_670, x = aw_chunk_67_cast_fp16)[name = string("op_1070_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1071_cast_fp16 = softmax(axis = var_670, x = aw_chunk_69_cast_fp16)[name = string("op_1071_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1072_cast_fp16 = softmax(axis = var_670, x = aw_chunk_71_cast_fp16)[name = string("op_1072_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1073_cast_fp16 = softmax(axis = var_670, x = aw_chunk_73_cast_fp16)[name = string("op_1073_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1074_cast_fp16 = softmax(axis = var_670, x = aw_chunk_75_cast_fp16)[name = string("op_1074_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1075_cast_fp16 = softmax(axis = var_670, x = aw_chunk_77_cast_fp16)[name = string("op_1075_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1076_cast_fp16 = softmax(axis = var_670, x = aw_chunk_79_cast_fp16)[name = string("op_1076_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1077_cast_fp16 = softmax(axis = var_670, x = aw_chunk_81_cast_fp16)[name = string("op_1077_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1078_cast_fp16 = softmax(axis = var_670, x = aw_chunk_83_cast_fp16)[name = string("op_1078_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1079_cast_fp16 = softmax(axis = var_670, x = aw_chunk_85_cast_fp16)[name = string("op_1079_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1080_cast_fp16 = softmax(axis = var_670, x = aw_chunk_87_cast_fp16)[name = string("op_1080_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1081_cast_fp16 = softmax(axis = var_670, x = aw_chunk_89_cast_fp16)[name = string("op_1081_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1082_cast_fp16 = softmax(axis = var_670, x = aw_chunk_91_cast_fp16)[name = string("op_1082_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1083_cast_fp16 = softmax(axis = var_670, x = aw_chunk_93_cast_fp16)[name = string("op_1083_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1084_cast_fp16 = softmax(axis = var_670, x = aw_chunk_95_cast_fp16)[name = string("op_1084_cast_fp16")];
            string var_1086_equation_0 = const()[name = string("op_1086_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1086_cast_fp16 = einsum(equation = var_1086_equation_0, values = (var_942_cast_fp16, var_1061_cast_fp16))[name = string("op_1086_cast_fp16")];
            string var_1088_equation_0 = const()[name = string("op_1088_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1088_cast_fp16 = einsum(equation = var_1088_equation_0, values = (var_942_cast_fp16, var_1062_cast_fp16))[name = string("op_1088_cast_fp16")];
            string var_1090_equation_0 = const()[name = string("op_1090_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1090_cast_fp16 = einsum(equation = var_1090_equation_0, values = (var_942_cast_fp16, var_1063_cast_fp16))[name = string("op_1090_cast_fp16")];
            string var_1092_equation_0 = const()[name = string("op_1092_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1092_cast_fp16 = einsum(equation = var_1092_equation_0, values = (var_942_cast_fp16, var_1064_cast_fp16))[name = string("op_1092_cast_fp16")];
            string var_1094_equation_0 = const()[name = string("op_1094_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1094_cast_fp16 = einsum(equation = var_1094_equation_0, values = (var_946_cast_fp16, var_1065_cast_fp16))[name = string("op_1094_cast_fp16")];
            string var_1096_equation_0 = const()[name = string("op_1096_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1096_cast_fp16 = einsum(equation = var_1096_equation_0, values = (var_946_cast_fp16, var_1066_cast_fp16))[name = string("op_1096_cast_fp16")];
            string var_1098_equation_0 = const()[name = string("op_1098_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1098_cast_fp16 = einsum(equation = var_1098_equation_0, values = (var_946_cast_fp16, var_1067_cast_fp16))[name = string("op_1098_cast_fp16")];
            string var_1100_equation_0 = const()[name = string("op_1100_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1100_cast_fp16 = einsum(equation = var_1100_equation_0, values = (var_946_cast_fp16, var_1068_cast_fp16))[name = string("op_1100_cast_fp16")];
            string var_1102_equation_0 = const()[name = string("op_1102_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1102_cast_fp16 = einsum(equation = var_1102_equation_0, values = (var_950_cast_fp16, var_1069_cast_fp16))[name = string("op_1102_cast_fp16")];
            string var_1104_equation_0 = const()[name = string("op_1104_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1104_cast_fp16 = einsum(equation = var_1104_equation_0, values = (var_950_cast_fp16, var_1070_cast_fp16))[name = string("op_1104_cast_fp16")];
            string var_1106_equation_0 = const()[name = string("op_1106_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1106_cast_fp16 = einsum(equation = var_1106_equation_0, values = (var_950_cast_fp16, var_1071_cast_fp16))[name = string("op_1106_cast_fp16")];
            string var_1108_equation_0 = const()[name = string("op_1108_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1108_cast_fp16 = einsum(equation = var_1108_equation_0, values = (var_950_cast_fp16, var_1072_cast_fp16))[name = string("op_1108_cast_fp16")];
            string var_1110_equation_0 = const()[name = string("op_1110_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1110_cast_fp16 = einsum(equation = var_1110_equation_0, values = (var_954_cast_fp16, var_1073_cast_fp16))[name = string("op_1110_cast_fp16")];
            string var_1112_equation_0 = const()[name = string("op_1112_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1112_cast_fp16 = einsum(equation = var_1112_equation_0, values = (var_954_cast_fp16, var_1074_cast_fp16))[name = string("op_1112_cast_fp16")];
            string var_1114_equation_0 = const()[name = string("op_1114_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1114_cast_fp16 = einsum(equation = var_1114_equation_0, values = (var_954_cast_fp16, var_1075_cast_fp16))[name = string("op_1114_cast_fp16")];
            string var_1116_equation_0 = const()[name = string("op_1116_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1116_cast_fp16 = einsum(equation = var_1116_equation_0, values = (var_954_cast_fp16, var_1076_cast_fp16))[name = string("op_1116_cast_fp16")];
            string var_1118_equation_0 = const()[name = string("op_1118_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1118_cast_fp16 = einsum(equation = var_1118_equation_0, values = (var_958_cast_fp16, var_1077_cast_fp16))[name = string("op_1118_cast_fp16")];
            string var_1120_equation_0 = const()[name = string("op_1120_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1120_cast_fp16 = einsum(equation = var_1120_equation_0, values = (var_958_cast_fp16, var_1078_cast_fp16))[name = string("op_1120_cast_fp16")];
            string var_1122_equation_0 = const()[name = string("op_1122_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1122_cast_fp16 = einsum(equation = var_1122_equation_0, values = (var_958_cast_fp16, var_1079_cast_fp16))[name = string("op_1122_cast_fp16")];
            string var_1124_equation_0 = const()[name = string("op_1124_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1124_cast_fp16 = einsum(equation = var_1124_equation_0, values = (var_958_cast_fp16, var_1080_cast_fp16))[name = string("op_1124_cast_fp16")];
            string var_1126_equation_0 = const()[name = string("op_1126_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1126_cast_fp16 = einsum(equation = var_1126_equation_0, values = (var_962_cast_fp16, var_1081_cast_fp16))[name = string("op_1126_cast_fp16")];
            string var_1128_equation_0 = const()[name = string("op_1128_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1128_cast_fp16 = einsum(equation = var_1128_equation_0, values = (var_962_cast_fp16, var_1082_cast_fp16))[name = string("op_1128_cast_fp16")];
            string var_1130_equation_0 = const()[name = string("op_1130_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1130_cast_fp16 = einsum(equation = var_1130_equation_0, values = (var_962_cast_fp16, var_1083_cast_fp16))[name = string("op_1130_cast_fp16")];
            string var_1132_equation_0 = const()[name = string("op_1132_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1132_cast_fp16 = einsum(equation = var_1132_equation_0, values = (var_962_cast_fp16, var_1084_cast_fp16))[name = string("op_1132_cast_fp16")];
            bool var_1134_interleave_0 = const()[name = string("op_1134_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1134_cast_fp16 = concat(axis = var_659, interleave = var_1134_interleave_0, values = (var_1086_cast_fp16, var_1088_cast_fp16, var_1090_cast_fp16, var_1092_cast_fp16))[name = string("op_1134_cast_fp16")];
            bool var_1136_interleave_0 = const()[name = string("op_1136_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1136_cast_fp16 = concat(axis = var_659, interleave = var_1136_interleave_0, values = (var_1094_cast_fp16, var_1096_cast_fp16, var_1098_cast_fp16, var_1100_cast_fp16))[name = string("op_1136_cast_fp16")];
            bool var_1138_interleave_0 = const()[name = string("op_1138_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1138_cast_fp16 = concat(axis = var_659, interleave = var_1138_interleave_0, values = (var_1102_cast_fp16, var_1104_cast_fp16, var_1106_cast_fp16, var_1108_cast_fp16))[name = string("op_1138_cast_fp16")];
            bool var_1140_interleave_0 = const()[name = string("op_1140_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1140_cast_fp16 = concat(axis = var_659, interleave = var_1140_interleave_0, values = (var_1110_cast_fp16, var_1112_cast_fp16, var_1114_cast_fp16, var_1116_cast_fp16))[name = string("op_1140_cast_fp16")];
            bool var_1142_interleave_0 = const()[name = string("op_1142_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1142_cast_fp16 = concat(axis = var_659, interleave = var_1142_interleave_0, values = (var_1118_cast_fp16, var_1120_cast_fp16, var_1122_cast_fp16, var_1124_cast_fp16))[name = string("op_1142_cast_fp16")];
            bool var_1144_interleave_0 = const()[name = string("op_1144_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1144_cast_fp16 = concat(axis = var_659, interleave = var_1144_interleave_0, values = (var_1126_cast_fp16, var_1128_cast_fp16, var_1130_cast_fp16, var_1132_cast_fp16))[name = string("op_1144_cast_fp16")];
            bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 384, 1, 1500]> input_9_cast_fp16 = concat(axis = var_670, interleave = input_9_interleave_0, values = (var_1134_cast_fp16, var_1136_cast_fp16, var_1138_cast_fp16, var_1140_cast_fp16, var_1142_cast_fp16, var_1144_cast_fp16))[name = string("input_9_cast_fp16")];
            string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6662016)))];
            tensor<fp16, [384]> layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6956992)))];
            tensor<fp16, [1, 384, 1, 1500]> obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")];
            tensor<int32, [1]> out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1163_to_fp16 = const()[name = string("op_1163_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_1163_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")];
            tensor<fp16, [384]> input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6957824)))];
            tensor<fp16, [384]> input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6958656)))];
            fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")];
            string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)];
            tensor<fp16, [1536, 384, 1, 1]> layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [1536, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6959488)))];
            tensor<fp16, [1536]> layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8139200)))];
            tensor<fp16, [1, 1536, 1, 1500]> input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")];
            string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1536, 1, 1500]> input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")];
            string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)];
            tensor<fp16, [384, 1536, 1, 1]> layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [384, 1536, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8142336)))];
            tensor<fp16, [384]> layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322048)))];
            tensor<fp16, [1, 384, 1, 1500]> hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")];
            int32 var_1192 = const()[name = string("op_1192"), val = int32(3)];
            int32 var_1203 = const()[name = string("op_1203"), val = int32(1)];
            tensor<int32, [1]> out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1220_to_fp16 = const()[name = string("op_1220_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_1220_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")];
            tensor<fp16, [384]> obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9322880)))];
            tensor<fp16, [384]> obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9323712)))];
            fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")];
            string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9324544)))];
            tensor<fp16, [384]> layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9619520)))];
            tensor<fp16, [1, 384, 1, 1500]> query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")];
            string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9620352)))];
            tensor<fp16, [1, 384, 1, 1500]> key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")];
            string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9915328)))];
            tensor<fp16, [384]> layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10210304)))];
            tensor<fp16, [1, 384, 1, 1500]> value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")];
            tensor<int32, [4]> var_1258_begin_0 = const()[name = string("op_1258_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1258_end_0 = const()[name = string("op_1258_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1258_end_mask_0 = const()[name = string("op_1258_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1258_cast_fp16 = slice_by_index(begin = var_1258_begin_0, end = var_1258_end_0, end_mask = var_1258_end_mask_0, x = query_5_cast_fp16)[name = string("op_1258_cast_fp16")];
            tensor<int32, [4]> var_1262_begin_0 = const()[name = string("op_1262_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1262_end_0 = const()[name = string("op_1262_end_0"), val = tensor<int32, [4]>([1, 128, 1, 1500])];
            tensor<bool, [4]> var_1262_end_mask_0 = const()[name = string("op_1262_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1262_cast_fp16 = slice_by_index(begin = var_1262_begin_0, end = var_1262_end_0, end_mask = var_1262_end_mask_0, x = query_5_cast_fp16)[name = string("op_1262_cast_fp16")];
            tensor<int32, [4]> var_1266_begin_0 = const()[name = string("op_1266_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1266_end_0 = const()[name = string("op_1266_end_0"), val = tensor<int32, [4]>([1, 192, 1, 1500])];
            tensor<bool, [4]> var_1266_end_mask_0 = const()[name = string("op_1266_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1266_cast_fp16 = slice_by_index(begin = var_1266_begin_0, end = var_1266_end_0, end_mask = var_1266_end_mask_0, x = query_5_cast_fp16)[name = string("op_1266_cast_fp16")];
            tensor<int32, [4]> var_1270_begin_0 = const()[name = string("op_1270_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1270_end_0 = const()[name = string("op_1270_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1500])];
            tensor<bool, [4]> var_1270_end_mask_0 = const()[name = string("op_1270_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1270_cast_fp16 = slice_by_index(begin = var_1270_begin_0, end = var_1270_end_0, end_mask = var_1270_end_mask_0, x = query_5_cast_fp16)[name = string("op_1270_cast_fp16")];
            tensor<int32, [4]> var_1274_begin_0 = const()[name = string("op_1274_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1274_end_0 = const()[name = string("op_1274_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1500])];
            tensor<bool, [4]> var_1274_end_mask_0 = const()[name = string("op_1274_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1274_cast_fp16 = slice_by_index(begin = var_1274_begin_0, end = var_1274_end_0, end_mask = var_1274_end_mask_0, x = query_5_cast_fp16)[name = string("op_1274_cast_fp16")];
            tensor<int32, [4]> var_1278_begin_0 = const()[name = string("op_1278_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1278_end_0 = const()[name = string("op_1278_end_0"), val = tensor<int32, [4]>([1, 384, 1, 1500])];
            tensor<bool, [4]> var_1278_end_mask_0 = const()[name = string("op_1278_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1278_cast_fp16 = slice_by_index(begin = var_1278_begin_0, end = var_1278_end_0, end_mask = var_1278_end_mask_0, x = query_5_cast_fp16)[name = string("op_1278_cast_fp16")];
            tensor<int32, [4]> var_1287_begin_0 = const()[name = string("op_1287_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1287_end_0 = const()[name = string("op_1287_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1287_end_mask_0 = const()[name = string("op_1287_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1287_cast_fp16 = slice_by_index(begin = var_1287_begin_0, end = var_1287_end_0, end_mask = var_1287_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1287_cast_fp16")];
            tensor<int32, [4]> var_1294_begin_0 = const()[name = string("op_1294_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1294_end_0 = const()[name = string("op_1294_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1294_end_mask_0 = const()[name = string("op_1294_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1294_cast_fp16 = slice_by_index(begin = var_1294_begin_0, end = var_1294_end_0, end_mask = var_1294_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1294_cast_fp16")];
            tensor<int32, [4]> var_1301_begin_0 = const()[name = string("op_1301_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1301_end_0 = const()[name = string("op_1301_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1301_end_mask_0 = const()[name = string("op_1301_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1301_cast_fp16 = slice_by_index(begin = var_1301_begin_0, end = var_1301_end_0, end_mask = var_1301_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1301_cast_fp16")];
            tensor<int32, [4]> var_1308_begin_0 = const()[name = string("op_1308_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1308_end_0 = const()[name = string("op_1308_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1308_end_mask_0 = const()[name = string("op_1308_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1308_cast_fp16 = slice_by_index(begin = var_1308_begin_0, end = var_1308_end_0, end_mask = var_1308_end_mask_0, x = var_1258_cast_fp16)[name = string("op_1308_cast_fp16")];
            tensor<int32, [4]> var_1315_begin_0 = const()[name = string("op_1315_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1315_end_0 = const()[name = string("op_1315_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1315_end_mask_0 = const()[name = string("op_1315_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1315_cast_fp16 = slice_by_index(begin = var_1315_begin_0, end = var_1315_end_0, end_mask = var_1315_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1315_cast_fp16")];
            tensor<int32, [4]> var_1322_begin_0 = const()[name = string("op_1322_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1322_end_0 = const()[name = string("op_1322_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1322_end_mask_0 = const()[name = string("op_1322_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1322_cast_fp16 = slice_by_index(begin = var_1322_begin_0, end = var_1322_end_0, end_mask = var_1322_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1322_cast_fp16")];
            tensor<int32, [4]> var_1329_begin_0 = const()[name = string("op_1329_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1329_end_0 = const()[name = string("op_1329_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1329_end_mask_0 = const()[name = string("op_1329_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1329_cast_fp16 = slice_by_index(begin = var_1329_begin_0, end = var_1329_end_0, end_mask = var_1329_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1329_cast_fp16")];
            tensor<int32, [4]> var_1336_begin_0 = const()[name = string("op_1336_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1336_end_0 = const()[name = string("op_1336_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1336_end_mask_0 = const()[name = string("op_1336_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1336_cast_fp16 = slice_by_index(begin = var_1336_begin_0, end = var_1336_end_0, end_mask = var_1336_end_mask_0, x = var_1262_cast_fp16)[name = string("op_1336_cast_fp16")];
            tensor<int32, [4]> var_1343_begin_0 = const()[name = string("op_1343_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1343_end_0 = const()[name = string("op_1343_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1343_end_mask_0 = const()[name = string("op_1343_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1343_cast_fp16 = slice_by_index(begin = var_1343_begin_0, end = var_1343_end_0, end_mask = var_1343_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1343_cast_fp16")];
            tensor<int32, [4]> var_1350_begin_0 = const()[name = string("op_1350_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1350_end_0 = const()[name = string("op_1350_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1350_end_mask_0 = const()[name = string("op_1350_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1350_cast_fp16 = slice_by_index(begin = var_1350_begin_0, end = var_1350_end_0, end_mask = var_1350_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1350_cast_fp16")];
            tensor<int32, [4]> var_1357_begin_0 = const()[name = string("op_1357_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1357_end_0 = const()[name = string("op_1357_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1357_end_mask_0 = const()[name = string("op_1357_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1357_cast_fp16 = slice_by_index(begin = var_1357_begin_0, end = var_1357_end_0, end_mask = var_1357_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1357_cast_fp16")];
            tensor<int32, [4]> var_1364_begin_0 = const()[name = string("op_1364_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1364_end_0 = const()[name = string("op_1364_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1364_end_mask_0 = const()[name = string("op_1364_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1364_cast_fp16 = slice_by_index(begin = var_1364_begin_0, end = var_1364_end_0, end_mask = var_1364_end_mask_0, x = var_1266_cast_fp16)[name = string("op_1364_cast_fp16")];
            tensor<int32, [4]> var_1371_begin_0 = const()[name = string("op_1371_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1371_end_0 = const()[name = string("op_1371_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1371_end_mask_0 = const()[name = string("op_1371_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1371_cast_fp16 = slice_by_index(begin = var_1371_begin_0, end = var_1371_end_0, end_mask = var_1371_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1371_cast_fp16")];
            tensor<int32, [4]> var_1378_begin_0 = const()[name = string("op_1378_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1378_end_0 = const()[name = string("op_1378_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1378_end_mask_0 = const()[name = string("op_1378_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1378_cast_fp16 = slice_by_index(begin = var_1378_begin_0, end = var_1378_end_0, end_mask = var_1378_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1378_cast_fp16")];
            tensor<int32, [4]> var_1385_begin_0 = const()[name = string("op_1385_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1385_end_0 = const()[name = string("op_1385_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1385_end_mask_0 = const()[name = string("op_1385_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1385_cast_fp16 = slice_by_index(begin = var_1385_begin_0, end = var_1385_end_0, end_mask = var_1385_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1385_cast_fp16")];
            tensor<int32, [4]> var_1392_begin_0 = const()[name = string("op_1392_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1392_end_0 = const()[name = string("op_1392_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1392_end_mask_0 = const()[name = string("op_1392_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1392_cast_fp16 = slice_by_index(begin = var_1392_begin_0, end = var_1392_end_0, end_mask = var_1392_end_mask_0, x = var_1270_cast_fp16)[name = string("op_1392_cast_fp16")];
            tensor<int32, [4]> var_1399_begin_0 = const()[name = string("op_1399_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1399_end_0 = const()[name = string("op_1399_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1399_end_mask_0 = const()[name = string("op_1399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1399_cast_fp16 = slice_by_index(begin = var_1399_begin_0, end = var_1399_end_0, end_mask = var_1399_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1399_cast_fp16")];
            tensor<int32, [4]> var_1406_begin_0 = const()[name = string("op_1406_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1406_end_0 = const()[name = string("op_1406_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1406_end_mask_0 = const()[name = string("op_1406_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1406_cast_fp16 = slice_by_index(begin = var_1406_begin_0, end = var_1406_end_0, end_mask = var_1406_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1406_cast_fp16")];
            tensor<int32, [4]> var_1413_begin_0 = const()[name = string("op_1413_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1413_end_0 = const()[name = string("op_1413_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1413_end_mask_0 = const()[name = string("op_1413_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1413_cast_fp16 = slice_by_index(begin = var_1413_begin_0, end = var_1413_end_0, end_mask = var_1413_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1413_cast_fp16")];
            tensor<int32, [4]> var_1420_begin_0 = const()[name = string("op_1420_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1420_end_0 = const()[name = string("op_1420_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1420_end_mask_0 = const()[name = string("op_1420_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1420_cast_fp16 = slice_by_index(begin = var_1420_begin_0, end = var_1420_end_0, end_mask = var_1420_end_mask_0, x = var_1274_cast_fp16)[name = string("op_1420_cast_fp16")];
            tensor<int32, [4]> var_1427_begin_0 = const()[name = string("op_1427_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1427_end_0 = const()[name = string("op_1427_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1427_end_mask_0 = const()[name = string("op_1427_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1427_cast_fp16 = slice_by_index(begin = var_1427_begin_0, end = var_1427_end_0, end_mask = var_1427_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1427_cast_fp16")];
            tensor<int32, [4]> var_1434_begin_0 = const()[name = string("op_1434_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1434_end_0 = const()[name = string("op_1434_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1434_end_mask_0 = const()[name = string("op_1434_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1434_cast_fp16 = slice_by_index(begin = var_1434_begin_0, end = var_1434_end_0, end_mask = var_1434_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1434_cast_fp16")];
            tensor<int32, [4]> var_1441_begin_0 = const()[name = string("op_1441_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1441_end_0 = const()[name = string("op_1441_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1441_end_mask_0 = const()[name = string("op_1441_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1441_cast_fp16 = slice_by_index(begin = var_1441_begin_0, end = var_1441_end_0, end_mask = var_1441_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1441_cast_fp16")];
            tensor<int32, [4]> var_1448_begin_0 = const()[name = string("op_1448_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1448_end_0 = const()[name = string("op_1448_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1448_end_mask_0 = const()[name = string("op_1448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1448_cast_fp16 = slice_by_index(begin = var_1448_begin_0, end = var_1448_end_0, end_mask = var_1448_end_mask_0, x = var_1278_cast_fp16)[name = string("op_1448_cast_fp16")];
            tensor<int32, [4]> k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1453_begin_0 = const()[name = string("op_1453_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1453_end_0 = const()[name = string("op_1453_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 64])];
            tensor<bool, [4]> var_1453_end_mask_0 = const()[name = string("op_1453_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 384]> k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 1500, 1, 64]> var_1453_cast_fp16 = slice_by_index(begin = var_1453_begin_0, end = var_1453_end_0, end_mask = var_1453_end_mask_0, x = k_5_cast_fp16)[name = string("op_1453_cast_fp16")];
            tensor<int32, [4]> var_1457_begin_0 = const()[name = string("op_1457_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1457_end_0 = const()[name = string("op_1457_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 128])];
            tensor<bool, [4]> var_1457_end_mask_0 = const()[name = string("op_1457_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_1457_cast_fp16 = slice_by_index(begin = var_1457_begin_0, end = var_1457_end_0, end_mask = var_1457_end_mask_0, x = k_5_cast_fp16)[name = string("op_1457_cast_fp16")];
            tensor<int32, [4]> var_1461_begin_0 = const()[name = string("op_1461_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1461_end_0 = const()[name = string("op_1461_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 192])];
            tensor<bool, [4]> var_1461_end_mask_0 = const()[name = string("op_1461_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_1461_cast_fp16 = slice_by_index(begin = var_1461_begin_0, end = var_1461_end_0, end_mask = var_1461_end_mask_0, x = k_5_cast_fp16)[name = string("op_1461_cast_fp16")];
            tensor<int32, [4]> var_1465_begin_0 = const()[name = string("op_1465_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1465_end_0 = const()[name = string("op_1465_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 256])];
            tensor<bool, [4]> var_1465_end_mask_0 = const()[name = string("op_1465_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_1465_cast_fp16 = slice_by_index(begin = var_1465_begin_0, end = var_1465_end_0, end_mask = var_1465_end_mask_0, x = k_5_cast_fp16)[name = string("op_1465_cast_fp16")];
            tensor<int32, [4]> var_1469_begin_0 = const()[name = string("op_1469_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_1469_end_0 = const()[name = string("op_1469_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 320])];
            tensor<bool, [4]> var_1469_end_mask_0 = const()[name = string("op_1469_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_1469_cast_fp16 = slice_by_index(begin = var_1469_begin_0, end = var_1469_end_0, end_mask = var_1469_end_mask_0, x = k_5_cast_fp16)[name = string("op_1469_cast_fp16")];
            tensor<int32, [4]> var_1473_begin_0 = const()[name = string("op_1473_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_1473_end_0 = const()[name = string("op_1473_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 384])];
            tensor<bool, [4]> var_1473_end_mask_0 = const()[name = string("op_1473_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_1473_cast_fp16 = slice_by_index(begin = var_1473_begin_0, end = var_1473_end_0, end_mask = var_1473_end_mask_0, x = k_5_cast_fp16)[name = string("op_1473_cast_fp16")];
            tensor<int32, [4]> var_1475_begin_0 = const()[name = string("op_1475_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1475_end_0 = const()[name = string("op_1475_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1475_end_mask_0 = const()[name = string("op_1475_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1475_cast_fp16 = slice_by_index(begin = var_1475_begin_0, end = var_1475_end_0, end_mask = var_1475_end_mask_0, x = value_5_cast_fp16)[name = string("op_1475_cast_fp16")];
            tensor<int32, [4]> var_1479_begin_0 = const()[name = string("op_1479_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1479_end_0 = const()[name = string("op_1479_end_0"), val = tensor<int32, [4]>([1, 128, 1, 1500])];
            tensor<bool, [4]> var_1479_end_mask_0 = const()[name = string("op_1479_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1479_cast_fp16 = slice_by_index(begin = var_1479_begin_0, end = var_1479_end_0, end_mask = var_1479_end_mask_0, x = value_5_cast_fp16)[name = string("op_1479_cast_fp16")];
            tensor<int32, [4]> var_1483_begin_0 = const()[name = string("op_1483_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1483_end_0 = const()[name = string("op_1483_end_0"), val = tensor<int32, [4]>([1, 192, 1, 1500])];
            tensor<bool, [4]> var_1483_end_mask_0 = const()[name = string("op_1483_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1483_cast_fp16 = slice_by_index(begin = var_1483_begin_0, end = var_1483_end_0, end_mask = var_1483_end_mask_0, x = value_5_cast_fp16)[name = string("op_1483_cast_fp16")];
            tensor<int32, [4]> var_1487_begin_0 = const()[name = string("op_1487_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1487_end_0 = const()[name = string("op_1487_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1500])];
            tensor<bool, [4]> var_1487_end_mask_0 = const()[name = string("op_1487_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1487_cast_fp16 = slice_by_index(begin = var_1487_begin_0, end = var_1487_end_0, end_mask = var_1487_end_mask_0, x = value_5_cast_fp16)[name = string("op_1487_cast_fp16")];
            tensor<int32, [4]> var_1491_begin_0 = const()[name = string("op_1491_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1491_end_0 = const()[name = string("op_1491_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1500])];
            tensor<bool, [4]> var_1491_end_mask_0 = const()[name = string("op_1491_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1491_cast_fp16 = slice_by_index(begin = var_1491_begin_0, end = var_1491_end_0, end_mask = var_1491_end_mask_0, x = value_5_cast_fp16)[name = string("op_1491_cast_fp16")];
            tensor<int32, [4]> var_1495_begin_0 = const()[name = string("op_1495_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1495_end_0 = const()[name = string("op_1495_end_0"), val = tensor<int32, [4]>([1, 384, 1, 1500])];
            tensor<bool, [4]> var_1495_end_mask_0 = const()[name = string("op_1495_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1495_cast_fp16 = slice_by_index(begin = var_1495_begin_0, end = var_1495_end_0, end_mask = var_1495_end_mask_0, x = value_5_cast_fp16)[name = string("op_1495_cast_fp16")];
            string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1453_cast_fp16, var_1287_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")];
            string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1453_cast_fp16, var_1294_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")];
            string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1453_cast_fp16, var_1301_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")];
            string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1453_cast_fp16, var_1308_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")];
            string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1457_cast_fp16, var_1315_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")];
            string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1457_cast_fp16, var_1322_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")];
            string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1457_cast_fp16, var_1329_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")];
            string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1457_cast_fp16, var_1336_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")];
            string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1461_cast_fp16, var_1343_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")];
            string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1461_cast_fp16, var_1350_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")];
            string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1461_cast_fp16, var_1357_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")];
            string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1461_cast_fp16, var_1364_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")];
            string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1465_cast_fp16, var_1371_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")];
            string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1465_cast_fp16, var_1378_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")];
            string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1465_cast_fp16, var_1385_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")];
            string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1465_cast_fp16, var_1392_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")];
            string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1469_cast_fp16, var_1399_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")];
            string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1469_cast_fp16, var_1406_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")];
            string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1469_cast_fp16, var_1413_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")];
            string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1469_cast_fp16, var_1420_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")];
            string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1473_cast_fp16, var_1427_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")];
            string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1473_cast_fp16, var_1434_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")];
            string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1473_cast_fp16, var_1441_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")];
            string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1473_cast_fp16, var_1448_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")];
            fp16 var_1546_to_fp16 = const()[name = string("op_1546_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1546_to_fp16)[name = string("aw_chunk_97_cast_fp16")];
            fp16 var_1548_to_fp16 = const()[name = string("op_1548_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1548_to_fp16)[name = string("aw_chunk_99_cast_fp16")];
            fp16 var_1550_to_fp16 = const()[name = string("op_1550_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1550_to_fp16)[name = string("aw_chunk_101_cast_fp16")];
            fp16 var_1552_to_fp16 = const()[name = string("op_1552_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1552_to_fp16)[name = string("aw_chunk_103_cast_fp16")];
            fp16 var_1554_to_fp16 = const()[name = string("op_1554_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1554_to_fp16)[name = string("aw_chunk_105_cast_fp16")];
            fp16 var_1556_to_fp16 = const()[name = string("op_1556_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1556_to_fp16)[name = string("aw_chunk_107_cast_fp16")];
            fp16 var_1558_to_fp16 = const()[name = string("op_1558_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1558_to_fp16)[name = string("aw_chunk_109_cast_fp16")];
            fp16 var_1560_to_fp16 = const()[name = string("op_1560_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1560_to_fp16)[name = string("aw_chunk_111_cast_fp16")];
            fp16 var_1562_to_fp16 = const()[name = string("op_1562_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1562_to_fp16)[name = string("aw_chunk_113_cast_fp16")];
            fp16 var_1564_to_fp16 = const()[name = string("op_1564_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1564_to_fp16)[name = string("aw_chunk_115_cast_fp16")];
            fp16 var_1566_to_fp16 = const()[name = string("op_1566_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1566_to_fp16)[name = string("aw_chunk_117_cast_fp16")];
            fp16 var_1568_to_fp16 = const()[name = string("op_1568_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1568_to_fp16)[name = string("aw_chunk_119_cast_fp16")];
            fp16 var_1570_to_fp16 = const()[name = string("op_1570_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1570_to_fp16)[name = string("aw_chunk_121_cast_fp16")];
            fp16 var_1572_to_fp16 = const()[name = string("op_1572_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1572_to_fp16)[name = string("aw_chunk_123_cast_fp16")];
            fp16 var_1574_to_fp16 = const()[name = string("op_1574_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1574_to_fp16)[name = string("aw_chunk_125_cast_fp16")];
            fp16 var_1576_to_fp16 = const()[name = string("op_1576_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1576_to_fp16)[name = string("aw_chunk_127_cast_fp16")];
            fp16 var_1578_to_fp16 = const()[name = string("op_1578_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1578_to_fp16)[name = string("aw_chunk_129_cast_fp16")];
            fp16 var_1580_to_fp16 = const()[name = string("op_1580_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1580_to_fp16)[name = string("aw_chunk_131_cast_fp16")];
            fp16 var_1582_to_fp16 = const()[name = string("op_1582_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1582_to_fp16)[name = string("aw_chunk_133_cast_fp16")];
            fp16 var_1584_to_fp16 = const()[name = string("op_1584_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1584_to_fp16)[name = string("aw_chunk_135_cast_fp16")];
            fp16 var_1586_to_fp16 = const()[name = string("op_1586_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1586_to_fp16)[name = string("aw_chunk_137_cast_fp16")];
            fp16 var_1588_to_fp16 = const()[name = string("op_1588_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1588_to_fp16)[name = string("aw_chunk_139_cast_fp16")];
            fp16 var_1590_to_fp16 = const()[name = string("op_1590_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1590_to_fp16)[name = string("aw_chunk_141_cast_fp16")];
            fp16 var_1592_to_fp16 = const()[name = string("op_1592_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1592_to_fp16)[name = string("aw_chunk_143_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1594_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_97_cast_fp16)[name = string("op_1594_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1595_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_99_cast_fp16)[name = string("op_1595_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1596_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_101_cast_fp16)[name = string("op_1596_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1597_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_103_cast_fp16)[name = string("op_1597_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1598_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_105_cast_fp16)[name = string("op_1598_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1599_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_107_cast_fp16)[name = string("op_1599_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1600_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_109_cast_fp16)[name = string("op_1600_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1601_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_111_cast_fp16)[name = string("op_1601_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1602_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_113_cast_fp16)[name = string("op_1602_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1603_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_115_cast_fp16)[name = string("op_1603_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1604_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_117_cast_fp16)[name = string("op_1604_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1605_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_119_cast_fp16)[name = string("op_1605_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1606_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_121_cast_fp16)[name = string("op_1606_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1607_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_123_cast_fp16)[name = string("op_1607_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1608_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_125_cast_fp16)[name = string("op_1608_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1609_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_127_cast_fp16)[name = string("op_1609_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1610_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_129_cast_fp16)[name = string("op_1610_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1611_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_131_cast_fp16)[name = string("op_1611_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1612_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_133_cast_fp16)[name = string("op_1612_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1613_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_135_cast_fp16)[name = string("op_1613_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1614_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_137_cast_fp16)[name = string("op_1614_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1615_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_139_cast_fp16)[name = string("op_1615_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1616_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_141_cast_fp16)[name = string("op_1616_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_1617_cast_fp16 = softmax(axis = var_1203, x = aw_chunk_143_cast_fp16)[name = string("op_1617_cast_fp16")];
            string var_1619_equation_0 = const()[name = string("op_1619_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1619_cast_fp16 = einsum(equation = var_1619_equation_0, values = (var_1475_cast_fp16, var_1594_cast_fp16))[name = string("op_1619_cast_fp16")];
            string var_1621_equation_0 = const()[name = string("op_1621_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1621_cast_fp16 = einsum(equation = var_1621_equation_0, values = (var_1475_cast_fp16, var_1595_cast_fp16))[name = string("op_1621_cast_fp16")];
            string var_1623_equation_0 = const()[name = string("op_1623_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1623_cast_fp16 = einsum(equation = var_1623_equation_0, values = (var_1475_cast_fp16, var_1596_cast_fp16))[name = string("op_1623_cast_fp16")];
            string var_1625_equation_0 = const()[name = string("op_1625_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1625_cast_fp16 = einsum(equation = var_1625_equation_0, values = (var_1475_cast_fp16, var_1597_cast_fp16))[name = string("op_1625_cast_fp16")];
            string var_1627_equation_0 = const()[name = string("op_1627_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1627_cast_fp16 = einsum(equation = var_1627_equation_0, values = (var_1479_cast_fp16, var_1598_cast_fp16))[name = string("op_1627_cast_fp16")];
            string var_1629_equation_0 = const()[name = string("op_1629_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1629_cast_fp16 = einsum(equation = var_1629_equation_0, values = (var_1479_cast_fp16, var_1599_cast_fp16))[name = string("op_1629_cast_fp16")];
            string var_1631_equation_0 = const()[name = string("op_1631_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1631_cast_fp16 = einsum(equation = var_1631_equation_0, values = (var_1479_cast_fp16, var_1600_cast_fp16))[name = string("op_1631_cast_fp16")];
            string var_1633_equation_0 = const()[name = string("op_1633_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1633_cast_fp16 = einsum(equation = var_1633_equation_0, values = (var_1479_cast_fp16, var_1601_cast_fp16))[name = string("op_1633_cast_fp16")];
            string var_1635_equation_0 = const()[name = string("op_1635_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1635_cast_fp16 = einsum(equation = var_1635_equation_0, values = (var_1483_cast_fp16, var_1602_cast_fp16))[name = string("op_1635_cast_fp16")];
            string var_1637_equation_0 = const()[name = string("op_1637_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1637_cast_fp16 = einsum(equation = var_1637_equation_0, values = (var_1483_cast_fp16, var_1603_cast_fp16))[name = string("op_1637_cast_fp16")];
            string var_1639_equation_0 = const()[name = string("op_1639_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1639_cast_fp16 = einsum(equation = var_1639_equation_0, values = (var_1483_cast_fp16, var_1604_cast_fp16))[name = string("op_1639_cast_fp16")];
            string var_1641_equation_0 = const()[name = string("op_1641_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1641_cast_fp16 = einsum(equation = var_1641_equation_0, values = (var_1483_cast_fp16, var_1605_cast_fp16))[name = string("op_1641_cast_fp16")];
            string var_1643_equation_0 = const()[name = string("op_1643_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1643_cast_fp16 = einsum(equation = var_1643_equation_0, values = (var_1487_cast_fp16, var_1606_cast_fp16))[name = string("op_1643_cast_fp16")];
            string var_1645_equation_0 = const()[name = string("op_1645_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1645_cast_fp16 = einsum(equation = var_1645_equation_0, values = (var_1487_cast_fp16, var_1607_cast_fp16))[name = string("op_1645_cast_fp16")];
            string var_1647_equation_0 = const()[name = string("op_1647_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1647_cast_fp16 = einsum(equation = var_1647_equation_0, values = (var_1487_cast_fp16, var_1608_cast_fp16))[name = string("op_1647_cast_fp16")];
            string var_1649_equation_0 = const()[name = string("op_1649_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1649_cast_fp16 = einsum(equation = var_1649_equation_0, values = (var_1487_cast_fp16, var_1609_cast_fp16))[name = string("op_1649_cast_fp16")];
            string var_1651_equation_0 = const()[name = string("op_1651_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1651_cast_fp16 = einsum(equation = var_1651_equation_0, values = (var_1491_cast_fp16, var_1610_cast_fp16))[name = string("op_1651_cast_fp16")];
            string var_1653_equation_0 = const()[name = string("op_1653_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1653_cast_fp16 = einsum(equation = var_1653_equation_0, values = (var_1491_cast_fp16, var_1611_cast_fp16))[name = string("op_1653_cast_fp16")];
            string var_1655_equation_0 = const()[name = string("op_1655_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1655_cast_fp16 = einsum(equation = var_1655_equation_0, values = (var_1491_cast_fp16, var_1612_cast_fp16))[name = string("op_1655_cast_fp16")];
            string var_1657_equation_0 = const()[name = string("op_1657_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1657_cast_fp16 = einsum(equation = var_1657_equation_0, values = (var_1491_cast_fp16, var_1613_cast_fp16))[name = string("op_1657_cast_fp16")];
            string var_1659_equation_0 = const()[name = string("op_1659_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1659_cast_fp16 = einsum(equation = var_1659_equation_0, values = (var_1495_cast_fp16, var_1614_cast_fp16))[name = string("op_1659_cast_fp16")];
            string var_1661_equation_0 = const()[name = string("op_1661_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1661_cast_fp16 = einsum(equation = var_1661_equation_0, values = (var_1495_cast_fp16, var_1615_cast_fp16))[name = string("op_1661_cast_fp16")];
            string var_1663_equation_0 = const()[name = string("op_1663_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1663_cast_fp16 = einsum(equation = var_1663_equation_0, values = (var_1495_cast_fp16, var_1616_cast_fp16))[name = string("op_1663_cast_fp16")];
            string var_1665_equation_0 = const()[name = string("op_1665_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_1665_cast_fp16 = einsum(equation = var_1665_equation_0, values = (var_1495_cast_fp16, var_1617_cast_fp16))[name = string("op_1665_cast_fp16")];
            bool var_1667_interleave_0 = const()[name = string("op_1667_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1667_cast_fp16 = concat(axis = var_1192, interleave = var_1667_interleave_0, values = (var_1619_cast_fp16, var_1621_cast_fp16, var_1623_cast_fp16, var_1625_cast_fp16))[name = string("op_1667_cast_fp16")];
            bool var_1669_interleave_0 = const()[name = string("op_1669_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1669_cast_fp16 = concat(axis = var_1192, interleave = var_1669_interleave_0, values = (var_1627_cast_fp16, var_1629_cast_fp16, var_1631_cast_fp16, var_1633_cast_fp16))[name = string("op_1669_cast_fp16")];
            bool var_1671_interleave_0 = const()[name = string("op_1671_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1671_cast_fp16 = concat(axis = var_1192, interleave = var_1671_interleave_0, values = (var_1635_cast_fp16, var_1637_cast_fp16, var_1639_cast_fp16, var_1641_cast_fp16))[name = string("op_1671_cast_fp16")];
            bool var_1673_interleave_0 = const()[name = string("op_1673_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1673_cast_fp16 = concat(axis = var_1192, interleave = var_1673_interleave_0, values = (var_1643_cast_fp16, var_1645_cast_fp16, var_1647_cast_fp16, var_1649_cast_fp16))[name = string("op_1673_cast_fp16")];
            bool var_1675_interleave_0 = const()[name = string("op_1675_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1675_cast_fp16 = concat(axis = var_1192, interleave = var_1675_interleave_0, values = (var_1651_cast_fp16, var_1653_cast_fp16, var_1655_cast_fp16, var_1657_cast_fp16))[name = string("op_1675_cast_fp16")];
            bool var_1677_interleave_0 = const()[name = string("op_1677_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_1677_cast_fp16 = concat(axis = var_1192, interleave = var_1677_interleave_0, values = (var_1659_cast_fp16, var_1661_cast_fp16, var_1663_cast_fp16, var_1665_cast_fp16))[name = string("op_1677_cast_fp16")];
            bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 384, 1, 1500]> input_17_cast_fp16 = concat(axis = var_1203, interleave = input_17_interleave_0, values = (var_1667_cast_fp16, var_1669_cast_fp16, var_1671_cast_fp16, var_1673_cast_fp16, var_1675_cast_fp16, var_1677_cast_fp16))[name = string("input_17_cast_fp16")];
            string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10211136)))];
            tensor<fp16, [384]> layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506112)))];
            tensor<fp16, [1, 384, 1, 1500]> obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")];
            tensor<int32, [1]> out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1696_to_fp16 = const()[name = string("op_1696_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_1696_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")];
            tensor<fp16, [384]> input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10506944)))];
            tensor<fp16, [384]> input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10507776)))];
            fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")];
            string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)];
            tensor<fp16, [1536, 384, 1, 1]> layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor<fp16, [1536, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10508608)))];
            tensor<fp16, [1536]> layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11688320)))];
            tensor<fp16, [1, 1536, 1, 1500]> input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")];
            string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1536, 1, 1500]> input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")];
            string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)];
            tensor<fp16, [384, 1536, 1, 1]> layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor<fp16, [384, 1536, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11691456)))];
            tensor<fp16, [384]> layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12871168)))];
            tensor<fp16, [1, 384, 1, 1500]> hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")];
            int32 var_1725 = const()[name = string("op_1725"), val = int32(3)];
            int32 var_1736 = const()[name = string("op_1736"), val = int32(1)];
            tensor<int32, [1]> out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_1753_to_fp16 = const()[name = string("op_1753_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_1753_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")];
            tensor<fp16, [384]> obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872000)))];
            tensor<fp16, [384]> obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12872832)))];
            fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")];
            string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> query_strides_0 = const()[name = string("query_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> query_pad_0 = const()[name = string("query_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12873664)))];
            tensor<fp16, [384]> layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13168640)))];
            tensor<fp16, [1, 384, 1, 1500]> query_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_cast_fp16")];
            string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> key_strides_0 = const()[name = string("key_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> key_pad_0 = const()[name = string("key_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13169472)))];
            tensor<fp16, [1, 384, 1, 1500]> key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_cast_fp16")];
            string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> value_strides_0 = const()[name = string("value_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> value_pad_0 = const()[name = string("value_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13464448)))];
            tensor<fp16, [384]> layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13759424)))];
            tensor<fp16, [1, 384, 1, 1500]> value_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_cast_fp16")];
            tensor<int32, [4]> var_1791_begin_0 = const()[name = string("op_1791_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1791_end_0 = const()[name = string("op_1791_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1791_end_mask_0 = const()[name = string("op_1791_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1791_cast_fp16 = slice_by_index(begin = var_1791_begin_0, end = var_1791_end_0, end_mask = var_1791_end_mask_0, x = query_cast_fp16)[name = string("op_1791_cast_fp16")];
            tensor<int32, [4]> var_1795_begin_0 = const()[name = string("op_1795_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_1795_end_0 = const()[name = string("op_1795_end_0"), val = tensor<int32, [4]>([1, 128, 1, 1500])];
            tensor<bool, [4]> var_1795_end_mask_0 = const()[name = string("op_1795_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1795_cast_fp16 = slice_by_index(begin = var_1795_begin_0, end = var_1795_end_0, end_mask = var_1795_end_mask_0, x = query_cast_fp16)[name = string("op_1795_cast_fp16")];
            tensor<int32, [4]> var_1799_begin_0 = const()[name = string("op_1799_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_1799_end_0 = const()[name = string("op_1799_end_0"), val = tensor<int32, [4]>([1, 192, 1, 1500])];
            tensor<bool, [4]> var_1799_end_mask_0 = const()[name = string("op_1799_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1799_cast_fp16 = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = query_cast_fp16)[name = string("op_1799_cast_fp16")];
            tensor<int32, [4]> var_1803_begin_0 = const()[name = string("op_1803_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_1803_end_0 = const()[name = string("op_1803_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1500])];
            tensor<bool, [4]> var_1803_end_mask_0 = const()[name = string("op_1803_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16 = slice_by_index(begin = var_1803_begin_0, end = var_1803_end_0, end_mask = var_1803_end_mask_0, x = query_cast_fp16)[name = string("op_1803_cast_fp16")];
            tensor<int32, [4]> var_1807_begin_0 = const()[name = string("op_1807_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_1807_end_0 = const()[name = string("op_1807_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1500])];
            tensor<bool, [4]> var_1807_end_mask_0 = const()[name = string("op_1807_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1807_cast_fp16 = slice_by_index(begin = var_1807_begin_0, end = var_1807_end_0, end_mask = var_1807_end_mask_0, x = query_cast_fp16)[name = string("op_1807_cast_fp16")];
            tensor<int32, [4]> var_1811_begin_0 = const()[name = string("op_1811_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_1811_end_0 = const()[name = string("op_1811_end_0"), val = tensor<int32, [4]>([1, 384, 1, 1500])];
            tensor<bool, [4]> var_1811_end_mask_0 = const()[name = string("op_1811_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_1811_cast_fp16 = slice_by_index(begin = var_1811_begin_0, end = var_1811_end_0, end_mask = var_1811_end_mask_0, x = query_cast_fp16)[name = string("op_1811_cast_fp16")];
            tensor<int32, [4]> var_1820_begin_0 = const()[name = string("op_1820_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1820_end_0 = const()[name = string("op_1820_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1820_end_mask_0 = const()[name = string("op_1820_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1820_cast_fp16 = slice_by_index(begin = var_1820_begin_0, end = var_1820_end_0, end_mask = var_1820_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1820_cast_fp16")];
            tensor<int32, [4]> var_1827_begin_0 = const()[name = string("op_1827_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1827_end_0 = const()[name = string("op_1827_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1827_end_mask_0 = const()[name = string("op_1827_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1827_cast_fp16 = slice_by_index(begin = var_1827_begin_0, end = var_1827_end_0, end_mask = var_1827_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1827_cast_fp16")];
            tensor<int32, [4]> var_1834_begin_0 = const()[name = string("op_1834_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1834_end_0 = const()[name = string("op_1834_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1834_end_mask_0 = const()[name = string("op_1834_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1834_cast_fp16 = slice_by_index(begin = var_1834_begin_0, end = var_1834_end_0, end_mask = var_1834_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1834_cast_fp16")];
            tensor<int32, [4]> var_1841_begin_0 = const()[name = string("op_1841_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1841_end_0 = const()[name = string("op_1841_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1841_end_mask_0 = const()[name = string("op_1841_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = var_1791_cast_fp16)[name = string("op_1841_cast_fp16")];
            tensor<int32, [4]> var_1848_begin_0 = const()[name = string("op_1848_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1848_end_0 = const()[name = string("op_1848_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1848_end_mask_0 = const()[name = string("op_1848_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1848_cast_fp16 = slice_by_index(begin = var_1848_begin_0, end = var_1848_end_0, end_mask = var_1848_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1848_cast_fp16")];
            tensor<int32, [4]> var_1855_begin_0 = const()[name = string("op_1855_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1855_end_0 = const()[name = string("op_1855_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1855_end_mask_0 = const()[name = string("op_1855_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1855_cast_fp16 = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1855_cast_fp16")];
            tensor<int32, [4]> var_1862_begin_0 = const()[name = string("op_1862_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1862_end_0 = const()[name = string("op_1862_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1862_end_mask_0 = const()[name = string("op_1862_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1862_cast_fp16 = slice_by_index(begin = var_1862_begin_0, end = var_1862_end_0, end_mask = var_1862_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1862_cast_fp16")];
            tensor<int32, [4]> var_1869_begin_0 = const()[name = string("op_1869_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1869_end_0 = const()[name = string("op_1869_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1869_end_mask_0 = const()[name = string("op_1869_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1869_cast_fp16 = slice_by_index(begin = var_1869_begin_0, end = var_1869_end_0, end_mask = var_1869_end_mask_0, x = var_1795_cast_fp16)[name = string("op_1869_cast_fp16")];
            tensor<int32, [4]> var_1876_begin_0 = const()[name = string("op_1876_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1876_end_0 = const()[name = string("op_1876_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1876_end_mask_0 = const()[name = string("op_1876_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1876_cast_fp16 = slice_by_index(begin = var_1876_begin_0, end = var_1876_end_0, end_mask = var_1876_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1876_cast_fp16")];
            tensor<int32, [4]> var_1883_begin_0 = const()[name = string("op_1883_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1883_end_0 = const()[name = string("op_1883_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1883_end_mask_0 = const()[name = string("op_1883_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1883_cast_fp16 = slice_by_index(begin = var_1883_begin_0, end = var_1883_end_0, end_mask = var_1883_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1883_cast_fp16")];
            tensor<int32, [4]> var_1890_begin_0 = const()[name = string("op_1890_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1890_end_0 = const()[name = string("op_1890_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1890_end_mask_0 = const()[name = string("op_1890_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1890_cast_fp16 = slice_by_index(begin = var_1890_begin_0, end = var_1890_end_0, end_mask = var_1890_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1890_cast_fp16")];
            tensor<int32, [4]> var_1897_begin_0 = const()[name = string("op_1897_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1897_end_0 = const()[name = string("op_1897_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1897_end_mask_0 = const()[name = string("op_1897_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1897_cast_fp16 = slice_by_index(begin = var_1897_begin_0, end = var_1897_end_0, end_mask = var_1897_end_mask_0, x = var_1799_cast_fp16)[name = string("op_1897_cast_fp16")];
            tensor<int32, [4]> var_1904_begin_0 = const()[name = string("op_1904_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1904_end_0 = const()[name = string("op_1904_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1904_end_mask_0 = const()[name = string("op_1904_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1904_cast_fp16 = slice_by_index(begin = var_1904_begin_0, end = var_1904_end_0, end_mask = var_1904_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1904_cast_fp16")];
            tensor<int32, [4]> var_1911_begin_0 = const()[name = string("op_1911_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1911_end_0 = const()[name = string("op_1911_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1911_end_mask_0 = const()[name = string("op_1911_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1911_cast_fp16 = slice_by_index(begin = var_1911_begin_0, end = var_1911_end_0, end_mask = var_1911_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1911_cast_fp16")];
            tensor<int32, [4]> var_1918_begin_0 = const()[name = string("op_1918_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1918_end_0 = const()[name = string("op_1918_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1918_end_mask_0 = const()[name = string("op_1918_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1918_cast_fp16 = slice_by_index(begin = var_1918_begin_0, end = var_1918_end_0, end_mask = var_1918_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1918_cast_fp16")];
            tensor<int32, [4]> var_1925_begin_0 = const()[name = string("op_1925_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1925_end_0 = const()[name = string("op_1925_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1925_end_mask_0 = const()[name = string("op_1925_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1925_cast_fp16 = slice_by_index(begin = var_1925_begin_0, end = var_1925_end_0, end_mask = var_1925_end_mask_0, x = var_1803_cast_fp16)[name = string("op_1925_cast_fp16")];
            tensor<int32, [4]> var_1932_begin_0 = const()[name = string("op_1932_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1932_end_0 = const()[name = string("op_1932_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1932_end_mask_0 = const()[name = string("op_1932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1932_cast_fp16 = slice_by_index(begin = var_1932_begin_0, end = var_1932_end_0, end_mask = var_1932_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1932_cast_fp16")];
            tensor<int32, [4]> var_1939_begin_0 = const()[name = string("op_1939_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1939_end_0 = const()[name = string("op_1939_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1939_end_mask_0 = const()[name = string("op_1939_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1939_cast_fp16 = slice_by_index(begin = var_1939_begin_0, end = var_1939_end_0, end_mask = var_1939_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1939_cast_fp16")];
            tensor<int32, [4]> var_1946_begin_0 = const()[name = string("op_1946_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1946_end_0 = const()[name = string("op_1946_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1946_end_mask_0 = const()[name = string("op_1946_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1946_cast_fp16 = slice_by_index(begin = var_1946_begin_0, end = var_1946_end_0, end_mask = var_1946_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1946_cast_fp16")];
            tensor<int32, [4]> var_1953_begin_0 = const()[name = string("op_1953_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1953_end_0 = const()[name = string("op_1953_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1953_end_mask_0 = const()[name = string("op_1953_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1953_cast_fp16 = slice_by_index(begin = var_1953_begin_0, end = var_1953_end_0, end_mask = var_1953_end_mask_0, x = var_1807_cast_fp16)[name = string("op_1953_cast_fp16")];
            tensor<int32, [4]> var_1960_begin_0 = const()[name = string("op_1960_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1960_end_0 = const()[name = string("op_1960_end_0"), val = tensor<int32, [4]>([1, 64, 1, 375])];
            tensor<bool, [4]> var_1960_end_mask_0 = const()[name = string("op_1960_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1960_cast_fp16 = slice_by_index(begin = var_1960_begin_0, end = var_1960_end_0, end_mask = var_1960_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1960_cast_fp16")];
            tensor<int32, [4]> var_1967_begin_0 = const()[name = string("op_1967_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 375])];
            tensor<int32, [4]> var_1967_end_0 = const()[name = string("op_1967_end_0"), val = tensor<int32, [4]>([1, 64, 1, 750])];
            tensor<bool, [4]> var_1967_end_mask_0 = const()[name = string("op_1967_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1967_cast_fp16 = slice_by_index(begin = var_1967_begin_0, end = var_1967_end_0, end_mask = var_1967_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1967_cast_fp16")];
            tensor<int32, [4]> var_1974_begin_0 = const()[name = string("op_1974_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 750])];
            tensor<int32, [4]> var_1974_end_0 = const()[name = string("op_1974_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1125])];
            tensor<bool, [4]> var_1974_end_mask_0 = const()[name = string("op_1974_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1974_cast_fp16 = slice_by_index(begin = var_1974_begin_0, end = var_1974_end_0, end_mask = var_1974_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1974_cast_fp16")];
            tensor<int32, [4]> var_1981_begin_0 = const()[name = string("op_1981_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 1125])];
            tensor<int32, [4]> var_1981_end_0 = const()[name = string("op_1981_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_1981_end_mask_0 = const()[name = string("op_1981_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 64, 1, 375]> var_1981_cast_fp16 = slice_by_index(begin = var_1981_begin_0, end = var_1981_end_0, end_mask = var_1981_end_mask_0, x = var_1811_cast_fp16)[name = string("op_1981_cast_fp16")];
            tensor<int32, [4]> k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])];
            tensor<int32, [4]> var_1986_begin_0 = const()[name = string("op_1986_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_1986_end_0 = const()[name = string("op_1986_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 64])];
            tensor<bool, [4]> var_1986_end_mask_0 = const()[name = string("op_1986_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 384]> k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [1, 1500, 1, 64]> var_1986_cast_fp16 = slice_by_index(begin = var_1986_begin_0, end = var_1986_end_0, end_mask = var_1986_end_mask_0, x = k_7_cast_fp16)[name = string("op_1986_cast_fp16")];
            tensor<int32, [4]> var_1990_begin_0 = const()[name = string("op_1990_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 64])];
            tensor<int32, [4]> var_1990_end_0 = const()[name = string("op_1990_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 128])];
            tensor<bool, [4]> var_1990_end_mask_0 = const()[name = string("op_1990_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_1990_cast_fp16 = slice_by_index(begin = var_1990_begin_0, end = var_1990_end_0, end_mask = var_1990_end_mask_0, x = k_7_cast_fp16)[name = string("op_1990_cast_fp16")];
            tensor<int32, [4]> var_1994_begin_0 = const()[name = string("op_1994_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 128])];
            tensor<int32, [4]> var_1994_end_0 = const()[name = string("op_1994_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 192])];
            tensor<bool, [4]> var_1994_end_mask_0 = const()[name = string("op_1994_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_1994_cast_fp16 = slice_by_index(begin = var_1994_begin_0, end = var_1994_end_0, end_mask = var_1994_end_mask_0, x = k_7_cast_fp16)[name = string("op_1994_cast_fp16")];
            tensor<int32, [4]> var_1998_begin_0 = const()[name = string("op_1998_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 192])];
            tensor<int32, [4]> var_1998_end_0 = const()[name = string("op_1998_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 256])];
            tensor<bool, [4]> var_1998_end_mask_0 = const()[name = string("op_1998_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_1998_cast_fp16 = slice_by_index(begin = var_1998_begin_0, end = var_1998_end_0, end_mask = var_1998_end_mask_0, x = k_7_cast_fp16)[name = string("op_1998_cast_fp16")];
            tensor<int32, [4]> var_2002_begin_0 = const()[name = string("op_2002_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 256])];
            tensor<int32, [4]> var_2002_end_0 = const()[name = string("op_2002_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 320])];
            tensor<bool, [4]> var_2002_end_mask_0 = const()[name = string("op_2002_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_2002_cast_fp16 = slice_by_index(begin = var_2002_begin_0, end = var_2002_end_0, end_mask = var_2002_end_mask_0, x = k_7_cast_fp16)[name = string("op_2002_cast_fp16")];
            tensor<int32, [4]> var_2006_begin_0 = const()[name = string("op_2006_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 320])];
            tensor<int32, [4]> var_2006_end_0 = const()[name = string("op_2006_end_0"), val = tensor<int32, [4]>([1, 1500, 1, 384])];
            tensor<bool, [4]> var_2006_end_mask_0 = const()[name = string("op_2006_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 1500, 1, 64]> var_2006_cast_fp16 = slice_by_index(begin = var_2006_begin_0, end = var_2006_end_0, end_mask = var_2006_end_mask_0, x = k_7_cast_fp16)[name = string("op_2006_cast_fp16")];
            tensor<int32, [4]> var_2008_begin_0 = const()[name = string("op_2008_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_2008_end_0 = const()[name = string("op_2008_end_0"), val = tensor<int32, [4]>([1, 64, 1, 1500])];
            tensor<bool, [4]> var_2008_end_mask_0 = const()[name = string("op_2008_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16 = slice_by_index(begin = var_2008_begin_0, end = var_2008_end_0, end_mask = var_2008_end_mask_0, x = value_cast_fp16)[name = string("op_2008_cast_fp16")];
            tensor<int32, [4]> var_2012_begin_0 = const()[name = string("op_2012_begin_0"), val = tensor<int32, [4]>([0, 64, 0, 0])];
            tensor<int32, [4]> var_2012_end_0 = const()[name = string("op_2012_end_0"), val = tensor<int32, [4]>([1, 128, 1, 1500])];
            tensor<bool, [4]> var_2012_end_mask_0 = const()[name = string("op_2012_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_2012_cast_fp16 = slice_by_index(begin = var_2012_begin_0, end = var_2012_end_0, end_mask = var_2012_end_mask_0, x = value_cast_fp16)[name = string("op_2012_cast_fp16")];
            tensor<int32, [4]> var_2016_begin_0 = const()[name = string("op_2016_begin_0"), val = tensor<int32, [4]>([0, 128, 0, 0])];
            tensor<int32, [4]> var_2016_end_0 = const()[name = string("op_2016_end_0"), val = tensor<int32, [4]>([1, 192, 1, 1500])];
            tensor<bool, [4]> var_2016_end_mask_0 = const()[name = string("op_2016_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_2016_cast_fp16 = slice_by_index(begin = var_2016_begin_0, end = var_2016_end_0, end_mask = var_2016_end_mask_0, x = value_cast_fp16)[name = string("op_2016_cast_fp16")];
            tensor<int32, [4]> var_2020_begin_0 = const()[name = string("op_2020_begin_0"), val = tensor<int32, [4]>([0, 192, 0, 0])];
            tensor<int32, [4]> var_2020_end_0 = const()[name = string("op_2020_end_0"), val = tensor<int32, [4]>([1, 256, 1, 1500])];
            tensor<bool, [4]> var_2020_end_mask_0 = const()[name = string("op_2020_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_2020_cast_fp16 = slice_by_index(begin = var_2020_begin_0, end = var_2020_end_0, end_mask = var_2020_end_mask_0, x = value_cast_fp16)[name = string("op_2020_cast_fp16")];
            tensor<int32, [4]> var_2024_begin_0 = const()[name = string("op_2024_begin_0"), val = tensor<int32, [4]>([0, 256, 0, 0])];
            tensor<int32, [4]> var_2024_end_0 = const()[name = string("op_2024_end_0"), val = tensor<int32, [4]>([1, 320, 1, 1500])];
            tensor<bool, [4]> var_2024_end_mask_0 = const()[name = string("op_2024_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_2024_cast_fp16 = slice_by_index(begin = var_2024_begin_0, end = var_2024_end_0, end_mask = var_2024_end_mask_0, x = value_cast_fp16)[name = string("op_2024_cast_fp16")];
            tensor<int32, [4]> var_2028_begin_0 = const()[name = string("op_2028_begin_0"), val = tensor<int32, [4]>([0, 320, 0, 0])];
            tensor<int32, [4]> var_2028_end_0 = const()[name = string("op_2028_end_0"), val = tensor<int32, [4]>([1, 384, 1, 1500])];
            tensor<bool, [4]> var_2028_end_mask_0 = const()[name = string("op_2028_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 64, 1, 1500]> var_2028_cast_fp16 = slice_by_index(begin = var_2028_begin_0, end = var_2028_end_0, end_mask = var_2028_end_mask_0, x = value_cast_fp16)[name = string("op_2028_cast_fp16")];
            string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1986_cast_fp16, var_1820_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")];
            string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1986_cast_fp16, var_1827_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")];
            string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1986_cast_fp16, var_1834_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")];
            string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1986_cast_fp16, var_1841_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")];
            string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1990_cast_fp16, var_1848_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")];
            string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1990_cast_fp16, var_1855_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")];
            string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1990_cast_fp16, var_1862_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")];
            string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1990_cast_fp16, var_1869_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")];
            string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1994_cast_fp16, var_1876_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")];
            string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1994_cast_fp16, var_1883_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")];
            string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1994_cast_fp16, var_1890_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")];
            string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1994_cast_fp16, var_1897_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")];
            string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1998_cast_fp16, var_1904_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")];
            string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1998_cast_fp16, var_1911_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")];
            string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1998_cast_fp16, var_1918_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")];
            string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1998_cast_fp16, var_1925_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")];
            string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_2002_cast_fp16, var_1932_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")];
            string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_2002_cast_fp16, var_1939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")];
            string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_2002_cast_fp16, var_1946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")];
            string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_2002_cast_fp16, var_1953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")];
            string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_2006_cast_fp16, var_1960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")];
            string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_2006_cast_fp16, var_1967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")];
            string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_2006_cast_fp16, var_1974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")];
            string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")];
            tensor<fp16, [1, 1500, 1, 375]> _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_2006_cast_fp16, var_1981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")];
            fp16 var_2079_to_fp16 = const()[name = string("op_2079_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_2079_to_fp16)[name = string("aw_chunk_145_cast_fp16")];
            fp16 var_2081_to_fp16 = const()[name = string("op_2081_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_2081_to_fp16)[name = string("aw_chunk_147_cast_fp16")];
            fp16 var_2083_to_fp16 = const()[name = string("op_2083_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_2083_to_fp16)[name = string("aw_chunk_149_cast_fp16")];
            fp16 var_2085_to_fp16 = const()[name = string("op_2085_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_2085_to_fp16)[name = string("aw_chunk_151_cast_fp16")];
            fp16 var_2087_to_fp16 = const()[name = string("op_2087_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_2087_to_fp16)[name = string("aw_chunk_153_cast_fp16")];
            fp16 var_2089_to_fp16 = const()[name = string("op_2089_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_2089_to_fp16)[name = string("aw_chunk_155_cast_fp16")];
            fp16 var_2091_to_fp16 = const()[name = string("op_2091_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_2091_to_fp16)[name = string("aw_chunk_157_cast_fp16")];
            fp16 var_2093_to_fp16 = const()[name = string("op_2093_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_2093_to_fp16)[name = string("aw_chunk_159_cast_fp16")];
            fp16 var_2095_to_fp16 = const()[name = string("op_2095_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_2095_to_fp16)[name = string("aw_chunk_161_cast_fp16")];
            fp16 var_2097_to_fp16 = const()[name = string("op_2097_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_2097_to_fp16)[name = string("aw_chunk_163_cast_fp16")];
            fp16 var_2099_to_fp16 = const()[name = string("op_2099_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_2099_to_fp16)[name = string("aw_chunk_165_cast_fp16")];
            fp16 var_2101_to_fp16 = const()[name = string("op_2101_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_2101_to_fp16)[name = string("aw_chunk_167_cast_fp16")];
            fp16 var_2103_to_fp16 = const()[name = string("op_2103_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_2103_to_fp16)[name = string("aw_chunk_169_cast_fp16")];
            fp16 var_2105_to_fp16 = const()[name = string("op_2105_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_2105_to_fp16)[name = string("aw_chunk_171_cast_fp16")];
            fp16 var_2107_to_fp16 = const()[name = string("op_2107_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_2107_to_fp16)[name = string("aw_chunk_173_cast_fp16")];
            fp16 var_2109_to_fp16 = const()[name = string("op_2109_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_2109_to_fp16)[name = string("aw_chunk_175_cast_fp16")];
            fp16 var_2111_to_fp16 = const()[name = string("op_2111_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_2111_to_fp16)[name = string("aw_chunk_177_cast_fp16")];
            fp16 var_2113_to_fp16 = const()[name = string("op_2113_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_2113_to_fp16)[name = string("aw_chunk_179_cast_fp16")];
            fp16 var_2115_to_fp16 = const()[name = string("op_2115_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_2115_to_fp16)[name = string("aw_chunk_181_cast_fp16")];
            fp16 var_2117_to_fp16 = const()[name = string("op_2117_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_2117_to_fp16)[name = string("aw_chunk_183_cast_fp16")];
            fp16 var_2119_to_fp16 = const()[name = string("op_2119_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_2119_to_fp16)[name = string("aw_chunk_185_cast_fp16")];
            fp16 var_2121_to_fp16 = const()[name = string("op_2121_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_2121_to_fp16)[name = string("aw_chunk_187_cast_fp16")];
            fp16 var_2123_to_fp16 = const()[name = string("op_2123_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_2123_to_fp16)[name = string("aw_chunk_189_cast_fp16")];
            fp16 var_2125_to_fp16 = const()[name = string("op_2125_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 1500, 1, 375]> aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_2125_to_fp16)[name = string("aw_chunk_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2127_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_145_cast_fp16)[name = string("op_2127_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2128_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_147_cast_fp16)[name = string("op_2128_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2129_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_149_cast_fp16)[name = string("op_2129_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2130_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_151_cast_fp16)[name = string("op_2130_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2131_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_153_cast_fp16)[name = string("op_2131_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2132_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_155_cast_fp16)[name = string("op_2132_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2133_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_157_cast_fp16)[name = string("op_2133_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2134_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_159_cast_fp16)[name = string("op_2134_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2135_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_161_cast_fp16)[name = string("op_2135_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2136_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_163_cast_fp16)[name = string("op_2136_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2137_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_165_cast_fp16)[name = string("op_2137_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2138_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_167_cast_fp16)[name = string("op_2138_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2139_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_169_cast_fp16)[name = string("op_2139_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2140_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_171_cast_fp16)[name = string("op_2140_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2141_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_173_cast_fp16)[name = string("op_2141_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2142_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_175_cast_fp16)[name = string("op_2142_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2143_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_177_cast_fp16)[name = string("op_2143_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2144_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_179_cast_fp16)[name = string("op_2144_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2145_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_181_cast_fp16)[name = string("op_2145_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2146_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_183_cast_fp16)[name = string("op_2146_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2147_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_185_cast_fp16)[name = string("op_2147_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2148_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_187_cast_fp16)[name = string("op_2148_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2149_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_189_cast_fp16)[name = string("op_2149_cast_fp16")];
            tensor<fp16, [1, 1500, 1, 375]> var_2150_cast_fp16 = softmax(axis = var_1736, x = aw_chunk_cast_fp16)[name = string("op_2150_cast_fp16")];
            string var_2152_equation_0 = const()[name = string("op_2152_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2152_cast_fp16 = einsum(equation = var_2152_equation_0, values = (var_2008_cast_fp16, var_2127_cast_fp16))[name = string("op_2152_cast_fp16")];
            string var_2154_equation_0 = const()[name = string("op_2154_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2154_cast_fp16 = einsum(equation = var_2154_equation_0, values = (var_2008_cast_fp16, var_2128_cast_fp16))[name = string("op_2154_cast_fp16")];
            string var_2156_equation_0 = const()[name = string("op_2156_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2156_cast_fp16 = einsum(equation = var_2156_equation_0, values = (var_2008_cast_fp16, var_2129_cast_fp16))[name = string("op_2156_cast_fp16")];
            string var_2158_equation_0 = const()[name = string("op_2158_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2158_cast_fp16 = einsum(equation = var_2158_equation_0, values = (var_2008_cast_fp16, var_2130_cast_fp16))[name = string("op_2158_cast_fp16")];
            string var_2160_equation_0 = const()[name = string("op_2160_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2160_cast_fp16 = einsum(equation = var_2160_equation_0, values = (var_2012_cast_fp16, var_2131_cast_fp16))[name = string("op_2160_cast_fp16")];
            string var_2162_equation_0 = const()[name = string("op_2162_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2162_cast_fp16 = einsum(equation = var_2162_equation_0, values = (var_2012_cast_fp16, var_2132_cast_fp16))[name = string("op_2162_cast_fp16")];
            string var_2164_equation_0 = const()[name = string("op_2164_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2164_cast_fp16 = einsum(equation = var_2164_equation_0, values = (var_2012_cast_fp16, var_2133_cast_fp16))[name = string("op_2164_cast_fp16")];
            string var_2166_equation_0 = const()[name = string("op_2166_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2166_cast_fp16 = einsum(equation = var_2166_equation_0, values = (var_2012_cast_fp16, var_2134_cast_fp16))[name = string("op_2166_cast_fp16")];
            string var_2168_equation_0 = const()[name = string("op_2168_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2168_cast_fp16 = einsum(equation = var_2168_equation_0, values = (var_2016_cast_fp16, var_2135_cast_fp16))[name = string("op_2168_cast_fp16")];
            string var_2170_equation_0 = const()[name = string("op_2170_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2170_cast_fp16 = einsum(equation = var_2170_equation_0, values = (var_2016_cast_fp16, var_2136_cast_fp16))[name = string("op_2170_cast_fp16")];
            string var_2172_equation_0 = const()[name = string("op_2172_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2172_cast_fp16 = einsum(equation = var_2172_equation_0, values = (var_2016_cast_fp16, var_2137_cast_fp16))[name = string("op_2172_cast_fp16")];
            string var_2174_equation_0 = const()[name = string("op_2174_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2174_cast_fp16 = einsum(equation = var_2174_equation_0, values = (var_2016_cast_fp16, var_2138_cast_fp16))[name = string("op_2174_cast_fp16")];
            string var_2176_equation_0 = const()[name = string("op_2176_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2176_cast_fp16 = einsum(equation = var_2176_equation_0, values = (var_2020_cast_fp16, var_2139_cast_fp16))[name = string("op_2176_cast_fp16")];
            string var_2178_equation_0 = const()[name = string("op_2178_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2178_cast_fp16 = einsum(equation = var_2178_equation_0, values = (var_2020_cast_fp16, var_2140_cast_fp16))[name = string("op_2178_cast_fp16")];
            string var_2180_equation_0 = const()[name = string("op_2180_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2180_cast_fp16 = einsum(equation = var_2180_equation_0, values = (var_2020_cast_fp16, var_2141_cast_fp16))[name = string("op_2180_cast_fp16")];
            string var_2182_equation_0 = const()[name = string("op_2182_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2182_cast_fp16 = einsum(equation = var_2182_equation_0, values = (var_2020_cast_fp16, var_2142_cast_fp16))[name = string("op_2182_cast_fp16")];
            string var_2184_equation_0 = const()[name = string("op_2184_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2184_cast_fp16 = einsum(equation = var_2184_equation_0, values = (var_2024_cast_fp16, var_2143_cast_fp16))[name = string("op_2184_cast_fp16")];
            string var_2186_equation_0 = const()[name = string("op_2186_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2186_cast_fp16 = einsum(equation = var_2186_equation_0, values = (var_2024_cast_fp16, var_2144_cast_fp16))[name = string("op_2186_cast_fp16")];
            string var_2188_equation_0 = const()[name = string("op_2188_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2188_cast_fp16 = einsum(equation = var_2188_equation_0, values = (var_2024_cast_fp16, var_2145_cast_fp16))[name = string("op_2188_cast_fp16")];
            string var_2190_equation_0 = const()[name = string("op_2190_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2190_cast_fp16 = einsum(equation = var_2190_equation_0, values = (var_2024_cast_fp16, var_2146_cast_fp16))[name = string("op_2190_cast_fp16")];
            string var_2192_equation_0 = const()[name = string("op_2192_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2192_cast_fp16 = einsum(equation = var_2192_equation_0, values = (var_2028_cast_fp16, var_2147_cast_fp16))[name = string("op_2192_cast_fp16")];
            string var_2194_equation_0 = const()[name = string("op_2194_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2194_cast_fp16 = einsum(equation = var_2194_equation_0, values = (var_2028_cast_fp16, var_2148_cast_fp16))[name = string("op_2194_cast_fp16")];
            string var_2196_equation_0 = const()[name = string("op_2196_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2196_cast_fp16 = einsum(equation = var_2196_equation_0, values = (var_2028_cast_fp16, var_2149_cast_fp16))[name = string("op_2196_cast_fp16")];
            string var_2198_equation_0 = const()[name = string("op_2198_equation_0"), val = string("bchk,bkhq->bchq")];
            tensor<fp16, [1, 64, 1, 375]> var_2198_cast_fp16 = einsum(equation = var_2198_equation_0, values = (var_2028_cast_fp16, var_2150_cast_fp16))[name = string("op_2198_cast_fp16")];
            bool var_2200_interleave_0 = const()[name = string("op_2200_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_2200_cast_fp16 = concat(axis = var_1725, interleave = var_2200_interleave_0, values = (var_2152_cast_fp16, var_2154_cast_fp16, var_2156_cast_fp16, var_2158_cast_fp16))[name = string("op_2200_cast_fp16")];
            bool var_2202_interleave_0 = const()[name = string("op_2202_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_2202_cast_fp16 = concat(axis = var_1725, interleave = var_2202_interleave_0, values = (var_2160_cast_fp16, var_2162_cast_fp16, var_2164_cast_fp16, var_2166_cast_fp16))[name = string("op_2202_cast_fp16")];
            bool var_2204_interleave_0 = const()[name = string("op_2204_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_2204_cast_fp16 = concat(axis = var_1725, interleave = var_2204_interleave_0, values = (var_2168_cast_fp16, var_2170_cast_fp16, var_2172_cast_fp16, var_2174_cast_fp16))[name = string("op_2204_cast_fp16")];
            bool var_2206_interleave_0 = const()[name = string("op_2206_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_2206_cast_fp16 = concat(axis = var_1725, interleave = var_2206_interleave_0, values = (var_2176_cast_fp16, var_2178_cast_fp16, var_2180_cast_fp16, var_2182_cast_fp16))[name = string("op_2206_cast_fp16")];
            bool var_2208_interleave_0 = const()[name = string("op_2208_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_2208_cast_fp16 = concat(axis = var_1725, interleave = var_2208_interleave_0, values = (var_2184_cast_fp16, var_2186_cast_fp16, var_2188_cast_fp16, var_2190_cast_fp16))[name = string("op_2208_cast_fp16")];
            bool var_2210_interleave_0 = const()[name = string("op_2210_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 64, 1, 1500]> var_2210_cast_fp16 = concat(axis = var_1725, interleave = var_2210_interleave_0, values = (var_2192_cast_fp16, var_2194_cast_fp16, var_2196_cast_fp16, var_2198_cast_fp16))[name = string("op_2210_cast_fp16")];
            bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 384, 1, 1500]> input_25_cast_fp16 = concat(axis = var_1736, interleave = input_25_interleave_0, values = (var_2200_cast_fp16, var_2202_cast_fp16, var_2204_cast_fp16, var_2206_cast_fp16, var_2208_cast_fp16, var_2210_cast_fp16))[name = string("input_25_cast_fp16")];
            string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13760256)))];
            tensor<fp16, [384]> layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14055232)))];
            tensor<fp16, [1, 384, 1, 1500]> obj_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_cast_fp16)[name = string("inputs_15_cast_fp16")];
            tensor<int32, [1]> out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2229_to_fp16 = const()[name = string("op_2229_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_2229_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")];
            tensor<fp16, [384]> input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056064)))];
            tensor<fp16, [384]> input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14056896)))];
            fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")];
            string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)];
            tensor<fp16, [1536, 384, 1, 1]> layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor<fp16, [1536, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14057728)))];
            tensor<fp16, [1536]> layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15237440)))];
            tensor<fp16, [1, 1536, 1, 1500]> input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")];
            string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")];
            tensor<fp16, [1, 1536, 1, 1500]> input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")];
            string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)];
            tensor<fp16, [384, 1536, 1, 1]> layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor<fp16, [384, 1536, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15240576)))];
            tensor<fp16, [384]> layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16420288)))];
            tensor<fp16, [1, 384, 1, 1500]> hidden_states_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_cast_fp16")];
            tensor<fp16, [1, 384, 1, 1500]> inputs_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")];
            tensor<int32, [1]> out_axes_0 = const()[name = string("out_axes_0"), val = tensor<int32, [1]>([1])];
            fp16 var_2267_to_fp16 = const()[name = string("op_2267_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_2267_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")];
            tensor<fp16, [384]> encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16421120)))];
            tensor<fp16, [384]> encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16421952)))];
            fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 384, 1, 1500]> encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")];
            string var_2291_pad_type_0 = const()[name = string("op_2291_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2291_strides_0 = const()[name = string("op_2291_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2291_pad_0 = const()[name = string("op_2291_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2291_dilations_0 = const()[name = string("op_2291_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2291_groups_0 = const()[name = string("op_2291_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16422784)))];
            tensor<fp16, [1, 384, 1, 1500]> var_2291_cast_fp16 = conv(dilations = var_2291_dilations_0, groups = var_2291_groups_0, pad = var_2291_pad_0, pad_type = var_2291_pad_type_0, strides = var_2291_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2291_cast_fp16")];
            string var_2298_pad_type_0 = const()[name = string("op_2298_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2298_strides_0 = const()[name = string("op_2298_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2298_pad_0 = const()[name = string("op_2298_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2298_dilations_0 = const()[name = string("op_2298_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2298_groups_0 = const()[name = string("op_2298_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16717760)))];
            tensor<fp16, [384]> decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17012736)))];
            tensor<fp16, [1, 384, 1, 1500]> var_2298_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_2298_dilations_0, groups = var_2298_groups_0, pad = var_2298_pad_0, pad_type = var_2298_pad_type_0, strides = var_2298_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2298_cast_fp16")];
            string var_2316_pad_type_0 = const()[name = string("op_2316_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2316_strides_0 = const()[name = string("op_2316_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2316_pad_0 = const()[name = string("op_2316_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2316_dilations_0 = const()[name = string("op_2316_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2316_groups_0 = const()[name = string("op_2316_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17013568)))];
            tensor<fp16, [1, 384, 1, 1500]> var_2316_cast_fp16 = conv(dilations = var_2316_dilations_0, groups = var_2316_groups_0, pad = var_2316_pad_0, pad_type = var_2316_pad_type_0, strides = var_2316_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2316_cast_fp16")];
            string var_2323_pad_type_0 = const()[name = string("op_2323_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2323_strides_0 = const()[name = string("op_2323_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2323_pad_0 = const()[name = string("op_2323_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2323_dilations_0 = const()[name = string("op_2323_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2323_groups_0 = const()[name = string("op_2323_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17308544)))];
            tensor<fp16, [384]> decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17603520)))];
            tensor<fp16, [1, 384, 1, 1500]> var_2323_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_2323_dilations_0, groups = var_2323_groups_0, pad = var_2323_pad_0, pad_type = var_2323_pad_type_0, strides = var_2323_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2323_cast_fp16")];
            string var_2341_pad_type_0 = const()[name = string("op_2341_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2341_strides_0 = const()[name = string("op_2341_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2341_pad_0 = const()[name = string("op_2341_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2341_dilations_0 = const()[name = string("op_2341_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2341_groups_0 = const()[name = string("op_2341_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17604352)))];
            tensor<fp16, [1, 384, 1, 1500]> var_2341_cast_fp16 = conv(dilations = var_2341_dilations_0, groups = var_2341_groups_0, pad = var_2341_pad_0, pad_type = var_2341_pad_type_0, strides = var_2341_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2341_cast_fp16")];
            string var_2348_pad_type_0 = const()[name = string("op_2348_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_2348_strides_0 = const()[name = string("op_2348_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_2348_pad_0 = const()[name = string("op_2348_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_2348_dilations_0 = const()[name = string("op_2348_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_2348_groups_0 = const()[name = string("op_2348_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17899328)))];
            tensor<fp16, [384]> decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18194304)))];
            tensor<fp16, [1, 384, 1, 1500]> var_2348_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_2348_dilations_0, groups = var_2348_groups_0, pad = var_2348_pad_0, pad_type = var_2348_pad_type_0, strides = var_2348_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_2348_cast_fp16")];
            string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> k_strides_0 = const()[name = string("k_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> k_pad_0 = const()[name = string("k_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18195136)))];
            tensor<fp16, [1, 384, 1, 1500]> k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")];
            string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> v_strides_0 = const()[name = string("v_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> v_pad_0 = const()[name = string("v_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)];
            tensor<fp16, [384, 384, 1, 1]> decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [384, 384, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18490112)))];
            tensor<fp16, [384]> decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18785088)))];
            tensor<fp16, [1, 384, 1, 1500]> v_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")];
            int32 var_2378 = const()[name = string("op_2378"), val = int32(0)];
            bool input_35_interleave_0 = const()[name = string("input_35_interleave_0"), val = bool(false)];
            tensor<fp16, [4, 384, 1, 1500]> input_35_cast_fp16 = concat(axis = var_2378, interleave = input_35_interleave_0, values = (var_2291_cast_fp16, var_2316_cast_fp16, var_2341_cast_fp16, k_cast_fp16))[name = string("input_35_cast_fp16")];
            int32 var_2381 = const()[name = string("op_2381"), val = int32(0)];
            bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)];
            tensor<fp16, [4, 384, 1, 1500]> input_cast_fp16 = concat(axis = var_2381, interleave = input_interleave_0, values = (var_2298_cast_fp16, var_2323_cast_fp16, var_2348_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")];
            tensor<int32, [8]> var_2388_pad_0 = const()[name = string("op_2388_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 36])];
            string var_2388_mode_0 = const()[name = string("op_2388_mode_0"), val = string("constant")];
            fp16 const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [4, 384, 1, 1536]> encoder_attn_key_cache = pad(constant_val = const_5_to_fp16, mode = var_2388_mode_0, pad = var_2388_pad_0, x = input_35_cast_fp16)[name = string("op_2388_cast_fp16")];
            tensor<int32, [8]> var_2394_pad_0 = const()[name = string("op_2394_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 36])];
            string var_2394_mode_0 = const()[name = string("op_2394_mode_0"), val = string("constant")];
            fp16 const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [4, 384, 1, 1536]> encoder_attn_value_cache = pad(constant_val = const_6_to_fp16, mode = var_2394_mode_0, pad = var_2394_pad_0, x = input_cast_fp16)[name = string("op_2394_cast_fp16")];
        } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache);
}