program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}, {"coremltools-component-torch", "2.6.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})]
{
    func main<ios18>(tensor<fp16, [1, 34, 1024]> cond_emb, tensor<int32, [1, 256]> text_token_ids) {
            int32 var_768_axis_0 = const()[name = string("op_768_axis_0"), val = int32(0)];
            int32 var_768_batch_dims_0 = const()[name = string("op_768_batch_dims_0"), val = int32(0)];
            bool var_768_validate_indices_0 = const()[name = string("op_768_validate_indices_0"), val = bool(false)];
            tensor<fp16, [704, 1024]> text_emb_weight_to_fp16 = const()[name = string("text_emb_weight_to_fp16"), val = tensor<fp16, [704, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
            string text_token_ids_to_int16_dtype_0 = const()[name = string("text_token_ids_to_int16_dtype_0"), val = string("int16")];
            tensor<int16, [1, 256]> text_token_ids_to_int16 = cast(dtype = text_token_ids_to_int16_dtype_0, x = text_token_ids)[name = string("cast_431")];
            tensor<fp16, [1, 256, 1024]> var_768_cast_fp16_cast_uint16 = gather(axis = var_768_axis_0, batch_dims = var_768_batch_dims_0, indices = text_token_ids_to_int16, validate_indices = var_768_validate_indices_0, x = text_emb_weight_to_fp16)[name = string("op_768_cast_fp16_cast_uint16")];
            tensor<fp16, [256, 1024]> var_779_to_fp16 = const()[name = string("op_779_to_fp16"), val = tensor<fp16, [256, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1441920)))];
            tensor<fp16, [1, 256, 1024]> text_emb_1_cast_fp16 = add(x = var_768_cast_fp16_cast_uint16, y = var_779_to_fp16)[name = string("text_emb_1_cast_fp16")];
            int32 var_811 = const()[name = string("op_811"), val = int32(1)];
            bool inputs_embeds_interleave_0 = const()[name = string("inputs_embeds_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 1, 1024]> sos_emb_to_fp16 = const()[name = string("sos_emb_to_fp16"), val = tensor<fp16, [1, 1, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1966272)))];
            tensor<fp16, [1, 291, 1024]> inputs_embeds_cast_fp16 = concat(axis = var_811, interleave = inputs_embeds_interleave_0, values = (cond_emb, text_emb_1_cast_fp16, sos_emb_to_fp16))[name = string("inputs_embeds_cast_fp16")];
            fp16 var_1030_promoted_to_fp16 = const()[name = string("op_1030_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_1036_cast_fp16 = pow(x = inputs_embeds_cast_fp16, y = var_1030_promoted_to_fp16)[name = string("op_1036_cast_fp16")];
            tensor<int32, [1]> variance_1_axes_0 = const()[name = string("variance_1_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_1_keep_dims_0 = const()[name = string("variance_1_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_1_cast_fp16 = reduce_mean(axes = variance_1_axes_0, keep_dims = variance_1_keep_dims_0, x = var_1036_cast_fp16)[name = string("variance_1_cast_fp16")];
            fp16 var_1039_to_fp16 = const()[name = string("op_1039_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_1040_cast_fp16 = add(x = variance_1_cast_fp16, y = var_1039_to_fp16)[name = string("op_1040_cast_fp16")];
            fp32 var_1041_epsilon_0 = const()[name = string("op_1041_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_1041_cast_fp16 = rsqrt(epsilon = var_1041_epsilon_0, x = var_1040_cast_fp16)[name = string("op_1041_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_3_cast_fp16 = mul(x = inputs_embeds_cast_fp16, y = var_1041_cast_fp16)[name = string("hidden_states_3_cast_fp16")];
            tensor<fp16, [1024]> layers_0_input_layernorm_weight_to_fp16 = const()[name = string("layers_0_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1968384)))];
            tensor<fp16, [1, 291, 1024]> hidden_1_cast_fp16 = mul(x = layers_0_input_layernorm_weight_to_fp16, y = hidden_states_3_cast_fp16)[name = string("hidden_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1970496)))];
            tensor<fp16, [1024]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4067712)))];
            tensor<fp16, [1, 291, 1024]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = hidden_1_cast_fp16)[name = string("linear_0_cast_fp16")];
            tensor<int32, [4]> var_1065 = const()[name = string("op_1065"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_1_cast_fp16 = reshape(shape = var_1065, x = linear_0_cast_fp16)[name = string("q_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4069824)))];
            tensor<fp16, [1, 291, 1024]> linear_1_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = hidden_1_cast_fp16)[name = string("linear_1_cast_fp16")];
            tensor<int32, [4]> var_1072 = const()[name = string("op_1072"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_1_cast_fp16 = reshape(shape = var_1072, x = linear_1_cast_fp16)[name = string("k_1_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6167040)))];
            tensor<fp16, [1, 291, 1024]> linear_2_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = hidden_1_cast_fp16)[name = string("linear_2_cast_fp16")];
            tensor<int32, [4]> var_1079 = const()[name = string("op_1079"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_1_cast_fp16 = reshape(shape = var_1079, x = linear_2_cast_fp16)[name = string("v_1_cast_fp16")];
            tensor<int32, [4]> q_3_perm_0 = const()[name = string("q_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_3_perm_0 = const()[name = string("v_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 1, 291, 64]> var_1091_to_fp16 = const()[name = string("op_1091_to_fp16"), val = tensor<fp16, [1, 1, 291, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8264256)))];
            tensor<fp16, [1, 16, 291, 64]> q_3_cast_fp16 = transpose(perm = q_3_perm_0, x = q_1_cast_fp16)[name = string("transpose_179")];
            tensor<fp16, [1, 16, 291, 64]> var_1092_cast_fp16 = mul(x = q_3_cast_fp16, y = var_1091_to_fp16)[name = string("op_1092_cast_fp16")];
            tensor<int32, [4]> x1_1_begin_0 = const()[name = string("x1_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_1_end_0 = const()[name = string("x1_1_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_1_end_mask_0 = const()[name = string("x1_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_1_cast_fp16 = slice_by_index(begin = x1_1_begin_0, end = x1_1_end_0, end_mask = x1_1_end_mask_0, x = q_3_cast_fp16)[name = string("x1_1_cast_fp16")];
            tensor<int32, [4]> x2_1_begin_0 = const()[name = string("x2_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_1_end_0 = const()[name = string("x2_1_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_1_end_mask_0 = const()[name = string("x2_1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_1_cast_fp16 = slice_by_index(begin = x2_1_begin_0, end = x2_1_end_0, end_mask = x2_1_end_mask_0, x = q_3_cast_fp16)[name = string("x2_1_cast_fp16")];
            fp16 const_17_promoted_to_fp16 = const()[name = string("const_17_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_1113_cast_fp16 = mul(x = x2_1_cast_fp16, y = const_17_promoted_to_fp16)[name = string("op_1113_cast_fp16")];
            int32 var_1115 = const()[name = string("op_1115"), val = int32(-1)];
            bool var_1116_interleave_0 = const()[name = string("op_1116_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_1116_cast_fp16 = concat(axis = var_1115, interleave = var_1116_interleave_0, values = (var_1113_cast_fp16, x1_1_cast_fp16))[name = string("op_1116_cast_fp16")];
            tensor<fp16, [1, 1, 291, 64]> var_1118_to_fp16 = const()[name = string("op_1118_to_fp16"), val = tensor<fp16, [1, 1, 291, 64]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8301568)))];
            tensor<fp16, [1, 16, 291, 64]> var_1119_cast_fp16 = mul(x = var_1116_cast_fp16, y = var_1118_to_fp16)[name = string("op_1119_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_5_cast_fp16 = add(x = var_1092_cast_fp16, y = var_1119_cast_fp16)[name = string("q_5_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = k_1_cast_fp16)[name = string("transpose_178")];
            tensor<fp16, [1, 16, 291, 64]> var_1124_cast_fp16 = mul(x = k_3_cast_fp16, y = var_1091_to_fp16)[name = string("op_1124_cast_fp16")];
            tensor<int32, [4]> x1_3_begin_0 = const()[name = string("x1_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_3_end_0 = const()[name = string("x1_3_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_3_end_mask_0 = const()[name = string("x1_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_3_cast_fp16 = slice_by_index(begin = x1_3_begin_0, end = x1_3_end_0, end_mask = x1_3_end_mask_0, x = k_3_cast_fp16)[name = string("x1_3_cast_fp16")];
            tensor<int32, [4]> x2_3_begin_0 = const()[name = string("x2_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_3_end_0 = const()[name = string("x2_3_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_3_end_mask_0 = const()[name = string("x2_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_3_cast_fp16 = slice_by_index(begin = x2_3_begin_0, end = x2_3_end_0, end_mask = x2_3_end_mask_0, x = k_3_cast_fp16)[name = string("x2_3_cast_fp16")];
            fp16 const_20_promoted_to_fp16 = const()[name = string("const_20_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_1145_cast_fp16 = mul(x = x2_3_cast_fp16, y = const_20_promoted_to_fp16)[name = string("op_1145_cast_fp16")];
            int32 var_1147 = const()[name = string("op_1147"), val = int32(-1)];
            bool var_1148_interleave_0 = const()[name = string("op_1148_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_1148_cast_fp16 = concat(axis = var_1147, interleave = var_1148_interleave_0, values = (var_1145_cast_fp16, x1_3_cast_fp16))[name = string("op_1148_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_1151_cast_fp16 = mul(x = var_1148_cast_fp16, y = var_1118_to_fp16)[name = string("op_1151_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_5_cast_fp16 = add(x = var_1124_cast_fp16, y = var_1151_cast_fp16)[name = string("k_5_cast_fp16")];
            bool var_1157_transpose_x_1 = const()[name = string("op_1157_transpose_x_1"), val = bool(false)];
            bool var_1157_transpose_y_1 = const()[name = string("op_1157_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_1157_cast_fp16 = matmul(transpose_x = var_1157_transpose_x_1, transpose_y = var_1157_transpose_y_1, x = q_5_cast_fp16, y = k_5_cast_fp16)[name = string("op_1157_cast_fp16")];
            fp16 _inversed_scores_1_y_0_to_fp16 = const()[name = string("_inversed_scores_1_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_1_cast_fp16 = mul(x = var_1157_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = string("_inversed_scores_1_cast_fp16")];
            tensor<fp16, [1, 1, 291, 291]> const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = tensor<fp16, [1, 1, 291, 291]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8338880)))];
            tensor<fp16, [1, 16, 291, 291]> scores_3_cast_fp16 = add(x = _inversed_scores_1_cast_fp16, y = const_21_to_fp16)[name = string("scores_3_cast_fp16")];
            int32 var_1172 = const()[name = string("op_1172"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_1174_cast_fp16 = softmax(axis = var_1172, x = scores_3_cast_fp16)[name = string("op_1174_cast_fp16")];
            bool attn_out_1_transpose_x_0 = const()[name = string("attn_out_1_transpose_x_0"), val = bool(false)];
            bool attn_out_1_transpose_y_0 = const()[name = string("attn_out_1_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_3_cast_fp16 = transpose(perm = v_3_perm_0, x = v_1_cast_fp16)[name = string("transpose_177")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_1_cast_fp16 = matmul(transpose_x = attn_out_1_transpose_x_0, transpose_y = attn_out_1_transpose_y_0, x = var_1174_cast_fp16, y = v_3_cast_fp16)[name = string("attn_out_1_cast_fp16")];
            tensor<int32, [4]> var_1183_perm_0 = const()[name = string("op_1183_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1185 = const()[name = string("op_1185"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_1183_cast_fp16 = transpose(perm = var_1183_perm_0, x = attn_out_1_cast_fp16)[name = string("transpose_176")];
            tensor<fp16, [1, 291, 1024]> input_9_cast_fp16 = reshape(shape = var_1185, x = var_1183_cast_fp16)[name = string("input_9_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8508352)))];
            tensor<fp16, [1, 291, 1024]> linear_3_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("linear_3_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_7_cast_fp16 = add(x = inputs_embeds_cast_fp16, y = linear_3_cast_fp16)[name = string("hidden_states_7_cast_fp16")];
            fp16 var_1195_promoted_to_fp16 = const()[name = string("op_1195_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_1201_cast_fp16 = pow(x = hidden_states_7_cast_fp16, y = var_1195_promoted_to_fp16)[name = string("op_1201_cast_fp16")];
            tensor<int32, [1]> variance_3_axes_0 = const()[name = string("variance_3_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_3_keep_dims_0 = const()[name = string("variance_3_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_3_cast_fp16 = reduce_mean(axes = variance_3_axes_0, keep_dims = variance_3_keep_dims_0, x = var_1201_cast_fp16)[name = string("variance_3_cast_fp16")];
            fp16 var_1204_to_fp16 = const()[name = string("op_1204_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_1205_cast_fp16 = add(x = variance_3_cast_fp16, y = var_1204_to_fp16)[name = string("op_1205_cast_fp16")];
            fp32 var_1206_epsilon_0 = const()[name = string("op_1206_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_1206_cast_fp16 = rsqrt(epsilon = var_1206_epsilon_0, x = var_1205_cast_fp16)[name = string("op_1206_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_11_cast_fp16 = mul(x = hidden_states_7_cast_fp16, y = var_1206_cast_fp16)[name = string("hidden_states_11_cast_fp16")];
            tensor<fp16, [1024]> layers_0_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_0_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10605568)))];
            tensor<fp16, [1, 291, 1024]> input_11_cast_fp16 = mul(x = layers_0_post_attention_layernorm_weight_to_fp16, y = hidden_states_11_cast_fp16)[name = string("input_11_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_0_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_0_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10607680)))];
            tensor<fp16, [4096]> linear_4_bias_0_to_fp16 = const()[name = string("linear_4_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18996352)))];
            tensor<fp16, [1, 291, 4096]> linear_4_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_0_mlp_gate_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_4_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_1219_cast_fp16 = silu(x = linear_4_cast_fp16)[name = string("op_1219_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_0_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_0_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(19004608)))];
            tensor<fp16, [1, 291, 4096]> linear_5_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_0_mlp_up_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_5_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_15_cast_fp16 = mul(x = var_1219_cast_fp16, y = linear_5_cast_fp16)[name = string("input_15_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_0_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_0_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(27393280)))];
            tensor<fp16, [1, 291, 1024]> linear_6_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_0_mlp_down_proj_weight_to_fp16, x = input_15_cast_fp16)[name = string("linear_6_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_15_cast_fp16 = add(x = hidden_states_7_cast_fp16, y = linear_6_cast_fp16)[name = string("hidden_states_15_cast_fp16")];
            tensor<int32, [4]> var_1231 = const()[name = string("op_1231"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1243 = const()[name = string("op_1243"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_1232_cast_fp16 = transpose(perm = var_1231, x = k_5_cast_fp16)[name = string("transpose_175")];
            tensor<fp16, [1, 1024, 1, 291]> input_17_cast_fp16 = reshape(shape = var_1243, x = var_1232_cast_fp16)[name = string("input_17_cast_fp16")];
            tensor<int32, [8]> var_1249_pad_0 = const()[name = string("op_1249_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_1249_mode_0 = const()[name = string("op_1249_mode_0"), val = string("constant")];
            fp16 const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_1249_cast_fp16 = pad(constant_val = const_24_to_fp16, mode = var_1249_mode_0, pad = var_1249_pad_0, x = input_17_cast_fp16)[name = string("op_1249_cast_fp16")];
            tensor<int32, [4]> var_1254 = const()[name = string("op_1254"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1266 = const()[name = string("op_1266"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_1255_cast_fp16 = transpose(perm = var_1254, x = v_3_cast_fp16)[name = string("transpose_174")];
            tensor<fp16, [1, 1024, 1, 291]> input_19_cast_fp16 = reshape(shape = var_1266, x = var_1255_cast_fp16)[name = string("input_19_cast_fp16")];
            tensor<int32, [8]> var_1272_pad_0 = const()[name = string("op_1272_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_1272_mode_0 = const()[name = string("op_1272_mode_0"), val = string("constant")];
            fp16 const_27_to_fp16 = const()[name = string("const_27_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_1272_cast_fp16 = pad(constant_val = const_27_to_fp16, mode = var_1272_mode_0, pad = var_1272_pad_0, x = input_19_cast_fp16)[name = string("op_1272_cast_fp16")];
            fp16 var_1276_promoted_to_fp16 = const()[name = string("op_1276_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_1282_cast_fp16 = pow(x = hidden_states_15_cast_fp16, y = var_1276_promoted_to_fp16)[name = string("op_1282_cast_fp16")];
            tensor<int32, [1]> variance_5_axes_0 = const()[name = string("variance_5_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_5_keep_dims_0 = const()[name = string("variance_5_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_5_cast_fp16 = reduce_mean(axes = variance_5_axes_0, keep_dims = variance_5_keep_dims_0, x = var_1282_cast_fp16)[name = string("variance_5_cast_fp16")];
            fp16 var_1285_to_fp16 = const()[name = string("op_1285_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_1286_cast_fp16 = add(x = variance_5_cast_fp16, y = var_1285_to_fp16)[name = string("op_1286_cast_fp16")];
            fp32 var_1287_epsilon_0 = const()[name = string("op_1287_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_1287_cast_fp16 = rsqrt(epsilon = var_1287_epsilon_0, x = var_1286_cast_fp16)[name = string("op_1287_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_19_cast_fp16 = mul(x = hidden_states_15_cast_fp16, y = var_1287_cast_fp16)[name = string("hidden_states_19_cast_fp16")];
            tensor<fp16, [1024]> layers_1_input_layernorm_weight_to_fp16 = const()[name = string("layers_1_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35781952)))];
            tensor<fp16, [1, 291, 1024]> hidden_5_cast_fp16 = mul(x = layers_1_input_layernorm_weight_to_fp16, y = hidden_states_19_cast_fp16)[name = string("hidden_5_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35784064)))];
            tensor<fp16, [1, 291, 1024]> linear_7_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = hidden_5_cast_fp16)[name = string("linear_7_cast_fp16")];
            tensor<int32, [4]> var_1311 = const()[name = string("op_1311"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_7_cast_fp16 = reshape(shape = var_1311, x = linear_7_cast_fp16)[name = string("q_7_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37881280)))];
            tensor<fp16, [1, 291, 1024]> linear_8_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = hidden_5_cast_fp16)[name = string("linear_8_cast_fp16")];
            tensor<int32, [4]> var_1318 = const()[name = string("op_1318"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_7_cast_fp16 = reshape(shape = var_1318, x = linear_8_cast_fp16)[name = string("k_7_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39978496)))];
            tensor<fp16, [1, 291, 1024]> linear_9_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = hidden_5_cast_fp16)[name = string("linear_9_cast_fp16")];
            tensor<int32, [4]> var_1325 = const()[name = string("op_1325"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_5_cast_fp16 = reshape(shape = var_1325, x = linear_9_cast_fp16)[name = string("v_5_cast_fp16")];
            tensor<int32, [4]> q_9_perm_0 = const()[name = string("q_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_7_perm_0 = const()[name = string("v_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_9_cast_fp16 = transpose(perm = q_9_perm_0, x = q_7_cast_fp16)[name = string("transpose_173")];
            tensor<fp16, [1, 16, 291, 64]> var_1338_cast_fp16 = mul(x = q_9_cast_fp16, y = var_1091_to_fp16)[name = string("op_1338_cast_fp16")];
            tensor<int32, [4]> x1_5_begin_0 = const()[name = string("x1_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_5_end_0 = const()[name = string("x1_5_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_5_end_mask_0 = const()[name = string("x1_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_5_cast_fp16 = slice_by_index(begin = x1_5_begin_0, end = x1_5_end_0, end_mask = x1_5_end_mask_0, x = q_9_cast_fp16)[name = string("x1_5_cast_fp16")];
            tensor<int32, [4]> x2_5_begin_0 = const()[name = string("x2_5_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_5_end_0 = const()[name = string("x2_5_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_5_end_mask_0 = const()[name = string("x2_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_5_cast_fp16 = slice_by_index(begin = x2_5_begin_0, end = x2_5_end_0, end_mask = x2_5_end_mask_0, x = q_9_cast_fp16)[name = string("x2_5_cast_fp16")];
            fp16 const_32_promoted_to_fp16 = const()[name = string("const_32_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_1359_cast_fp16 = mul(x = x2_5_cast_fp16, y = const_32_promoted_to_fp16)[name = string("op_1359_cast_fp16")];
            int32 var_1361 = const()[name = string("op_1361"), val = int32(-1)];
            bool var_1362_interleave_0 = const()[name = string("op_1362_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_1362_cast_fp16 = concat(axis = var_1361, interleave = var_1362_interleave_0, values = (var_1359_cast_fp16, x1_5_cast_fp16))[name = string("op_1362_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_1365_cast_fp16 = mul(x = var_1362_cast_fp16, y = var_1118_to_fp16)[name = string("op_1365_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_11_cast_fp16 = add(x = var_1338_cast_fp16, y = var_1365_cast_fp16)[name = string("q_11_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = k_7_cast_fp16)[name = string("transpose_172")];
            tensor<fp16, [1, 16, 291, 64]> var_1370_cast_fp16 = mul(x = k_9_cast_fp16, y = var_1091_to_fp16)[name = string("op_1370_cast_fp16")];
            tensor<int32, [4]> x1_7_begin_0 = const()[name = string("x1_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_7_end_0 = const()[name = string("x1_7_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_7_end_mask_0 = const()[name = string("x1_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_7_cast_fp16 = slice_by_index(begin = x1_7_begin_0, end = x1_7_end_0, end_mask = x1_7_end_mask_0, x = k_9_cast_fp16)[name = string("x1_7_cast_fp16")];
            tensor<int32, [4]> x2_7_begin_0 = const()[name = string("x2_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_7_end_0 = const()[name = string("x2_7_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_7_end_mask_0 = const()[name = string("x2_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_7_cast_fp16 = slice_by_index(begin = x2_7_begin_0, end = x2_7_end_0, end_mask = x2_7_end_mask_0, x = k_9_cast_fp16)[name = string("x2_7_cast_fp16")];
            fp16 const_35_promoted_to_fp16 = const()[name = string("const_35_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_1391_cast_fp16 = mul(x = x2_7_cast_fp16, y = const_35_promoted_to_fp16)[name = string("op_1391_cast_fp16")];
            int32 var_1393 = const()[name = string("op_1393"), val = int32(-1)];
            bool var_1394_interleave_0 = const()[name = string("op_1394_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_1394_cast_fp16 = concat(axis = var_1393, interleave = var_1394_interleave_0, values = (var_1391_cast_fp16, x1_7_cast_fp16))[name = string("op_1394_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_1397_cast_fp16 = mul(x = var_1394_cast_fp16, y = var_1118_to_fp16)[name = string("op_1397_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_11_cast_fp16 = add(x = var_1370_cast_fp16, y = var_1397_cast_fp16)[name = string("k_11_cast_fp16")];
            bool var_1403_transpose_x_1 = const()[name = string("op_1403_transpose_x_1"), val = bool(false)];
            bool var_1403_transpose_y_1 = const()[name = string("op_1403_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_1403_cast_fp16 = matmul(transpose_x = var_1403_transpose_x_1, transpose_y = var_1403_transpose_y_1, x = q_11_cast_fp16, y = k_11_cast_fp16)[name = string("op_1403_cast_fp16")];
            fp16 _inversed_scores_7_y_0_to_fp16 = const()[name = string("_inversed_scores_7_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_7_cast_fp16 = mul(x = var_1403_cast_fp16, y = _inversed_scores_7_y_0_to_fp16)[name = string("_inversed_scores_7_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_9_cast_fp16 = add(x = _inversed_scores_7_cast_fp16, y = const_21_to_fp16)[name = string("scores_9_cast_fp16")];
            int32 var_1418 = const()[name = string("op_1418"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_1420_cast_fp16 = softmax(axis = var_1418, x = scores_9_cast_fp16)[name = string("op_1420_cast_fp16")];
            bool attn_out_5_transpose_x_0 = const()[name = string("attn_out_5_transpose_x_0"), val = bool(false)];
            bool attn_out_5_transpose_y_0 = const()[name = string("attn_out_5_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_7_cast_fp16 = transpose(perm = v_7_perm_0, x = v_5_cast_fp16)[name = string("transpose_171")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_5_cast_fp16 = matmul(transpose_x = attn_out_5_transpose_x_0, transpose_y = attn_out_5_transpose_y_0, x = var_1420_cast_fp16, y = v_7_cast_fp16)[name = string("attn_out_5_cast_fp16")];
            tensor<int32, [4]> var_1429_perm_0 = const()[name = string("op_1429_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1431 = const()[name = string("op_1431"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_1429_cast_fp16 = transpose(perm = var_1429_perm_0, x = attn_out_5_cast_fp16)[name = string("transpose_170")];
            tensor<fp16, [1, 291, 1024]> input_21_cast_fp16 = reshape(shape = var_1431, x = var_1429_cast_fp16)[name = string("input_21_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42075712)))];
            tensor<fp16, [1, 291, 1024]> linear_10_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_10_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_23_cast_fp16 = add(x = hidden_states_15_cast_fp16, y = linear_10_cast_fp16)[name = string("hidden_states_23_cast_fp16")];
            fp16 var_1441_promoted_to_fp16 = const()[name = string("op_1441_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_1447_cast_fp16 = pow(x = hidden_states_23_cast_fp16, y = var_1441_promoted_to_fp16)[name = string("op_1447_cast_fp16")];
            tensor<int32, [1]> variance_7_axes_0 = const()[name = string("variance_7_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_7_keep_dims_0 = const()[name = string("variance_7_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_7_cast_fp16 = reduce_mean(axes = variance_7_axes_0, keep_dims = variance_7_keep_dims_0, x = var_1447_cast_fp16)[name = string("variance_7_cast_fp16")];
            fp16 var_1450_to_fp16 = const()[name = string("op_1450_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_1451_cast_fp16 = add(x = variance_7_cast_fp16, y = var_1450_to_fp16)[name = string("op_1451_cast_fp16")];
            fp32 var_1452_epsilon_0 = const()[name = string("op_1452_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_1452_cast_fp16 = rsqrt(epsilon = var_1452_epsilon_0, x = var_1451_cast_fp16)[name = string("op_1452_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_27_cast_fp16 = mul(x = hidden_states_23_cast_fp16, y = var_1452_cast_fp16)[name = string("hidden_states_27_cast_fp16")];
            tensor<fp16, [1024]> layers_1_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_1_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44172928)))];
            tensor<fp16, [1, 291, 1024]> input_23_cast_fp16 = mul(x = layers_1_post_attention_layernorm_weight_to_fp16, y = hidden_states_27_cast_fp16)[name = string("input_23_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_1_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_1_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44175040)))];
            tensor<fp16, [1, 291, 4096]> linear_11_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_1_mlp_gate_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_11_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_1465_cast_fp16 = silu(x = linear_11_cast_fp16)[name = string("op_1465_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_1_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_1_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52563712)))];
            tensor<fp16, [1, 291, 4096]> linear_12_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_1_mlp_up_proj_weight_to_fp16, x = input_23_cast_fp16)[name = string("linear_12_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_27_cast_fp16 = mul(x = var_1465_cast_fp16, y = linear_12_cast_fp16)[name = string("input_27_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_1_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_1_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60952384)))];
            tensor<fp16, [1, 291, 1024]> linear_13_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_1_mlp_down_proj_weight_to_fp16, x = input_27_cast_fp16)[name = string("linear_13_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_31_cast_fp16 = add(x = hidden_states_23_cast_fp16, y = linear_13_cast_fp16)[name = string("hidden_states_31_cast_fp16")];
            tensor<int32, [4]> var_1477 = const()[name = string("op_1477"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1489 = const()[name = string("op_1489"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_1478_cast_fp16 = transpose(perm = var_1477, x = k_11_cast_fp16)[name = string("transpose_169")];
            tensor<fp16, [1, 1024, 1, 291]> input_29_cast_fp16 = reshape(shape = var_1489, x = var_1478_cast_fp16)[name = string("input_29_cast_fp16")];
            tensor<int32, [8]> var_1495_pad_0 = const()[name = string("op_1495_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_1495_mode_0 = const()[name = string("op_1495_mode_0"), val = string("constant")];
            fp16 const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_1495_cast_fp16 = pad(constant_val = const_39_to_fp16, mode = var_1495_mode_0, pad = var_1495_pad_0, x = input_29_cast_fp16)[name = string("op_1495_cast_fp16")];
            tensor<int32, [4]> var_1500 = const()[name = string("op_1500"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1512 = const()[name = string("op_1512"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_1501_cast_fp16 = transpose(perm = var_1500, x = v_7_cast_fp16)[name = string("transpose_168")];
            tensor<fp16, [1, 1024, 1, 291]> input_31_cast_fp16 = reshape(shape = var_1512, x = var_1501_cast_fp16)[name = string("input_31_cast_fp16")];
            tensor<int32, [8]> var_1518_pad_0 = const()[name = string("op_1518_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_1518_mode_0 = const()[name = string("op_1518_mode_0"), val = string("constant")];
            fp16 const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_1518_cast_fp16 = pad(constant_val = const_42_to_fp16, mode = var_1518_mode_0, pad = var_1518_pad_0, x = input_31_cast_fp16)[name = string("op_1518_cast_fp16")];
            fp16 var_1522_promoted_to_fp16 = const()[name = string("op_1522_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_1528_cast_fp16 = pow(x = hidden_states_31_cast_fp16, y = var_1522_promoted_to_fp16)[name = string("op_1528_cast_fp16")];
            tensor<int32, [1]> variance_9_axes_0 = const()[name = string("variance_9_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_9_keep_dims_0 = const()[name = string("variance_9_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_9_cast_fp16 = reduce_mean(axes = variance_9_axes_0, keep_dims = variance_9_keep_dims_0, x = var_1528_cast_fp16)[name = string("variance_9_cast_fp16")];
            fp16 var_1531_to_fp16 = const()[name = string("op_1531_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_1532_cast_fp16 = add(x = variance_9_cast_fp16, y = var_1531_to_fp16)[name = string("op_1532_cast_fp16")];
            fp32 var_1533_epsilon_0 = const()[name = string("op_1533_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_1533_cast_fp16 = rsqrt(epsilon = var_1533_epsilon_0, x = var_1532_cast_fp16)[name = string("op_1533_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_35_cast_fp16 = mul(x = hidden_states_31_cast_fp16, y = var_1533_cast_fp16)[name = string("hidden_states_35_cast_fp16")];
            tensor<fp16, [1024]> layers_2_input_layernorm_weight_to_fp16 = const()[name = string("layers_2_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69341056)))];
            tensor<fp16, [1, 291, 1024]> hidden_9_cast_fp16 = mul(x = layers_2_input_layernorm_weight_to_fp16, y = hidden_states_35_cast_fp16)[name = string("hidden_9_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69343168)))];
            tensor<fp16, [1, 291, 1024]> linear_14_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = hidden_9_cast_fp16)[name = string("linear_14_cast_fp16")];
            tensor<int32, [4]> var_1557 = const()[name = string("op_1557"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_13_cast_fp16 = reshape(shape = var_1557, x = linear_14_cast_fp16)[name = string("q_13_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(71440384)))];
            tensor<fp16, [1, 291, 1024]> linear_15_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = hidden_9_cast_fp16)[name = string("linear_15_cast_fp16")];
            tensor<int32, [4]> var_1564 = const()[name = string("op_1564"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_13_cast_fp16 = reshape(shape = var_1564, x = linear_15_cast_fp16)[name = string("k_13_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73537600)))];
            tensor<fp16, [1, 291, 1024]> linear_16_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = hidden_9_cast_fp16)[name = string("linear_16_cast_fp16")];
            tensor<int32, [4]> var_1571 = const()[name = string("op_1571"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_9_cast_fp16 = reshape(shape = var_1571, x = linear_16_cast_fp16)[name = string("v_9_cast_fp16")];
            tensor<int32, [4]> q_15_perm_0 = const()[name = string("q_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_15_perm_0 = const()[name = string("k_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_11_perm_0 = const()[name = string("v_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_15_cast_fp16 = transpose(perm = q_15_perm_0, x = q_13_cast_fp16)[name = string("transpose_167")];
            tensor<fp16, [1, 16, 291, 64]> var_1584_cast_fp16 = mul(x = q_15_cast_fp16, y = var_1091_to_fp16)[name = string("op_1584_cast_fp16")];
            tensor<int32, [4]> x1_9_begin_0 = const()[name = string("x1_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_9_end_0 = const()[name = string("x1_9_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_9_end_mask_0 = const()[name = string("x1_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_9_cast_fp16 = slice_by_index(begin = x1_9_begin_0, end = x1_9_end_0, end_mask = x1_9_end_mask_0, x = q_15_cast_fp16)[name = string("x1_9_cast_fp16")];
            tensor<int32, [4]> x2_9_begin_0 = const()[name = string("x2_9_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_9_end_0 = const()[name = string("x2_9_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_9_end_mask_0 = const()[name = string("x2_9_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_9_cast_fp16 = slice_by_index(begin = x2_9_begin_0, end = x2_9_end_0, end_mask = x2_9_end_mask_0, x = q_15_cast_fp16)[name = string("x2_9_cast_fp16")];
            fp16 const_47_promoted_to_fp16 = const()[name = string("const_47_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_1605_cast_fp16 = mul(x = x2_9_cast_fp16, y = const_47_promoted_to_fp16)[name = string("op_1605_cast_fp16")];
            int32 var_1607 = const()[name = string("op_1607"), val = int32(-1)];
            bool var_1608_interleave_0 = const()[name = string("op_1608_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_1608_cast_fp16 = concat(axis = var_1607, interleave = var_1608_interleave_0, values = (var_1605_cast_fp16, x1_9_cast_fp16))[name = string("op_1608_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_1611_cast_fp16 = mul(x = var_1608_cast_fp16, y = var_1118_to_fp16)[name = string("op_1611_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_17_cast_fp16 = add(x = var_1584_cast_fp16, y = var_1611_cast_fp16)[name = string("q_17_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_15_cast_fp16 = transpose(perm = k_15_perm_0, x = k_13_cast_fp16)[name = string("transpose_166")];
            tensor<fp16, [1, 16, 291, 64]> var_1616_cast_fp16 = mul(x = k_15_cast_fp16, y = var_1091_to_fp16)[name = string("op_1616_cast_fp16")];
            tensor<int32, [4]> x1_11_begin_0 = const()[name = string("x1_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_11_end_0 = const()[name = string("x1_11_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_11_end_mask_0 = const()[name = string("x1_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_11_cast_fp16 = slice_by_index(begin = x1_11_begin_0, end = x1_11_end_0, end_mask = x1_11_end_mask_0, x = k_15_cast_fp16)[name = string("x1_11_cast_fp16")];
            tensor<int32, [4]> x2_11_begin_0 = const()[name = string("x2_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_11_end_0 = const()[name = string("x2_11_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_11_end_mask_0 = const()[name = string("x2_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_11_cast_fp16 = slice_by_index(begin = x2_11_begin_0, end = x2_11_end_0, end_mask = x2_11_end_mask_0, x = k_15_cast_fp16)[name = string("x2_11_cast_fp16")];
            fp16 const_50_promoted_to_fp16 = const()[name = string("const_50_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_1637_cast_fp16 = mul(x = x2_11_cast_fp16, y = const_50_promoted_to_fp16)[name = string("op_1637_cast_fp16")];
            int32 var_1639 = const()[name = string("op_1639"), val = int32(-1)];
            bool var_1640_interleave_0 = const()[name = string("op_1640_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_1640_cast_fp16 = concat(axis = var_1639, interleave = var_1640_interleave_0, values = (var_1637_cast_fp16, x1_11_cast_fp16))[name = string("op_1640_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_1643_cast_fp16 = mul(x = var_1640_cast_fp16, y = var_1118_to_fp16)[name = string("op_1643_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_17_cast_fp16 = add(x = var_1616_cast_fp16, y = var_1643_cast_fp16)[name = string("k_17_cast_fp16")];
            bool var_1649_transpose_x_1 = const()[name = string("op_1649_transpose_x_1"), val = bool(false)];
            bool var_1649_transpose_y_1 = const()[name = string("op_1649_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_1649_cast_fp16 = matmul(transpose_x = var_1649_transpose_x_1, transpose_y = var_1649_transpose_y_1, x = q_17_cast_fp16, y = k_17_cast_fp16)[name = string("op_1649_cast_fp16")];
            fp16 _inversed_scores_13_y_0_to_fp16 = const()[name = string("_inversed_scores_13_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_13_cast_fp16 = mul(x = var_1649_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = string("_inversed_scores_13_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_15_cast_fp16 = add(x = _inversed_scores_13_cast_fp16, y = const_21_to_fp16)[name = string("scores_15_cast_fp16")];
            int32 var_1664 = const()[name = string("op_1664"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_1666_cast_fp16 = softmax(axis = var_1664, x = scores_15_cast_fp16)[name = string("op_1666_cast_fp16")];
            bool attn_out_9_transpose_x_0 = const()[name = string("attn_out_9_transpose_x_0"), val = bool(false)];
            bool attn_out_9_transpose_y_0 = const()[name = string("attn_out_9_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_11_cast_fp16 = transpose(perm = v_11_perm_0, x = v_9_cast_fp16)[name = string("transpose_165")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_9_cast_fp16 = matmul(transpose_x = attn_out_9_transpose_x_0, transpose_y = attn_out_9_transpose_y_0, x = var_1666_cast_fp16, y = v_11_cast_fp16)[name = string("attn_out_9_cast_fp16")];
            tensor<int32, [4]> var_1675_perm_0 = const()[name = string("op_1675_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1677 = const()[name = string("op_1677"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_1675_cast_fp16 = transpose(perm = var_1675_perm_0, x = attn_out_9_cast_fp16)[name = string("transpose_164")];
            tensor<fp16, [1, 291, 1024]> input_33_cast_fp16 = reshape(shape = var_1677, x = var_1675_cast_fp16)[name = string("input_33_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(75634816)))];
            tensor<fp16, [1, 291, 1024]> linear_17_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("linear_17_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_39_cast_fp16 = add(x = hidden_states_31_cast_fp16, y = linear_17_cast_fp16)[name = string("hidden_states_39_cast_fp16")];
            fp16 var_1687_promoted_to_fp16 = const()[name = string("op_1687_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_1693_cast_fp16 = pow(x = hidden_states_39_cast_fp16, y = var_1687_promoted_to_fp16)[name = string("op_1693_cast_fp16")];
            tensor<int32, [1]> variance_11_axes_0 = const()[name = string("variance_11_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_11_keep_dims_0 = const()[name = string("variance_11_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_11_cast_fp16 = reduce_mean(axes = variance_11_axes_0, keep_dims = variance_11_keep_dims_0, x = var_1693_cast_fp16)[name = string("variance_11_cast_fp16")];
            fp16 var_1696_to_fp16 = const()[name = string("op_1696_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_1697_cast_fp16 = add(x = variance_11_cast_fp16, y = var_1696_to_fp16)[name = string("op_1697_cast_fp16")];
            fp32 var_1698_epsilon_0 = const()[name = string("op_1698_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_1698_cast_fp16 = rsqrt(epsilon = var_1698_epsilon_0, x = var_1697_cast_fp16)[name = string("op_1698_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_43_cast_fp16 = mul(x = hidden_states_39_cast_fp16, y = var_1698_cast_fp16)[name = string("hidden_states_43_cast_fp16")];
            tensor<fp16, [1024]> layers_2_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_2_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77732032)))];
            tensor<fp16, [1, 291, 1024]> input_35_cast_fp16 = mul(x = layers_2_post_attention_layernorm_weight_to_fp16, y = hidden_states_43_cast_fp16)[name = string("input_35_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_2_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_2_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77734144)))];
            tensor<fp16, [1, 291, 4096]> linear_18_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_2_mlp_gate_proj_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_18_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_1711_cast_fp16 = silu(x = linear_18_cast_fp16)[name = string("op_1711_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_2_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_2_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86122816)))];
            tensor<fp16, [1, 291, 4096]> linear_19_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_2_mlp_up_proj_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_19_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_39_cast_fp16 = mul(x = var_1711_cast_fp16, y = linear_19_cast_fp16)[name = string("input_39_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_2_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_2_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94511488)))];
            tensor<fp16, [1, 291, 1024]> linear_20_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_2_mlp_down_proj_weight_to_fp16, x = input_39_cast_fp16)[name = string("linear_20_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_47_cast_fp16 = add(x = hidden_states_39_cast_fp16, y = linear_20_cast_fp16)[name = string("hidden_states_47_cast_fp16")];
            tensor<int32, [4]> var_1723 = const()[name = string("op_1723"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1735 = const()[name = string("op_1735"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_1724_cast_fp16 = transpose(perm = var_1723, x = k_17_cast_fp16)[name = string("transpose_163")];
            tensor<fp16, [1, 1024, 1, 291]> input_41_cast_fp16 = reshape(shape = var_1735, x = var_1724_cast_fp16)[name = string("input_41_cast_fp16")];
            tensor<int32, [8]> var_1741_pad_0 = const()[name = string("op_1741_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_1741_mode_0 = const()[name = string("op_1741_mode_0"), val = string("constant")];
            fp16 const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_1741_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = var_1741_mode_0, pad = var_1741_pad_0, x = input_41_cast_fp16)[name = string("op_1741_cast_fp16")];
            tensor<int32, [4]> var_1746 = const()[name = string("op_1746"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1758 = const()[name = string("op_1758"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_1747_cast_fp16 = transpose(perm = var_1746, x = v_11_cast_fp16)[name = string("transpose_162")];
            tensor<fp16, [1, 1024, 1, 291]> input_43_cast_fp16 = reshape(shape = var_1758, x = var_1747_cast_fp16)[name = string("input_43_cast_fp16")];
            tensor<int32, [8]> var_1764_pad_0 = const()[name = string("op_1764_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_1764_mode_0 = const()[name = string("op_1764_mode_0"), val = string("constant")];
            fp16 const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_1764_cast_fp16 = pad(constant_val = const_57_to_fp16, mode = var_1764_mode_0, pad = var_1764_pad_0, x = input_43_cast_fp16)[name = string("op_1764_cast_fp16")];
            fp16 var_1768_promoted_to_fp16 = const()[name = string("op_1768_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_1774_cast_fp16 = pow(x = hidden_states_47_cast_fp16, y = var_1768_promoted_to_fp16)[name = string("op_1774_cast_fp16")];
            tensor<int32, [1]> variance_13_axes_0 = const()[name = string("variance_13_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_13_keep_dims_0 = const()[name = string("variance_13_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_13_cast_fp16 = reduce_mean(axes = variance_13_axes_0, keep_dims = variance_13_keep_dims_0, x = var_1774_cast_fp16)[name = string("variance_13_cast_fp16")];
            fp16 var_1777_to_fp16 = const()[name = string("op_1777_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_1778_cast_fp16 = add(x = variance_13_cast_fp16, y = var_1777_to_fp16)[name = string("op_1778_cast_fp16")];
            fp32 var_1779_epsilon_0 = const()[name = string("op_1779_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_1779_cast_fp16 = rsqrt(epsilon = var_1779_epsilon_0, x = var_1778_cast_fp16)[name = string("op_1779_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_51_cast_fp16 = mul(x = hidden_states_47_cast_fp16, y = var_1779_cast_fp16)[name = string("hidden_states_51_cast_fp16")];
            tensor<fp16, [1024]> layers_3_input_layernorm_weight_to_fp16 = const()[name = string("layers_3_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102900160)))];
            tensor<fp16, [1, 291, 1024]> hidden_13_cast_fp16 = mul(x = layers_3_input_layernorm_weight_to_fp16, y = hidden_states_51_cast_fp16)[name = string("hidden_13_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102902272)))];
            tensor<fp16, [1, 291, 1024]> linear_21_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = hidden_13_cast_fp16)[name = string("linear_21_cast_fp16")];
            tensor<int32, [4]> var_1803 = const()[name = string("op_1803"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_19_cast_fp16 = reshape(shape = var_1803, x = linear_21_cast_fp16)[name = string("q_19_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104999488)))];
            tensor<fp16, [1, 291, 1024]> linear_22_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = hidden_13_cast_fp16)[name = string("linear_22_cast_fp16")];
            tensor<int32, [4]> var_1810 = const()[name = string("op_1810"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_19_cast_fp16 = reshape(shape = var_1810, x = linear_22_cast_fp16)[name = string("k_19_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107096704)))];
            tensor<fp16, [1, 291, 1024]> linear_23_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = hidden_13_cast_fp16)[name = string("linear_23_cast_fp16")];
            tensor<int32, [4]> var_1817 = const()[name = string("op_1817"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_13_cast_fp16 = reshape(shape = var_1817, x = linear_23_cast_fp16)[name = string("v_13_cast_fp16")];
            tensor<int32, [4]> q_21_perm_0 = const()[name = string("q_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_21_perm_0 = const()[name = string("k_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_15_perm_0 = const()[name = string("v_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_21_cast_fp16 = transpose(perm = q_21_perm_0, x = q_19_cast_fp16)[name = string("transpose_161")];
            tensor<fp16, [1, 16, 291, 64]> var_1830_cast_fp16 = mul(x = q_21_cast_fp16, y = var_1091_to_fp16)[name = string("op_1830_cast_fp16")];
            tensor<int32, [4]> x1_13_begin_0 = const()[name = string("x1_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_13_end_0 = const()[name = string("x1_13_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_13_end_mask_0 = const()[name = string("x1_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_13_cast_fp16 = slice_by_index(begin = x1_13_begin_0, end = x1_13_end_0, end_mask = x1_13_end_mask_0, x = q_21_cast_fp16)[name = string("x1_13_cast_fp16")];
            tensor<int32, [4]> x2_13_begin_0 = const()[name = string("x2_13_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_13_end_0 = const()[name = string("x2_13_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_13_end_mask_0 = const()[name = string("x2_13_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_13_cast_fp16 = slice_by_index(begin = x2_13_begin_0, end = x2_13_end_0, end_mask = x2_13_end_mask_0, x = q_21_cast_fp16)[name = string("x2_13_cast_fp16")];
            fp16 const_62_promoted_to_fp16 = const()[name = string("const_62_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_1851_cast_fp16 = mul(x = x2_13_cast_fp16, y = const_62_promoted_to_fp16)[name = string("op_1851_cast_fp16")];
            int32 var_1853 = const()[name = string("op_1853"), val = int32(-1)];
            bool var_1854_interleave_0 = const()[name = string("op_1854_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_1854_cast_fp16 = concat(axis = var_1853, interleave = var_1854_interleave_0, values = (var_1851_cast_fp16, x1_13_cast_fp16))[name = string("op_1854_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_1857_cast_fp16 = mul(x = var_1854_cast_fp16, y = var_1118_to_fp16)[name = string("op_1857_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_23_cast_fp16 = add(x = var_1830_cast_fp16, y = var_1857_cast_fp16)[name = string("q_23_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_21_cast_fp16 = transpose(perm = k_21_perm_0, x = k_19_cast_fp16)[name = string("transpose_160")];
            tensor<fp16, [1, 16, 291, 64]> var_1862_cast_fp16 = mul(x = k_21_cast_fp16, y = var_1091_to_fp16)[name = string("op_1862_cast_fp16")];
            tensor<int32, [4]> x1_15_begin_0 = const()[name = string("x1_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_15_end_0 = const()[name = string("x1_15_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_15_end_mask_0 = const()[name = string("x1_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_15_cast_fp16 = slice_by_index(begin = x1_15_begin_0, end = x1_15_end_0, end_mask = x1_15_end_mask_0, x = k_21_cast_fp16)[name = string("x1_15_cast_fp16")];
            tensor<int32, [4]> x2_15_begin_0 = const()[name = string("x2_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_15_end_0 = const()[name = string("x2_15_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_15_end_mask_0 = const()[name = string("x2_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_15_cast_fp16 = slice_by_index(begin = x2_15_begin_0, end = x2_15_end_0, end_mask = x2_15_end_mask_0, x = k_21_cast_fp16)[name = string("x2_15_cast_fp16")];
            fp16 const_65_promoted_to_fp16 = const()[name = string("const_65_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_1883_cast_fp16 = mul(x = x2_15_cast_fp16, y = const_65_promoted_to_fp16)[name = string("op_1883_cast_fp16")];
            int32 var_1885 = const()[name = string("op_1885"), val = int32(-1)];
            bool var_1886_interleave_0 = const()[name = string("op_1886_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_1886_cast_fp16 = concat(axis = var_1885, interleave = var_1886_interleave_0, values = (var_1883_cast_fp16, x1_15_cast_fp16))[name = string("op_1886_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_1889_cast_fp16 = mul(x = var_1886_cast_fp16, y = var_1118_to_fp16)[name = string("op_1889_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_23_cast_fp16 = add(x = var_1862_cast_fp16, y = var_1889_cast_fp16)[name = string("k_23_cast_fp16")];
            bool var_1895_transpose_x_1 = const()[name = string("op_1895_transpose_x_1"), val = bool(false)];
            bool var_1895_transpose_y_1 = const()[name = string("op_1895_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_1895_cast_fp16 = matmul(transpose_x = var_1895_transpose_x_1, transpose_y = var_1895_transpose_y_1, x = q_23_cast_fp16, y = k_23_cast_fp16)[name = string("op_1895_cast_fp16")];
            fp16 _inversed_scores_19_y_0_to_fp16 = const()[name = string("_inversed_scores_19_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_19_cast_fp16 = mul(x = var_1895_cast_fp16, y = _inversed_scores_19_y_0_to_fp16)[name = string("_inversed_scores_19_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_21_cast_fp16 = add(x = _inversed_scores_19_cast_fp16, y = const_21_to_fp16)[name = string("scores_21_cast_fp16")];
            int32 var_1910 = const()[name = string("op_1910"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_1912_cast_fp16 = softmax(axis = var_1910, x = scores_21_cast_fp16)[name = string("op_1912_cast_fp16")];
            bool attn_out_13_transpose_x_0 = const()[name = string("attn_out_13_transpose_x_0"), val = bool(false)];
            bool attn_out_13_transpose_y_0 = const()[name = string("attn_out_13_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_15_cast_fp16 = transpose(perm = v_15_perm_0, x = v_13_cast_fp16)[name = string("transpose_159")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_13_cast_fp16 = matmul(transpose_x = attn_out_13_transpose_x_0, transpose_y = attn_out_13_transpose_y_0, x = var_1912_cast_fp16, y = v_15_cast_fp16)[name = string("attn_out_13_cast_fp16")];
            tensor<int32, [4]> var_1921_perm_0 = const()[name = string("op_1921_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1923 = const()[name = string("op_1923"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_1921_cast_fp16 = transpose(perm = var_1921_perm_0, x = attn_out_13_cast_fp16)[name = string("transpose_158")];
            tensor<fp16, [1, 291, 1024]> input_45_cast_fp16 = reshape(shape = var_1923, x = var_1921_cast_fp16)[name = string("input_45_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(109193920)))];
            tensor<fp16, [1, 291, 1024]> linear_24_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_45_cast_fp16)[name = string("linear_24_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_55_cast_fp16 = add(x = hidden_states_47_cast_fp16, y = linear_24_cast_fp16)[name = string("hidden_states_55_cast_fp16")];
            fp16 var_1933_promoted_to_fp16 = const()[name = string("op_1933_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_1939_cast_fp16 = pow(x = hidden_states_55_cast_fp16, y = var_1933_promoted_to_fp16)[name = string("op_1939_cast_fp16")];
            tensor<int32, [1]> variance_15_axes_0 = const()[name = string("variance_15_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_15_keep_dims_0 = const()[name = string("variance_15_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_15_cast_fp16 = reduce_mean(axes = variance_15_axes_0, keep_dims = variance_15_keep_dims_0, x = var_1939_cast_fp16)[name = string("variance_15_cast_fp16")];
            fp16 var_1942_to_fp16 = const()[name = string("op_1942_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_1943_cast_fp16 = add(x = variance_15_cast_fp16, y = var_1942_to_fp16)[name = string("op_1943_cast_fp16")];
            fp32 var_1944_epsilon_0 = const()[name = string("op_1944_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_1944_cast_fp16 = rsqrt(epsilon = var_1944_epsilon_0, x = var_1943_cast_fp16)[name = string("op_1944_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_59_cast_fp16 = mul(x = hidden_states_55_cast_fp16, y = var_1944_cast_fp16)[name = string("hidden_states_59_cast_fp16")];
            tensor<fp16, [1024]> layers_3_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_3_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111291136)))];
            tensor<fp16, [1, 291, 1024]> input_47_cast_fp16 = mul(x = layers_3_post_attention_layernorm_weight_to_fp16, y = hidden_states_59_cast_fp16)[name = string("input_47_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_3_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_3_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(111293248)))];
            tensor<fp16, [1, 291, 4096]> linear_25_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_3_mlp_gate_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_25_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_1957_cast_fp16 = silu(x = linear_25_cast_fp16)[name = string("op_1957_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_3_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_3_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119681920)))];
            tensor<fp16, [1, 291, 4096]> linear_26_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_3_mlp_up_proj_weight_to_fp16, x = input_47_cast_fp16)[name = string("linear_26_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_51_cast_fp16 = mul(x = var_1957_cast_fp16, y = linear_26_cast_fp16)[name = string("input_51_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_3_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_3_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(128070592)))];
            tensor<fp16, [1, 291, 1024]> linear_27_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_3_mlp_down_proj_weight_to_fp16, x = input_51_cast_fp16)[name = string("linear_27_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_63_cast_fp16 = add(x = hidden_states_55_cast_fp16, y = linear_27_cast_fp16)[name = string("hidden_states_63_cast_fp16")];
            tensor<int32, [4]> var_1969 = const()[name = string("op_1969"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_1981 = const()[name = string("op_1981"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_1970_cast_fp16 = transpose(perm = var_1969, x = k_23_cast_fp16)[name = string("transpose_157")];
            tensor<fp16, [1, 1024, 1, 291]> input_53_cast_fp16 = reshape(shape = var_1981, x = var_1970_cast_fp16)[name = string("input_53_cast_fp16")];
            tensor<int32, [8]> var_1987_pad_0 = const()[name = string("op_1987_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_1987_mode_0 = const()[name = string("op_1987_mode_0"), val = string("constant")];
            fp16 const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_1987_cast_fp16 = pad(constant_val = const_69_to_fp16, mode = var_1987_mode_0, pad = var_1987_pad_0, x = input_53_cast_fp16)[name = string("op_1987_cast_fp16")];
            tensor<int32, [4]> var_1992 = const()[name = string("op_1992"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2004 = const()[name = string("op_2004"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_1993_cast_fp16 = transpose(perm = var_1992, x = v_15_cast_fp16)[name = string("transpose_156")];
            tensor<fp16, [1, 1024, 1, 291]> input_55_cast_fp16 = reshape(shape = var_2004, x = var_1993_cast_fp16)[name = string("input_55_cast_fp16")];
            tensor<int32, [8]> var_2010_pad_0 = const()[name = string("op_2010_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2010_mode_0 = const()[name = string("op_2010_mode_0"), val = string("constant")];
            fp16 const_72_to_fp16 = const()[name = string("const_72_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2010_cast_fp16 = pad(constant_val = const_72_to_fp16, mode = var_2010_mode_0, pad = var_2010_pad_0, x = input_55_cast_fp16)[name = string("op_2010_cast_fp16")];
            fp16 var_2014_promoted_to_fp16 = const()[name = string("op_2014_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_2020_cast_fp16 = pow(x = hidden_states_63_cast_fp16, y = var_2014_promoted_to_fp16)[name = string("op_2020_cast_fp16")];
            tensor<int32, [1]> variance_17_axes_0 = const()[name = string("variance_17_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_17_keep_dims_0 = const()[name = string("variance_17_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_17_cast_fp16 = reduce_mean(axes = variance_17_axes_0, keep_dims = variance_17_keep_dims_0, x = var_2020_cast_fp16)[name = string("variance_17_cast_fp16")];
            fp16 var_2023_to_fp16 = const()[name = string("op_2023_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_2024_cast_fp16 = add(x = variance_17_cast_fp16, y = var_2023_to_fp16)[name = string("op_2024_cast_fp16")];
            fp32 var_2025_epsilon_0 = const()[name = string("op_2025_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_2025_cast_fp16 = rsqrt(epsilon = var_2025_epsilon_0, x = var_2024_cast_fp16)[name = string("op_2025_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_67_cast_fp16 = mul(x = hidden_states_63_cast_fp16, y = var_2025_cast_fp16)[name = string("hidden_states_67_cast_fp16")];
            tensor<fp16, [1024]> layers_4_input_layernorm_weight_to_fp16 = const()[name = string("layers_4_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136459264)))];
            tensor<fp16, [1, 291, 1024]> hidden_17_cast_fp16 = mul(x = layers_4_input_layernorm_weight_to_fp16, y = hidden_states_67_cast_fp16)[name = string("hidden_17_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136461376)))];
            tensor<fp16, [1, 291, 1024]> linear_28_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = hidden_17_cast_fp16)[name = string("linear_28_cast_fp16")];
            tensor<int32, [4]> var_2049 = const()[name = string("op_2049"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_25_cast_fp16 = reshape(shape = var_2049, x = linear_28_cast_fp16)[name = string("q_25_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138558592)))];
            tensor<fp16, [1, 291, 1024]> linear_29_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = hidden_17_cast_fp16)[name = string("linear_29_cast_fp16")];
            tensor<int32, [4]> var_2056 = const()[name = string("op_2056"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_25_cast_fp16 = reshape(shape = var_2056, x = linear_29_cast_fp16)[name = string("k_25_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140655808)))];
            tensor<fp16, [1, 291, 1024]> linear_30_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = hidden_17_cast_fp16)[name = string("linear_30_cast_fp16")];
            tensor<int32, [4]> var_2063 = const()[name = string("op_2063"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_17_cast_fp16 = reshape(shape = var_2063, x = linear_30_cast_fp16)[name = string("v_17_cast_fp16")];
            tensor<int32, [4]> q_27_perm_0 = const()[name = string("q_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_27_perm_0 = const()[name = string("k_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_19_perm_0 = const()[name = string("v_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_27_cast_fp16 = transpose(perm = q_27_perm_0, x = q_25_cast_fp16)[name = string("transpose_155")];
            tensor<fp16, [1, 16, 291, 64]> var_2076_cast_fp16 = mul(x = q_27_cast_fp16, y = var_1091_to_fp16)[name = string("op_2076_cast_fp16")];
            tensor<int32, [4]> x1_17_begin_0 = const()[name = string("x1_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_17_end_0 = const()[name = string("x1_17_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_17_end_mask_0 = const()[name = string("x1_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_17_cast_fp16 = slice_by_index(begin = x1_17_begin_0, end = x1_17_end_0, end_mask = x1_17_end_mask_0, x = q_27_cast_fp16)[name = string("x1_17_cast_fp16")];
            tensor<int32, [4]> x2_17_begin_0 = const()[name = string("x2_17_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_17_end_0 = const()[name = string("x2_17_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_17_end_mask_0 = const()[name = string("x2_17_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_17_cast_fp16 = slice_by_index(begin = x2_17_begin_0, end = x2_17_end_0, end_mask = x2_17_end_mask_0, x = q_27_cast_fp16)[name = string("x2_17_cast_fp16")];
            fp16 const_77_promoted_to_fp16 = const()[name = string("const_77_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_2097_cast_fp16 = mul(x = x2_17_cast_fp16, y = const_77_promoted_to_fp16)[name = string("op_2097_cast_fp16")];
            int32 var_2099 = const()[name = string("op_2099"), val = int32(-1)];
            bool var_2100_interleave_0 = const()[name = string("op_2100_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_2100_cast_fp16 = concat(axis = var_2099, interleave = var_2100_interleave_0, values = (var_2097_cast_fp16, x1_17_cast_fp16))[name = string("op_2100_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_2103_cast_fp16 = mul(x = var_2100_cast_fp16, y = var_1118_to_fp16)[name = string("op_2103_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_29_cast_fp16 = add(x = var_2076_cast_fp16, y = var_2103_cast_fp16)[name = string("q_29_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_27_cast_fp16 = transpose(perm = k_27_perm_0, x = k_25_cast_fp16)[name = string("transpose_154")];
            tensor<fp16, [1, 16, 291, 64]> var_2108_cast_fp16 = mul(x = k_27_cast_fp16, y = var_1091_to_fp16)[name = string("op_2108_cast_fp16")];
            tensor<int32, [4]> x1_19_begin_0 = const()[name = string("x1_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_19_end_0 = const()[name = string("x1_19_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_19_end_mask_0 = const()[name = string("x1_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_19_cast_fp16 = slice_by_index(begin = x1_19_begin_0, end = x1_19_end_0, end_mask = x1_19_end_mask_0, x = k_27_cast_fp16)[name = string("x1_19_cast_fp16")];
            tensor<int32, [4]> x2_19_begin_0 = const()[name = string("x2_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_19_end_0 = const()[name = string("x2_19_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_19_end_mask_0 = const()[name = string("x2_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_19_cast_fp16 = slice_by_index(begin = x2_19_begin_0, end = x2_19_end_0, end_mask = x2_19_end_mask_0, x = k_27_cast_fp16)[name = string("x2_19_cast_fp16")];
            fp16 const_80_promoted_to_fp16 = const()[name = string("const_80_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_2129_cast_fp16 = mul(x = x2_19_cast_fp16, y = const_80_promoted_to_fp16)[name = string("op_2129_cast_fp16")];
            int32 var_2131 = const()[name = string("op_2131"), val = int32(-1)];
            bool var_2132_interleave_0 = const()[name = string("op_2132_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_2132_cast_fp16 = concat(axis = var_2131, interleave = var_2132_interleave_0, values = (var_2129_cast_fp16, x1_19_cast_fp16))[name = string("op_2132_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_2135_cast_fp16 = mul(x = var_2132_cast_fp16, y = var_1118_to_fp16)[name = string("op_2135_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_29_cast_fp16 = add(x = var_2108_cast_fp16, y = var_2135_cast_fp16)[name = string("k_29_cast_fp16")];
            bool var_2141_transpose_x_1 = const()[name = string("op_2141_transpose_x_1"), val = bool(false)];
            bool var_2141_transpose_y_1 = const()[name = string("op_2141_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_2141_cast_fp16 = matmul(transpose_x = var_2141_transpose_x_1, transpose_y = var_2141_transpose_y_1, x = q_29_cast_fp16, y = k_29_cast_fp16)[name = string("op_2141_cast_fp16")];
            fp16 _inversed_scores_25_y_0_to_fp16 = const()[name = string("_inversed_scores_25_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_25_cast_fp16 = mul(x = var_2141_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = string("_inversed_scores_25_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_27_cast_fp16 = add(x = _inversed_scores_25_cast_fp16, y = const_21_to_fp16)[name = string("scores_27_cast_fp16")];
            int32 var_2156 = const()[name = string("op_2156"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_2158_cast_fp16 = softmax(axis = var_2156, x = scores_27_cast_fp16)[name = string("op_2158_cast_fp16")];
            bool attn_out_17_transpose_x_0 = const()[name = string("attn_out_17_transpose_x_0"), val = bool(false)];
            bool attn_out_17_transpose_y_0 = const()[name = string("attn_out_17_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_19_cast_fp16 = transpose(perm = v_19_perm_0, x = v_17_cast_fp16)[name = string("transpose_153")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_17_cast_fp16 = matmul(transpose_x = attn_out_17_transpose_x_0, transpose_y = attn_out_17_transpose_y_0, x = var_2158_cast_fp16, y = v_19_cast_fp16)[name = string("attn_out_17_cast_fp16")];
            tensor<int32, [4]> var_2167_perm_0 = const()[name = string("op_2167_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2169 = const()[name = string("op_2169"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_2167_cast_fp16 = transpose(perm = var_2167_perm_0, x = attn_out_17_cast_fp16)[name = string("transpose_152")];
            tensor<fp16, [1, 291, 1024]> input_57_cast_fp16 = reshape(shape = var_2169, x = var_2167_cast_fp16)[name = string("input_57_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142753024)))];
            tensor<fp16, [1, 291, 1024]> linear_31_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_57_cast_fp16)[name = string("linear_31_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_71_cast_fp16 = add(x = hidden_states_63_cast_fp16, y = linear_31_cast_fp16)[name = string("hidden_states_71_cast_fp16")];
            fp16 var_2179_promoted_to_fp16 = const()[name = string("op_2179_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_2185_cast_fp16 = pow(x = hidden_states_71_cast_fp16, y = var_2179_promoted_to_fp16)[name = string("op_2185_cast_fp16")];
            tensor<int32, [1]> variance_19_axes_0 = const()[name = string("variance_19_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_19_keep_dims_0 = const()[name = string("variance_19_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_19_cast_fp16 = reduce_mean(axes = variance_19_axes_0, keep_dims = variance_19_keep_dims_0, x = var_2185_cast_fp16)[name = string("variance_19_cast_fp16")];
            fp16 var_2188_to_fp16 = const()[name = string("op_2188_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_2189_cast_fp16 = add(x = variance_19_cast_fp16, y = var_2188_to_fp16)[name = string("op_2189_cast_fp16")];
            fp32 var_2190_epsilon_0 = const()[name = string("op_2190_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_2190_cast_fp16 = rsqrt(epsilon = var_2190_epsilon_0, x = var_2189_cast_fp16)[name = string("op_2190_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_75_cast_fp16 = mul(x = hidden_states_71_cast_fp16, y = var_2190_cast_fp16)[name = string("hidden_states_75_cast_fp16")];
            tensor<fp16, [1024]> layers_4_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_4_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144850240)))];
            tensor<fp16, [1, 291, 1024]> input_59_cast_fp16 = mul(x = layers_4_post_attention_layernorm_weight_to_fp16, y = hidden_states_75_cast_fp16)[name = string("input_59_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_4_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_4_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144852352)))];
            tensor<fp16, [1, 291, 4096]> linear_32_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_4_mlp_gate_proj_weight_to_fp16, x = input_59_cast_fp16)[name = string("linear_32_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_2203_cast_fp16 = silu(x = linear_32_cast_fp16)[name = string("op_2203_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_4_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_4_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(153241024)))];
            tensor<fp16, [1, 291, 4096]> linear_33_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_4_mlp_up_proj_weight_to_fp16, x = input_59_cast_fp16)[name = string("linear_33_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_63_cast_fp16 = mul(x = var_2203_cast_fp16, y = linear_33_cast_fp16)[name = string("input_63_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_4_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_4_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161629696)))];
            tensor<fp16, [1, 291, 1024]> linear_34_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_4_mlp_down_proj_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_34_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_79_cast_fp16 = add(x = hidden_states_71_cast_fp16, y = linear_34_cast_fp16)[name = string("hidden_states_79_cast_fp16")];
            tensor<int32, [4]> var_2215 = const()[name = string("op_2215"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2227 = const()[name = string("op_2227"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_2216_cast_fp16 = transpose(perm = var_2215, x = k_29_cast_fp16)[name = string("transpose_151")];
            tensor<fp16, [1, 1024, 1, 291]> input_65_cast_fp16 = reshape(shape = var_2227, x = var_2216_cast_fp16)[name = string("input_65_cast_fp16")];
            tensor<int32, [8]> var_2233_pad_0 = const()[name = string("op_2233_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2233_mode_0 = const()[name = string("op_2233_mode_0"), val = string("constant")];
            fp16 const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2233_cast_fp16 = pad(constant_val = const_84_to_fp16, mode = var_2233_mode_0, pad = var_2233_pad_0, x = input_65_cast_fp16)[name = string("op_2233_cast_fp16")];
            tensor<int32, [4]> var_2238 = const()[name = string("op_2238"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2250 = const()[name = string("op_2250"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_2239_cast_fp16 = transpose(perm = var_2238, x = v_19_cast_fp16)[name = string("transpose_150")];
            tensor<fp16, [1, 1024, 1, 291]> input_67_cast_fp16 = reshape(shape = var_2250, x = var_2239_cast_fp16)[name = string("input_67_cast_fp16")];
            tensor<int32, [8]> var_2256_pad_0 = const()[name = string("op_2256_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2256_mode_0 = const()[name = string("op_2256_mode_0"), val = string("constant")];
            fp16 const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2256_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = var_2256_mode_0, pad = var_2256_pad_0, x = input_67_cast_fp16)[name = string("op_2256_cast_fp16")];
            fp16 var_2260_promoted_to_fp16 = const()[name = string("op_2260_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_2266_cast_fp16 = pow(x = hidden_states_79_cast_fp16, y = var_2260_promoted_to_fp16)[name = string("op_2266_cast_fp16")];
            tensor<int32, [1]> variance_21_axes_0 = const()[name = string("variance_21_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_21_keep_dims_0 = const()[name = string("variance_21_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_21_cast_fp16 = reduce_mean(axes = variance_21_axes_0, keep_dims = variance_21_keep_dims_0, x = var_2266_cast_fp16)[name = string("variance_21_cast_fp16")];
            fp16 var_2269_to_fp16 = const()[name = string("op_2269_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_2270_cast_fp16 = add(x = variance_21_cast_fp16, y = var_2269_to_fp16)[name = string("op_2270_cast_fp16")];
            fp32 var_2271_epsilon_0 = const()[name = string("op_2271_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_2271_cast_fp16 = rsqrt(epsilon = var_2271_epsilon_0, x = var_2270_cast_fp16)[name = string("op_2271_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_83_cast_fp16 = mul(x = hidden_states_79_cast_fp16, y = var_2271_cast_fp16)[name = string("hidden_states_83_cast_fp16")];
            tensor<fp16, [1024]> layers_5_input_layernorm_weight_to_fp16 = const()[name = string("layers_5_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170018368)))];
            tensor<fp16, [1, 291, 1024]> hidden_21_cast_fp16 = mul(x = layers_5_input_layernorm_weight_to_fp16, y = hidden_states_83_cast_fp16)[name = string("hidden_21_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170020480)))];
            tensor<fp16, [1, 291, 1024]> linear_35_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = hidden_21_cast_fp16)[name = string("linear_35_cast_fp16")];
            tensor<int32, [4]> var_2295 = const()[name = string("op_2295"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_31_cast_fp16 = reshape(shape = var_2295, x = linear_35_cast_fp16)[name = string("q_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(172117696)))];
            tensor<fp16, [1, 291, 1024]> linear_36_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = hidden_21_cast_fp16)[name = string("linear_36_cast_fp16")];
            tensor<int32, [4]> var_2302 = const()[name = string("op_2302"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_31_cast_fp16 = reshape(shape = var_2302, x = linear_36_cast_fp16)[name = string("k_31_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(174214912)))];
            tensor<fp16, [1, 291, 1024]> linear_37_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = hidden_21_cast_fp16)[name = string("linear_37_cast_fp16")];
            tensor<int32, [4]> var_2309 = const()[name = string("op_2309"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_21_cast_fp16 = reshape(shape = var_2309, x = linear_37_cast_fp16)[name = string("v_21_cast_fp16")];
            tensor<int32, [4]> q_33_perm_0 = const()[name = string("q_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_33_perm_0 = const()[name = string("k_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_23_perm_0 = const()[name = string("v_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_33_cast_fp16 = transpose(perm = q_33_perm_0, x = q_31_cast_fp16)[name = string("transpose_149")];
            tensor<fp16, [1, 16, 291, 64]> var_2322_cast_fp16 = mul(x = q_33_cast_fp16, y = var_1091_to_fp16)[name = string("op_2322_cast_fp16")];
            tensor<int32, [4]> x1_21_begin_0 = const()[name = string("x1_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_21_end_0 = const()[name = string("x1_21_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_21_end_mask_0 = const()[name = string("x1_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_21_cast_fp16 = slice_by_index(begin = x1_21_begin_0, end = x1_21_end_0, end_mask = x1_21_end_mask_0, x = q_33_cast_fp16)[name = string("x1_21_cast_fp16")];
            tensor<int32, [4]> x2_21_begin_0 = const()[name = string("x2_21_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_21_end_0 = const()[name = string("x2_21_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_21_end_mask_0 = const()[name = string("x2_21_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_21_cast_fp16 = slice_by_index(begin = x2_21_begin_0, end = x2_21_end_0, end_mask = x2_21_end_mask_0, x = q_33_cast_fp16)[name = string("x2_21_cast_fp16")];
            fp16 const_92_promoted_to_fp16 = const()[name = string("const_92_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_2343_cast_fp16 = mul(x = x2_21_cast_fp16, y = const_92_promoted_to_fp16)[name = string("op_2343_cast_fp16")];
            int32 var_2345 = const()[name = string("op_2345"), val = int32(-1)];
            bool var_2346_interleave_0 = const()[name = string("op_2346_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_2346_cast_fp16 = concat(axis = var_2345, interleave = var_2346_interleave_0, values = (var_2343_cast_fp16, x1_21_cast_fp16))[name = string("op_2346_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_2349_cast_fp16 = mul(x = var_2346_cast_fp16, y = var_1118_to_fp16)[name = string("op_2349_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_35_cast_fp16 = add(x = var_2322_cast_fp16, y = var_2349_cast_fp16)[name = string("q_35_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_33_cast_fp16 = transpose(perm = k_33_perm_0, x = k_31_cast_fp16)[name = string("transpose_148")];
            tensor<fp16, [1, 16, 291, 64]> var_2354_cast_fp16 = mul(x = k_33_cast_fp16, y = var_1091_to_fp16)[name = string("op_2354_cast_fp16")];
            tensor<int32, [4]> x1_23_begin_0 = const()[name = string("x1_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_23_end_0 = const()[name = string("x1_23_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_23_end_mask_0 = const()[name = string("x1_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_23_cast_fp16 = slice_by_index(begin = x1_23_begin_0, end = x1_23_end_0, end_mask = x1_23_end_mask_0, x = k_33_cast_fp16)[name = string("x1_23_cast_fp16")];
            tensor<int32, [4]> x2_23_begin_0 = const()[name = string("x2_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_23_end_0 = const()[name = string("x2_23_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_23_end_mask_0 = const()[name = string("x2_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_23_cast_fp16 = slice_by_index(begin = x2_23_begin_0, end = x2_23_end_0, end_mask = x2_23_end_mask_0, x = k_33_cast_fp16)[name = string("x2_23_cast_fp16")];
            fp16 const_95_promoted_to_fp16 = const()[name = string("const_95_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_2375_cast_fp16 = mul(x = x2_23_cast_fp16, y = const_95_promoted_to_fp16)[name = string("op_2375_cast_fp16")];
            int32 var_2377 = const()[name = string("op_2377"), val = int32(-1)];
            bool var_2378_interleave_0 = const()[name = string("op_2378_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_2378_cast_fp16 = concat(axis = var_2377, interleave = var_2378_interleave_0, values = (var_2375_cast_fp16, x1_23_cast_fp16))[name = string("op_2378_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_2381_cast_fp16 = mul(x = var_2378_cast_fp16, y = var_1118_to_fp16)[name = string("op_2381_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_35_cast_fp16 = add(x = var_2354_cast_fp16, y = var_2381_cast_fp16)[name = string("k_35_cast_fp16")];
            bool var_2387_transpose_x_1 = const()[name = string("op_2387_transpose_x_1"), val = bool(false)];
            bool var_2387_transpose_y_1 = const()[name = string("op_2387_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_2387_cast_fp16 = matmul(transpose_x = var_2387_transpose_x_1, transpose_y = var_2387_transpose_y_1, x = q_35_cast_fp16, y = k_35_cast_fp16)[name = string("op_2387_cast_fp16")];
            fp16 _inversed_scores_31_y_0_to_fp16 = const()[name = string("_inversed_scores_31_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_31_cast_fp16 = mul(x = var_2387_cast_fp16, y = _inversed_scores_31_y_0_to_fp16)[name = string("_inversed_scores_31_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_33_cast_fp16 = add(x = _inversed_scores_31_cast_fp16, y = const_21_to_fp16)[name = string("scores_33_cast_fp16")];
            int32 var_2402 = const()[name = string("op_2402"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_2404_cast_fp16 = softmax(axis = var_2402, x = scores_33_cast_fp16)[name = string("op_2404_cast_fp16")];
            bool attn_out_21_transpose_x_0 = const()[name = string("attn_out_21_transpose_x_0"), val = bool(false)];
            bool attn_out_21_transpose_y_0 = const()[name = string("attn_out_21_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_23_cast_fp16 = transpose(perm = v_23_perm_0, x = v_21_cast_fp16)[name = string("transpose_147")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_21_cast_fp16 = matmul(transpose_x = attn_out_21_transpose_x_0, transpose_y = attn_out_21_transpose_y_0, x = var_2404_cast_fp16, y = v_23_cast_fp16)[name = string("attn_out_21_cast_fp16")];
            tensor<int32, [4]> var_2413_perm_0 = const()[name = string("op_2413_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2415 = const()[name = string("op_2415"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_2413_cast_fp16 = transpose(perm = var_2413_perm_0, x = attn_out_21_cast_fp16)[name = string("transpose_146")];
            tensor<fp16, [1, 291, 1024]> input_69_cast_fp16 = reshape(shape = var_2415, x = var_2413_cast_fp16)[name = string("input_69_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176312128)))];
            tensor<fp16, [1, 291, 1024]> linear_38_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_69_cast_fp16)[name = string("linear_38_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_87_cast_fp16 = add(x = hidden_states_79_cast_fp16, y = linear_38_cast_fp16)[name = string("hidden_states_87_cast_fp16")];
            fp16 var_2425_promoted_to_fp16 = const()[name = string("op_2425_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_2431_cast_fp16 = pow(x = hidden_states_87_cast_fp16, y = var_2425_promoted_to_fp16)[name = string("op_2431_cast_fp16")];
            tensor<int32, [1]> variance_23_axes_0 = const()[name = string("variance_23_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_23_keep_dims_0 = const()[name = string("variance_23_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_23_cast_fp16 = reduce_mean(axes = variance_23_axes_0, keep_dims = variance_23_keep_dims_0, x = var_2431_cast_fp16)[name = string("variance_23_cast_fp16")];
            fp16 var_2434_to_fp16 = const()[name = string("op_2434_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_2435_cast_fp16 = add(x = variance_23_cast_fp16, y = var_2434_to_fp16)[name = string("op_2435_cast_fp16")];
            fp32 var_2436_epsilon_0 = const()[name = string("op_2436_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_2436_cast_fp16 = rsqrt(epsilon = var_2436_epsilon_0, x = var_2435_cast_fp16)[name = string("op_2436_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_91_cast_fp16 = mul(x = hidden_states_87_cast_fp16, y = var_2436_cast_fp16)[name = string("hidden_states_91_cast_fp16")];
            tensor<fp16, [1024]> layers_5_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_5_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178409344)))];
            tensor<fp16, [1, 291, 1024]> input_71_cast_fp16 = mul(x = layers_5_post_attention_layernorm_weight_to_fp16, y = hidden_states_91_cast_fp16)[name = string("input_71_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_5_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_5_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178411456)))];
            tensor<fp16, [1, 291, 4096]> linear_39_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_5_mlp_gate_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_39_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_2449_cast_fp16 = silu(x = linear_39_cast_fp16)[name = string("op_2449_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_5_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_5_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186800128)))];
            tensor<fp16, [1, 291, 4096]> linear_40_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_5_mlp_up_proj_weight_to_fp16, x = input_71_cast_fp16)[name = string("linear_40_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_75_cast_fp16 = mul(x = var_2449_cast_fp16, y = linear_40_cast_fp16)[name = string("input_75_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_5_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_5_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195188800)))];
            tensor<fp16, [1, 291, 1024]> linear_41_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_5_mlp_down_proj_weight_to_fp16, x = input_75_cast_fp16)[name = string("linear_41_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_95_cast_fp16 = add(x = hidden_states_87_cast_fp16, y = linear_41_cast_fp16)[name = string("hidden_states_95_cast_fp16")];
            tensor<int32, [4]> var_2461 = const()[name = string("op_2461"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2473 = const()[name = string("op_2473"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_2462_cast_fp16 = transpose(perm = var_2461, x = k_35_cast_fp16)[name = string("transpose_145")];
            tensor<fp16, [1, 1024, 1, 291]> input_77_cast_fp16 = reshape(shape = var_2473, x = var_2462_cast_fp16)[name = string("input_77_cast_fp16")];
            tensor<int32, [8]> var_2479_pad_0 = const()[name = string("op_2479_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2479_mode_0 = const()[name = string("op_2479_mode_0"), val = string("constant")];
            fp16 const_99_to_fp16 = const()[name = string("const_99_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2479_cast_fp16 = pad(constant_val = const_99_to_fp16, mode = var_2479_mode_0, pad = var_2479_pad_0, x = input_77_cast_fp16)[name = string("op_2479_cast_fp16")];
            tensor<int32, [4]> var_2484 = const()[name = string("op_2484"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2496 = const()[name = string("op_2496"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_2485_cast_fp16 = transpose(perm = var_2484, x = v_23_cast_fp16)[name = string("transpose_144")];
            tensor<fp16, [1, 1024, 1, 291]> input_79_cast_fp16 = reshape(shape = var_2496, x = var_2485_cast_fp16)[name = string("input_79_cast_fp16")];
            tensor<int32, [8]> var_2502_pad_0 = const()[name = string("op_2502_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2502_mode_0 = const()[name = string("op_2502_mode_0"), val = string("constant")];
            fp16 const_102_to_fp16 = const()[name = string("const_102_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2502_cast_fp16 = pad(constant_val = const_102_to_fp16, mode = var_2502_mode_0, pad = var_2502_pad_0, x = input_79_cast_fp16)[name = string("op_2502_cast_fp16")];
            fp16 var_2506_promoted_to_fp16 = const()[name = string("op_2506_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_2512_cast_fp16 = pow(x = hidden_states_95_cast_fp16, y = var_2506_promoted_to_fp16)[name = string("op_2512_cast_fp16")];
            tensor<int32, [1]> variance_25_axes_0 = const()[name = string("variance_25_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_25_keep_dims_0 = const()[name = string("variance_25_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_25_cast_fp16 = reduce_mean(axes = variance_25_axes_0, keep_dims = variance_25_keep_dims_0, x = var_2512_cast_fp16)[name = string("variance_25_cast_fp16")];
            fp16 var_2515_to_fp16 = const()[name = string("op_2515_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_2516_cast_fp16 = add(x = variance_25_cast_fp16, y = var_2515_to_fp16)[name = string("op_2516_cast_fp16")];
            fp32 var_2517_epsilon_0 = const()[name = string("op_2517_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_2517_cast_fp16 = rsqrt(epsilon = var_2517_epsilon_0, x = var_2516_cast_fp16)[name = string("op_2517_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_99_cast_fp16 = mul(x = hidden_states_95_cast_fp16, y = var_2517_cast_fp16)[name = string("hidden_states_99_cast_fp16")];
            tensor<fp16, [1024]> layers_6_input_layernorm_weight_to_fp16 = const()[name = string("layers_6_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203577472)))];
            tensor<fp16, [1, 291, 1024]> hidden_25_cast_fp16 = mul(x = layers_6_input_layernorm_weight_to_fp16, y = hidden_states_99_cast_fp16)[name = string("hidden_25_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203579584)))];
            tensor<fp16, [1, 291, 1024]> linear_42_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_6_self_attn_q_proj_weight_to_fp16, x = hidden_25_cast_fp16)[name = string("linear_42_cast_fp16")];
            tensor<int32, [4]> var_2541 = const()[name = string("op_2541"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_37_cast_fp16 = reshape(shape = var_2541, x = linear_42_cast_fp16)[name = string("q_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(205676800)))];
            tensor<fp16, [1, 291, 1024]> linear_43_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_6_self_attn_k_proj_weight_to_fp16, x = hidden_25_cast_fp16)[name = string("linear_43_cast_fp16")];
            tensor<int32, [4]> var_2548 = const()[name = string("op_2548"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_37_cast_fp16 = reshape(shape = var_2548, x = linear_43_cast_fp16)[name = string("k_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(207774016)))];
            tensor<fp16, [1, 291, 1024]> linear_44_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_6_self_attn_v_proj_weight_to_fp16, x = hidden_25_cast_fp16)[name = string("linear_44_cast_fp16")];
            tensor<int32, [4]> var_2555 = const()[name = string("op_2555"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_25_cast_fp16 = reshape(shape = var_2555, x = linear_44_cast_fp16)[name = string("v_25_cast_fp16")];
            tensor<int32, [4]> q_39_perm_0 = const()[name = string("q_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_39_perm_0 = const()[name = string("k_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_27_perm_0 = const()[name = string("v_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_39_cast_fp16 = transpose(perm = q_39_perm_0, x = q_37_cast_fp16)[name = string("transpose_143")];
            tensor<fp16, [1, 16, 291, 64]> var_2568_cast_fp16 = mul(x = q_39_cast_fp16, y = var_1091_to_fp16)[name = string("op_2568_cast_fp16")];
            tensor<int32, [4]> x1_25_begin_0 = const()[name = string("x1_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_25_end_0 = const()[name = string("x1_25_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_25_end_mask_0 = const()[name = string("x1_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_25_cast_fp16 = slice_by_index(begin = x1_25_begin_0, end = x1_25_end_0, end_mask = x1_25_end_mask_0, x = q_39_cast_fp16)[name = string("x1_25_cast_fp16")];
            tensor<int32, [4]> x2_25_begin_0 = const()[name = string("x2_25_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_25_end_0 = const()[name = string("x2_25_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_25_end_mask_0 = const()[name = string("x2_25_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_25_cast_fp16 = slice_by_index(begin = x2_25_begin_0, end = x2_25_end_0, end_mask = x2_25_end_mask_0, x = q_39_cast_fp16)[name = string("x2_25_cast_fp16")];
            fp16 const_107_promoted_to_fp16 = const()[name = string("const_107_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_2589_cast_fp16 = mul(x = x2_25_cast_fp16, y = const_107_promoted_to_fp16)[name = string("op_2589_cast_fp16")];
            int32 var_2591 = const()[name = string("op_2591"), val = int32(-1)];
            bool var_2592_interleave_0 = const()[name = string("op_2592_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_2592_cast_fp16 = concat(axis = var_2591, interleave = var_2592_interleave_0, values = (var_2589_cast_fp16, x1_25_cast_fp16))[name = string("op_2592_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_2595_cast_fp16 = mul(x = var_2592_cast_fp16, y = var_1118_to_fp16)[name = string("op_2595_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_41_cast_fp16 = add(x = var_2568_cast_fp16, y = var_2595_cast_fp16)[name = string("q_41_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_39_cast_fp16 = transpose(perm = k_39_perm_0, x = k_37_cast_fp16)[name = string("transpose_142")];
            tensor<fp16, [1, 16, 291, 64]> var_2600_cast_fp16 = mul(x = k_39_cast_fp16, y = var_1091_to_fp16)[name = string("op_2600_cast_fp16")];
            tensor<int32, [4]> x1_27_begin_0 = const()[name = string("x1_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_27_end_0 = const()[name = string("x1_27_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_27_end_mask_0 = const()[name = string("x1_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_27_cast_fp16 = slice_by_index(begin = x1_27_begin_0, end = x1_27_end_0, end_mask = x1_27_end_mask_0, x = k_39_cast_fp16)[name = string("x1_27_cast_fp16")];
            tensor<int32, [4]> x2_27_begin_0 = const()[name = string("x2_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_27_end_0 = const()[name = string("x2_27_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_27_end_mask_0 = const()[name = string("x2_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_27_cast_fp16 = slice_by_index(begin = x2_27_begin_0, end = x2_27_end_0, end_mask = x2_27_end_mask_0, x = k_39_cast_fp16)[name = string("x2_27_cast_fp16")];
            fp16 const_110_promoted_to_fp16 = const()[name = string("const_110_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_2621_cast_fp16 = mul(x = x2_27_cast_fp16, y = const_110_promoted_to_fp16)[name = string("op_2621_cast_fp16")];
            int32 var_2623 = const()[name = string("op_2623"), val = int32(-1)];
            bool var_2624_interleave_0 = const()[name = string("op_2624_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_2624_cast_fp16 = concat(axis = var_2623, interleave = var_2624_interleave_0, values = (var_2621_cast_fp16, x1_27_cast_fp16))[name = string("op_2624_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_2627_cast_fp16 = mul(x = var_2624_cast_fp16, y = var_1118_to_fp16)[name = string("op_2627_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_41_cast_fp16 = add(x = var_2600_cast_fp16, y = var_2627_cast_fp16)[name = string("k_41_cast_fp16")];
            bool var_2633_transpose_x_1 = const()[name = string("op_2633_transpose_x_1"), val = bool(false)];
            bool var_2633_transpose_y_1 = const()[name = string("op_2633_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_2633_cast_fp16 = matmul(transpose_x = var_2633_transpose_x_1, transpose_y = var_2633_transpose_y_1, x = q_41_cast_fp16, y = k_41_cast_fp16)[name = string("op_2633_cast_fp16")];
            fp16 _inversed_scores_37_y_0_to_fp16 = const()[name = string("_inversed_scores_37_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_37_cast_fp16 = mul(x = var_2633_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = string("_inversed_scores_37_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_39_cast_fp16 = add(x = _inversed_scores_37_cast_fp16, y = const_21_to_fp16)[name = string("scores_39_cast_fp16")];
            int32 var_2648 = const()[name = string("op_2648"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_2650_cast_fp16 = softmax(axis = var_2648, x = scores_39_cast_fp16)[name = string("op_2650_cast_fp16")];
            bool attn_out_25_transpose_x_0 = const()[name = string("attn_out_25_transpose_x_0"), val = bool(false)];
            bool attn_out_25_transpose_y_0 = const()[name = string("attn_out_25_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_27_cast_fp16 = transpose(perm = v_27_perm_0, x = v_25_cast_fp16)[name = string("transpose_141")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_25_cast_fp16 = matmul(transpose_x = attn_out_25_transpose_x_0, transpose_y = attn_out_25_transpose_y_0, x = var_2650_cast_fp16, y = v_27_cast_fp16)[name = string("attn_out_25_cast_fp16")];
            tensor<int32, [4]> var_2659_perm_0 = const()[name = string("op_2659_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2661 = const()[name = string("op_2661"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_2659_cast_fp16 = transpose(perm = var_2659_perm_0, x = attn_out_25_cast_fp16)[name = string("transpose_140")];
            tensor<fp16, [1, 291, 1024]> input_81_cast_fp16 = reshape(shape = var_2661, x = var_2659_cast_fp16)[name = string("input_81_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_6_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_6_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(209871232)))];
            tensor<fp16, [1, 291, 1024]> linear_45_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_6_self_attn_o_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_45_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_103_cast_fp16 = add(x = hidden_states_95_cast_fp16, y = linear_45_cast_fp16)[name = string("hidden_states_103_cast_fp16")];
            fp16 var_2671_promoted_to_fp16 = const()[name = string("op_2671_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_2677_cast_fp16 = pow(x = hidden_states_103_cast_fp16, y = var_2671_promoted_to_fp16)[name = string("op_2677_cast_fp16")];
            tensor<int32, [1]> variance_27_axes_0 = const()[name = string("variance_27_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_27_keep_dims_0 = const()[name = string("variance_27_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_27_cast_fp16 = reduce_mean(axes = variance_27_axes_0, keep_dims = variance_27_keep_dims_0, x = var_2677_cast_fp16)[name = string("variance_27_cast_fp16")];
            fp16 var_2680_to_fp16 = const()[name = string("op_2680_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_2681_cast_fp16 = add(x = variance_27_cast_fp16, y = var_2680_to_fp16)[name = string("op_2681_cast_fp16")];
            fp32 var_2682_epsilon_0 = const()[name = string("op_2682_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_2682_cast_fp16 = rsqrt(epsilon = var_2682_epsilon_0, x = var_2681_cast_fp16)[name = string("op_2682_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_107_cast_fp16 = mul(x = hidden_states_103_cast_fp16, y = var_2682_cast_fp16)[name = string("hidden_states_107_cast_fp16")];
            tensor<fp16, [1024]> layers_6_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_6_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211968448)))];
            tensor<fp16, [1, 291, 1024]> input_83_cast_fp16 = mul(x = layers_6_post_attention_layernorm_weight_to_fp16, y = hidden_states_107_cast_fp16)[name = string("input_83_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_6_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_6_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211970560)))];
            tensor<fp16, [1, 291, 4096]> linear_46_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_6_mlp_gate_proj_weight_to_fp16, x = input_83_cast_fp16)[name = string("linear_46_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_2695_cast_fp16 = silu(x = linear_46_cast_fp16)[name = string("op_2695_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_6_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_6_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(220359232)))];
            tensor<fp16, [1, 291, 4096]> linear_47_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_6_mlp_up_proj_weight_to_fp16, x = input_83_cast_fp16)[name = string("linear_47_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_87_cast_fp16 = mul(x = var_2695_cast_fp16, y = linear_47_cast_fp16)[name = string("input_87_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_6_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_6_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228747904)))];
            tensor<fp16, [1, 291, 1024]> linear_48_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_6_mlp_down_proj_weight_to_fp16, x = input_87_cast_fp16)[name = string("linear_48_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_111_cast_fp16 = add(x = hidden_states_103_cast_fp16, y = linear_48_cast_fp16)[name = string("hidden_states_111_cast_fp16")];
            tensor<int32, [4]> var_2707 = const()[name = string("op_2707"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2719 = const()[name = string("op_2719"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_2708_cast_fp16 = transpose(perm = var_2707, x = k_41_cast_fp16)[name = string("transpose_139")];
            tensor<fp16, [1, 1024, 1, 291]> input_89_cast_fp16 = reshape(shape = var_2719, x = var_2708_cast_fp16)[name = string("input_89_cast_fp16")];
            tensor<int32, [8]> var_2725_pad_0 = const()[name = string("op_2725_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2725_mode_0 = const()[name = string("op_2725_mode_0"), val = string("constant")];
            fp16 const_114_to_fp16 = const()[name = string("const_114_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2725_cast_fp16 = pad(constant_val = const_114_to_fp16, mode = var_2725_mode_0, pad = var_2725_pad_0, x = input_89_cast_fp16)[name = string("op_2725_cast_fp16")];
            tensor<int32, [4]> var_2730 = const()[name = string("op_2730"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2742 = const()[name = string("op_2742"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_2731_cast_fp16 = transpose(perm = var_2730, x = v_27_cast_fp16)[name = string("transpose_138")];
            tensor<fp16, [1, 1024, 1, 291]> input_91_cast_fp16 = reshape(shape = var_2742, x = var_2731_cast_fp16)[name = string("input_91_cast_fp16")];
            tensor<int32, [8]> var_2748_pad_0 = const()[name = string("op_2748_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2748_mode_0 = const()[name = string("op_2748_mode_0"), val = string("constant")];
            fp16 const_117_to_fp16 = const()[name = string("const_117_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2748_cast_fp16 = pad(constant_val = const_117_to_fp16, mode = var_2748_mode_0, pad = var_2748_pad_0, x = input_91_cast_fp16)[name = string("op_2748_cast_fp16")];
            fp16 var_2752_promoted_to_fp16 = const()[name = string("op_2752_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_2758_cast_fp16 = pow(x = hidden_states_111_cast_fp16, y = var_2752_promoted_to_fp16)[name = string("op_2758_cast_fp16")];
            tensor<int32, [1]> variance_29_axes_0 = const()[name = string("variance_29_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_29_keep_dims_0 = const()[name = string("variance_29_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_29_cast_fp16 = reduce_mean(axes = variance_29_axes_0, keep_dims = variance_29_keep_dims_0, x = var_2758_cast_fp16)[name = string("variance_29_cast_fp16")];
            fp16 var_2761_to_fp16 = const()[name = string("op_2761_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_2762_cast_fp16 = add(x = variance_29_cast_fp16, y = var_2761_to_fp16)[name = string("op_2762_cast_fp16")];
            fp32 var_2763_epsilon_0 = const()[name = string("op_2763_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_2763_cast_fp16 = rsqrt(epsilon = var_2763_epsilon_0, x = var_2762_cast_fp16)[name = string("op_2763_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_115_cast_fp16 = mul(x = hidden_states_111_cast_fp16, y = var_2763_cast_fp16)[name = string("hidden_states_115_cast_fp16")];
            tensor<fp16, [1024]> layers_7_input_layernorm_weight_to_fp16 = const()[name = string("layers_7_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237136576)))];
            tensor<fp16, [1, 291, 1024]> hidden_29_cast_fp16 = mul(x = layers_7_input_layernorm_weight_to_fp16, y = hidden_states_115_cast_fp16)[name = string("hidden_29_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237138688)))];
            tensor<fp16, [1, 291, 1024]> linear_49_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_7_self_attn_q_proj_weight_to_fp16, x = hidden_29_cast_fp16)[name = string("linear_49_cast_fp16")];
            tensor<int32, [4]> var_2787 = const()[name = string("op_2787"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_43_cast_fp16 = reshape(shape = var_2787, x = linear_49_cast_fp16)[name = string("q_43_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239235904)))];
            tensor<fp16, [1, 291, 1024]> linear_50_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_7_self_attn_k_proj_weight_to_fp16, x = hidden_29_cast_fp16)[name = string("linear_50_cast_fp16")];
            tensor<int32, [4]> var_2794 = const()[name = string("op_2794"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_43_cast_fp16 = reshape(shape = var_2794, x = linear_50_cast_fp16)[name = string("k_43_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(241333120)))];
            tensor<fp16, [1, 291, 1024]> linear_51_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_7_self_attn_v_proj_weight_to_fp16, x = hidden_29_cast_fp16)[name = string("linear_51_cast_fp16")];
            tensor<int32, [4]> var_2801 = const()[name = string("op_2801"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_29_cast_fp16 = reshape(shape = var_2801, x = linear_51_cast_fp16)[name = string("v_29_cast_fp16")];
            tensor<int32, [4]> q_45_perm_0 = const()[name = string("q_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_45_perm_0 = const()[name = string("k_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_31_perm_0 = const()[name = string("v_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_45_cast_fp16 = transpose(perm = q_45_perm_0, x = q_43_cast_fp16)[name = string("transpose_137")];
            tensor<fp16, [1, 16, 291, 64]> var_2814_cast_fp16 = mul(x = q_45_cast_fp16, y = var_1091_to_fp16)[name = string("op_2814_cast_fp16")];
            tensor<int32, [4]> x1_29_begin_0 = const()[name = string("x1_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_29_end_0 = const()[name = string("x1_29_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_29_end_mask_0 = const()[name = string("x1_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_29_cast_fp16 = slice_by_index(begin = x1_29_begin_0, end = x1_29_end_0, end_mask = x1_29_end_mask_0, x = q_45_cast_fp16)[name = string("x1_29_cast_fp16")];
            tensor<int32, [4]> x2_29_begin_0 = const()[name = string("x2_29_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_29_end_0 = const()[name = string("x2_29_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_29_end_mask_0 = const()[name = string("x2_29_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_29_cast_fp16 = slice_by_index(begin = x2_29_begin_0, end = x2_29_end_0, end_mask = x2_29_end_mask_0, x = q_45_cast_fp16)[name = string("x2_29_cast_fp16")];
            fp16 const_122_promoted_to_fp16 = const()[name = string("const_122_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_2835_cast_fp16 = mul(x = x2_29_cast_fp16, y = const_122_promoted_to_fp16)[name = string("op_2835_cast_fp16")];
            int32 var_2837 = const()[name = string("op_2837"), val = int32(-1)];
            bool var_2838_interleave_0 = const()[name = string("op_2838_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_2838_cast_fp16 = concat(axis = var_2837, interleave = var_2838_interleave_0, values = (var_2835_cast_fp16, x1_29_cast_fp16))[name = string("op_2838_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_2841_cast_fp16 = mul(x = var_2838_cast_fp16, y = var_1118_to_fp16)[name = string("op_2841_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_47_cast_fp16 = add(x = var_2814_cast_fp16, y = var_2841_cast_fp16)[name = string("q_47_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_45_cast_fp16 = transpose(perm = k_45_perm_0, x = k_43_cast_fp16)[name = string("transpose_136")];
            tensor<fp16, [1, 16, 291, 64]> var_2846_cast_fp16 = mul(x = k_45_cast_fp16, y = var_1091_to_fp16)[name = string("op_2846_cast_fp16")];
            tensor<int32, [4]> x1_31_begin_0 = const()[name = string("x1_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_31_end_0 = const()[name = string("x1_31_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_31_end_mask_0 = const()[name = string("x1_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_31_cast_fp16 = slice_by_index(begin = x1_31_begin_0, end = x1_31_end_0, end_mask = x1_31_end_mask_0, x = k_45_cast_fp16)[name = string("x1_31_cast_fp16")];
            tensor<int32, [4]> x2_31_begin_0 = const()[name = string("x2_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_31_end_0 = const()[name = string("x2_31_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_31_end_mask_0 = const()[name = string("x2_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_31_cast_fp16 = slice_by_index(begin = x2_31_begin_0, end = x2_31_end_0, end_mask = x2_31_end_mask_0, x = k_45_cast_fp16)[name = string("x2_31_cast_fp16")];
            fp16 const_125_promoted_to_fp16 = const()[name = string("const_125_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_2867_cast_fp16 = mul(x = x2_31_cast_fp16, y = const_125_promoted_to_fp16)[name = string("op_2867_cast_fp16")];
            int32 var_2869 = const()[name = string("op_2869"), val = int32(-1)];
            bool var_2870_interleave_0 = const()[name = string("op_2870_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_2870_cast_fp16 = concat(axis = var_2869, interleave = var_2870_interleave_0, values = (var_2867_cast_fp16, x1_31_cast_fp16))[name = string("op_2870_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_2873_cast_fp16 = mul(x = var_2870_cast_fp16, y = var_1118_to_fp16)[name = string("op_2873_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_47_cast_fp16 = add(x = var_2846_cast_fp16, y = var_2873_cast_fp16)[name = string("k_47_cast_fp16")];
            bool var_2879_transpose_x_1 = const()[name = string("op_2879_transpose_x_1"), val = bool(false)];
            bool var_2879_transpose_y_1 = const()[name = string("op_2879_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_2879_cast_fp16 = matmul(transpose_x = var_2879_transpose_x_1, transpose_y = var_2879_transpose_y_1, x = q_47_cast_fp16, y = k_47_cast_fp16)[name = string("op_2879_cast_fp16")];
            fp16 _inversed_scores_43_y_0_to_fp16 = const()[name = string("_inversed_scores_43_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_43_cast_fp16 = mul(x = var_2879_cast_fp16, y = _inversed_scores_43_y_0_to_fp16)[name = string("_inversed_scores_43_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_45_cast_fp16 = add(x = _inversed_scores_43_cast_fp16, y = const_21_to_fp16)[name = string("scores_45_cast_fp16")];
            int32 var_2894 = const()[name = string("op_2894"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_2896_cast_fp16 = softmax(axis = var_2894, x = scores_45_cast_fp16)[name = string("op_2896_cast_fp16")];
            bool attn_out_29_transpose_x_0 = const()[name = string("attn_out_29_transpose_x_0"), val = bool(false)];
            bool attn_out_29_transpose_y_0 = const()[name = string("attn_out_29_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_31_cast_fp16 = transpose(perm = v_31_perm_0, x = v_29_cast_fp16)[name = string("transpose_135")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_29_cast_fp16 = matmul(transpose_x = attn_out_29_transpose_x_0, transpose_y = attn_out_29_transpose_y_0, x = var_2896_cast_fp16, y = v_31_cast_fp16)[name = string("attn_out_29_cast_fp16")];
            tensor<int32, [4]> var_2905_perm_0 = const()[name = string("op_2905_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2907 = const()[name = string("op_2907"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_2905_cast_fp16 = transpose(perm = var_2905_perm_0, x = attn_out_29_cast_fp16)[name = string("transpose_134")];
            tensor<fp16, [1, 291, 1024]> input_93_cast_fp16 = reshape(shape = var_2907, x = var_2905_cast_fp16)[name = string("input_93_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_7_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_7_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(243430336)))];
            tensor<fp16, [1, 291, 1024]> linear_52_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_7_self_attn_o_proj_weight_to_fp16, x = input_93_cast_fp16)[name = string("linear_52_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_119_cast_fp16 = add(x = hidden_states_111_cast_fp16, y = linear_52_cast_fp16)[name = string("hidden_states_119_cast_fp16")];
            fp16 var_2917_promoted_to_fp16 = const()[name = string("op_2917_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_2923_cast_fp16 = pow(x = hidden_states_119_cast_fp16, y = var_2917_promoted_to_fp16)[name = string("op_2923_cast_fp16")];
            tensor<int32, [1]> variance_31_axes_0 = const()[name = string("variance_31_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_31_keep_dims_0 = const()[name = string("variance_31_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_31_cast_fp16 = reduce_mean(axes = variance_31_axes_0, keep_dims = variance_31_keep_dims_0, x = var_2923_cast_fp16)[name = string("variance_31_cast_fp16")];
            fp16 var_2926_to_fp16 = const()[name = string("op_2926_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_2927_cast_fp16 = add(x = variance_31_cast_fp16, y = var_2926_to_fp16)[name = string("op_2927_cast_fp16")];
            fp32 var_2928_epsilon_0 = const()[name = string("op_2928_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_2928_cast_fp16 = rsqrt(epsilon = var_2928_epsilon_0, x = var_2927_cast_fp16)[name = string("op_2928_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_123_cast_fp16 = mul(x = hidden_states_119_cast_fp16, y = var_2928_cast_fp16)[name = string("hidden_states_123_cast_fp16")];
            tensor<fp16, [1024]> layers_7_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_7_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245527552)))];
            tensor<fp16, [1, 291, 1024]> input_95_cast_fp16 = mul(x = layers_7_post_attention_layernorm_weight_to_fp16, y = hidden_states_123_cast_fp16)[name = string("input_95_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_7_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_7_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245529664)))];
            tensor<fp16, [1, 291, 4096]> linear_53_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_7_mlp_gate_proj_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_53_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_2941_cast_fp16 = silu(x = linear_53_cast_fp16)[name = string("op_2941_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_7_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_7_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253918336)))];
            tensor<fp16, [1, 291, 4096]> linear_54_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_7_mlp_up_proj_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_54_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_99_cast_fp16 = mul(x = var_2941_cast_fp16, y = linear_54_cast_fp16)[name = string("input_99_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_7_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_7_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(262307008)))];
            tensor<fp16, [1, 291, 1024]> linear_55_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_7_mlp_down_proj_weight_to_fp16, x = input_99_cast_fp16)[name = string("linear_55_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_127_cast_fp16 = add(x = hidden_states_119_cast_fp16, y = linear_55_cast_fp16)[name = string("hidden_states_127_cast_fp16")];
            tensor<int32, [4]> var_2953 = const()[name = string("op_2953"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2965 = const()[name = string("op_2965"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_2954_cast_fp16 = transpose(perm = var_2953, x = k_47_cast_fp16)[name = string("transpose_133")];
            tensor<fp16, [1, 1024, 1, 291]> input_101_cast_fp16 = reshape(shape = var_2965, x = var_2954_cast_fp16)[name = string("input_101_cast_fp16")];
            tensor<int32, [8]> var_2971_pad_0 = const()[name = string("op_2971_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2971_mode_0 = const()[name = string("op_2971_mode_0"), val = string("constant")];
            fp16 const_129_to_fp16 = const()[name = string("const_129_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2971_cast_fp16 = pad(constant_val = const_129_to_fp16, mode = var_2971_mode_0, pad = var_2971_pad_0, x = input_101_cast_fp16)[name = string("op_2971_cast_fp16")];
            tensor<int32, [4]> var_2976 = const()[name = string("op_2976"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_2988 = const()[name = string("op_2988"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_2977_cast_fp16 = transpose(perm = var_2976, x = v_31_cast_fp16)[name = string("transpose_132")];
            tensor<fp16, [1, 1024, 1, 291]> input_103_cast_fp16 = reshape(shape = var_2988, x = var_2977_cast_fp16)[name = string("input_103_cast_fp16")];
            tensor<int32, [8]> var_2994_pad_0 = const()[name = string("op_2994_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_2994_mode_0 = const()[name = string("op_2994_mode_0"), val = string("constant")];
            fp16 const_132_to_fp16 = const()[name = string("const_132_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_2994_cast_fp16 = pad(constant_val = const_132_to_fp16, mode = var_2994_mode_0, pad = var_2994_pad_0, x = input_103_cast_fp16)[name = string("op_2994_cast_fp16")];
            fp16 var_2998_promoted_to_fp16 = const()[name = string("op_2998_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3004_cast_fp16 = pow(x = hidden_states_127_cast_fp16, y = var_2998_promoted_to_fp16)[name = string("op_3004_cast_fp16")];
            tensor<int32, [1]> variance_33_axes_0 = const()[name = string("variance_33_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_33_keep_dims_0 = const()[name = string("variance_33_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_33_cast_fp16 = reduce_mean(axes = variance_33_axes_0, keep_dims = variance_33_keep_dims_0, x = var_3004_cast_fp16)[name = string("variance_33_cast_fp16")];
            fp16 var_3007_to_fp16 = const()[name = string("op_3007_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3008_cast_fp16 = add(x = variance_33_cast_fp16, y = var_3007_to_fp16)[name = string("op_3008_cast_fp16")];
            fp32 var_3009_epsilon_0 = const()[name = string("op_3009_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3009_cast_fp16 = rsqrt(epsilon = var_3009_epsilon_0, x = var_3008_cast_fp16)[name = string("op_3009_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_131_cast_fp16 = mul(x = hidden_states_127_cast_fp16, y = var_3009_cast_fp16)[name = string("hidden_states_131_cast_fp16")];
            tensor<fp16, [1024]> layers_8_input_layernorm_weight_to_fp16 = const()[name = string("layers_8_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270695680)))];
            tensor<fp16, [1, 291, 1024]> hidden_33_cast_fp16 = mul(x = layers_8_input_layernorm_weight_to_fp16, y = hidden_states_131_cast_fp16)[name = string("hidden_33_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270697792)))];
            tensor<fp16, [1, 291, 1024]> linear_56_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_8_self_attn_q_proj_weight_to_fp16, x = hidden_33_cast_fp16)[name = string("linear_56_cast_fp16")];
            tensor<int32, [4]> var_3033 = const()[name = string("op_3033"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_49_cast_fp16 = reshape(shape = var_3033, x = linear_56_cast_fp16)[name = string("q_49_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272795008)))];
            tensor<fp16, [1, 291, 1024]> linear_57_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_8_self_attn_k_proj_weight_to_fp16, x = hidden_33_cast_fp16)[name = string("linear_57_cast_fp16")];
            tensor<int32, [4]> var_3040 = const()[name = string("op_3040"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_49_cast_fp16 = reshape(shape = var_3040, x = linear_57_cast_fp16)[name = string("k_49_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274892224)))];
            tensor<fp16, [1, 291, 1024]> linear_58_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_8_self_attn_v_proj_weight_to_fp16, x = hidden_33_cast_fp16)[name = string("linear_58_cast_fp16")];
            tensor<int32, [4]> var_3047 = const()[name = string("op_3047"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_33_cast_fp16 = reshape(shape = var_3047, x = linear_58_cast_fp16)[name = string("v_33_cast_fp16")];
            tensor<int32, [4]> q_51_perm_0 = const()[name = string("q_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_51_perm_0 = const()[name = string("k_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_35_perm_0 = const()[name = string("v_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_51_cast_fp16 = transpose(perm = q_51_perm_0, x = q_49_cast_fp16)[name = string("transpose_131")];
            tensor<fp16, [1, 16, 291, 64]> var_3060_cast_fp16 = mul(x = q_51_cast_fp16, y = var_1091_to_fp16)[name = string("op_3060_cast_fp16")];
            tensor<int32, [4]> x1_33_begin_0 = const()[name = string("x1_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_33_end_0 = const()[name = string("x1_33_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_33_end_mask_0 = const()[name = string("x1_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_33_cast_fp16 = slice_by_index(begin = x1_33_begin_0, end = x1_33_end_0, end_mask = x1_33_end_mask_0, x = q_51_cast_fp16)[name = string("x1_33_cast_fp16")];
            tensor<int32, [4]> x2_33_begin_0 = const()[name = string("x2_33_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_33_end_0 = const()[name = string("x2_33_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_33_end_mask_0 = const()[name = string("x2_33_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_33_cast_fp16 = slice_by_index(begin = x2_33_begin_0, end = x2_33_end_0, end_mask = x2_33_end_mask_0, x = q_51_cast_fp16)[name = string("x2_33_cast_fp16")];
            fp16 const_137_promoted_to_fp16 = const()[name = string("const_137_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_3081_cast_fp16 = mul(x = x2_33_cast_fp16, y = const_137_promoted_to_fp16)[name = string("op_3081_cast_fp16")];
            int32 var_3083 = const()[name = string("op_3083"), val = int32(-1)];
            bool var_3084_interleave_0 = const()[name = string("op_3084_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_3084_cast_fp16 = concat(axis = var_3083, interleave = var_3084_interleave_0, values = (var_3081_cast_fp16, x1_33_cast_fp16))[name = string("op_3084_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_3087_cast_fp16 = mul(x = var_3084_cast_fp16, y = var_1118_to_fp16)[name = string("op_3087_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_53_cast_fp16 = add(x = var_3060_cast_fp16, y = var_3087_cast_fp16)[name = string("q_53_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_51_cast_fp16 = transpose(perm = k_51_perm_0, x = k_49_cast_fp16)[name = string("transpose_130")];
            tensor<fp16, [1, 16, 291, 64]> var_3092_cast_fp16 = mul(x = k_51_cast_fp16, y = var_1091_to_fp16)[name = string("op_3092_cast_fp16")];
            tensor<int32, [4]> x1_35_begin_0 = const()[name = string("x1_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_35_end_0 = const()[name = string("x1_35_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_35_end_mask_0 = const()[name = string("x1_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_35_cast_fp16 = slice_by_index(begin = x1_35_begin_0, end = x1_35_end_0, end_mask = x1_35_end_mask_0, x = k_51_cast_fp16)[name = string("x1_35_cast_fp16")];
            tensor<int32, [4]> x2_35_begin_0 = const()[name = string("x2_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_35_end_0 = const()[name = string("x2_35_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_35_end_mask_0 = const()[name = string("x2_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_35_cast_fp16 = slice_by_index(begin = x2_35_begin_0, end = x2_35_end_0, end_mask = x2_35_end_mask_0, x = k_51_cast_fp16)[name = string("x2_35_cast_fp16")];
            fp16 const_140_promoted_to_fp16 = const()[name = string("const_140_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_3113_cast_fp16 = mul(x = x2_35_cast_fp16, y = const_140_promoted_to_fp16)[name = string("op_3113_cast_fp16")];
            int32 var_3115 = const()[name = string("op_3115"), val = int32(-1)];
            bool var_3116_interleave_0 = const()[name = string("op_3116_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_3116_cast_fp16 = concat(axis = var_3115, interleave = var_3116_interleave_0, values = (var_3113_cast_fp16, x1_35_cast_fp16))[name = string("op_3116_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_3119_cast_fp16 = mul(x = var_3116_cast_fp16, y = var_1118_to_fp16)[name = string("op_3119_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_53_cast_fp16 = add(x = var_3092_cast_fp16, y = var_3119_cast_fp16)[name = string("k_53_cast_fp16")];
            bool var_3125_transpose_x_1 = const()[name = string("op_3125_transpose_x_1"), val = bool(false)];
            bool var_3125_transpose_y_1 = const()[name = string("op_3125_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_3125_cast_fp16 = matmul(transpose_x = var_3125_transpose_x_1, transpose_y = var_3125_transpose_y_1, x = q_53_cast_fp16, y = k_53_cast_fp16)[name = string("op_3125_cast_fp16")];
            fp16 _inversed_scores_49_y_0_to_fp16 = const()[name = string("_inversed_scores_49_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_49_cast_fp16 = mul(x = var_3125_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = string("_inversed_scores_49_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_51_cast_fp16 = add(x = _inversed_scores_49_cast_fp16, y = const_21_to_fp16)[name = string("scores_51_cast_fp16")];
            int32 var_3140 = const()[name = string("op_3140"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_3142_cast_fp16 = softmax(axis = var_3140, x = scores_51_cast_fp16)[name = string("op_3142_cast_fp16")];
            bool attn_out_33_transpose_x_0 = const()[name = string("attn_out_33_transpose_x_0"), val = bool(false)];
            bool attn_out_33_transpose_y_0 = const()[name = string("attn_out_33_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_35_cast_fp16 = transpose(perm = v_35_perm_0, x = v_33_cast_fp16)[name = string("transpose_129")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_33_cast_fp16 = matmul(transpose_x = attn_out_33_transpose_x_0, transpose_y = attn_out_33_transpose_y_0, x = var_3142_cast_fp16, y = v_35_cast_fp16)[name = string("attn_out_33_cast_fp16")];
            tensor<int32, [4]> var_3151_perm_0 = const()[name = string("op_3151_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3153 = const()[name = string("op_3153"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_3151_cast_fp16 = transpose(perm = var_3151_perm_0, x = attn_out_33_cast_fp16)[name = string("transpose_128")];
            tensor<fp16, [1, 291, 1024]> input_105_cast_fp16 = reshape(shape = var_3153, x = var_3151_cast_fp16)[name = string("input_105_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_8_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_8_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276989440)))];
            tensor<fp16, [1, 291, 1024]> linear_59_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_8_self_attn_o_proj_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_59_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_135_cast_fp16 = add(x = hidden_states_127_cast_fp16, y = linear_59_cast_fp16)[name = string("hidden_states_135_cast_fp16")];
            fp16 var_3163_promoted_to_fp16 = const()[name = string("op_3163_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3169_cast_fp16 = pow(x = hidden_states_135_cast_fp16, y = var_3163_promoted_to_fp16)[name = string("op_3169_cast_fp16")];
            tensor<int32, [1]> variance_35_axes_0 = const()[name = string("variance_35_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_35_keep_dims_0 = const()[name = string("variance_35_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_35_cast_fp16 = reduce_mean(axes = variance_35_axes_0, keep_dims = variance_35_keep_dims_0, x = var_3169_cast_fp16)[name = string("variance_35_cast_fp16")];
            fp16 var_3172_to_fp16 = const()[name = string("op_3172_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3173_cast_fp16 = add(x = variance_35_cast_fp16, y = var_3172_to_fp16)[name = string("op_3173_cast_fp16")];
            fp32 var_3174_epsilon_0 = const()[name = string("op_3174_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3174_cast_fp16 = rsqrt(epsilon = var_3174_epsilon_0, x = var_3173_cast_fp16)[name = string("op_3174_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_139_cast_fp16 = mul(x = hidden_states_135_cast_fp16, y = var_3174_cast_fp16)[name = string("hidden_states_139_cast_fp16")];
            tensor<fp16, [1024]> layers_8_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_8_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279086656)))];
            tensor<fp16, [1, 291, 1024]> input_107_cast_fp16 = mul(x = layers_8_post_attention_layernorm_weight_to_fp16, y = hidden_states_139_cast_fp16)[name = string("input_107_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_8_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_8_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279088768)))];
            tensor<fp16, [1, 291, 4096]> linear_60_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_8_mlp_gate_proj_weight_to_fp16, x = input_107_cast_fp16)[name = string("linear_60_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_3187_cast_fp16 = silu(x = linear_60_cast_fp16)[name = string("op_3187_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_8_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_8_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(287477440)))];
            tensor<fp16, [1, 291, 4096]> linear_61_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_8_mlp_up_proj_weight_to_fp16, x = input_107_cast_fp16)[name = string("linear_61_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_111_cast_fp16 = mul(x = var_3187_cast_fp16, y = linear_61_cast_fp16)[name = string("input_111_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_8_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_8_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295866112)))];
            tensor<fp16, [1, 291, 1024]> linear_62_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_8_mlp_down_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("linear_62_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_143_cast_fp16 = add(x = hidden_states_135_cast_fp16, y = linear_62_cast_fp16)[name = string("hidden_states_143_cast_fp16")];
            tensor<int32, [4]> var_3199 = const()[name = string("op_3199"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_3211 = const()[name = string("op_3211"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_3200_cast_fp16 = transpose(perm = var_3199, x = k_53_cast_fp16)[name = string("transpose_127")];
            tensor<fp16, [1, 1024, 1, 291]> input_113_cast_fp16 = reshape(shape = var_3211, x = var_3200_cast_fp16)[name = string("input_113_cast_fp16")];
            tensor<int32, [8]> var_3217_pad_0 = const()[name = string("op_3217_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_3217_mode_0 = const()[name = string("op_3217_mode_0"), val = string("constant")];
            fp16 const_144_to_fp16 = const()[name = string("const_144_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_3217_cast_fp16 = pad(constant_val = const_144_to_fp16, mode = var_3217_mode_0, pad = var_3217_pad_0, x = input_113_cast_fp16)[name = string("op_3217_cast_fp16")];
            tensor<int32, [4]> var_3222 = const()[name = string("op_3222"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_3234 = const()[name = string("op_3234"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_3223_cast_fp16 = transpose(perm = var_3222, x = v_35_cast_fp16)[name = string("transpose_126")];
            tensor<fp16, [1, 1024, 1, 291]> input_115_cast_fp16 = reshape(shape = var_3234, x = var_3223_cast_fp16)[name = string("input_115_cast_fp16")];
            tensor<int32, [8]> var_3240_pad_0 = const()[name = string("op_3240_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_3240_mode_0 = const()[name = string("op_3240_mode_0"), val = string("constant")];
            fp16 const_147_to_fp16 = const()[name = string("const_147_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_3240_cast_fp16 = pad(constant_val = const_147_to_fp16, mode = var_3240_mode_0, pad = var_3240_pad_0, x = input_115_cast_fp16)[name = string("op_3240_cast_fp16")];
            fp16 var_3244_promoted_to_fp16 = const()[name = string("op_3244_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3250_cast_fp16 = pow(x = hidden_states_143_cast_fp16, y = var_3244_promoted_to_fp16)[name = string("op_3250_cast_fp16")];
            tensor<int32, [1]> variance_37_axes_0 = const()[name = string("variance_37_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_37_keep_dims_0 = const()[name = string("variance_37_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_37_cast_fp16 = reduce_mean(axes = variance_37_axes_0, keep_dims = variance_37_keep_dims_0, x = var_3250_cast_fp16)[name = string("variance_37_cast_fp16")];
            fp16 var_3253_to_fp16 = const()[name = string("op_3253_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3254_cast_fp16 = add(x = variance_37_cast_fp16, y = var_3253_to_fp16)[name = string("op_3254_cast_fp16")];
            fp32 var_3255_epsilon_0 = const()[name = string("op_3255_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3255_cast_fp16 = rsqrt(epsilon = var_3255_epsilon_0, x = var_3254_cast_fp16)[name = string("op_3255_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_147_cast_fp16 = mul(x = hidden_states_143_cast_fp16, y = var_3255_cast_fp16)[name = string("hidden_states_147_cast_fp16")];
            tensor<fp16, [1024]> layers_9_input_layernorm_weight_to_fp16 = const()[name = string("layers_9_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304254784)))];
            tensor<fp16, [1, 291, 1024]> hidden_37_cast_fp16 = mul(x = layers_9_input_layernorm_weight_to_fp16, y = hidden_states_147_cast_fp16)[name = string("hidden_37_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(304256896)))];
            tensor<fp16, [1, 291, 1024]> linear_63_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_9_self_attn_q_proj_weight_to_fp16, x = hidden_37_cast_fp16)[name = string("linear_63_cast_fp16")];
            tensor<int32, [4]> var_3279 = const()[name = string("op_3279"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_55_cast_fp16 = reshape(shape = var_3279, x = linear_63_cast_fp16)[name = string("q_55_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(306354112)))];
            tensor<fp16, [1, 291, 1024]> linear_64_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_9_self_attn_k_proj_weight_to_fp16, x = hidden_37_cast_fp16)[name = string("linear_64_cast_fp16")];
            tensor<int32, [4]> var_3286 = const()[name = string("op_3286"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_55_cast_fp16 = reshape(shape = var_3286, x = linear_64_cast_fp16)[name = string("k_55_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(308451328)))];
            tensor<fp16, [1, 291, 1024]> linear_65_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_9_self_attn_v_proj_weight_to_fp16, x = hidden_37_cast_fp16)[name = string("linear_65_cast_fp16")];
            tensor<int32, [4]> var_3293 = const()[name = string("op_3293"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_37_cast_fp16 = reshape(shape = var_3293, x = linear_65_cast_fp16)[name = string("v_37_cast_fp16")];
            tensor<int32, [4]> q_57_perm_0 = const()[name = string("q_57_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_57_perm_0 = const()[name = string("k_57_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_39_perm_0 = const()[name = string("v_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_57_cast_fp16 = transpose(perm = q_57_perm_0, x = q_55_cast_fp16)[name = string("transpose_125")];
            tensor<fp16, [1, 16, 291, 64]> var_3306_cast_fp16 = mul(x = q_57_cast_fp16, y = var_1091_to_fp16)[name = string("op_3306_cast_fp16")];
            tensor<int32, [4]> x1_37_begin_0 = const()[name = string("x1_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_37_end_0 = const()[name = string("x1_37_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_37_end_mask_0 = const()[name = string("x1_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_37_cast_fp16 = slice_by_index(begin = x1_37_begin_0, end = x1_37_end_0, end_mask = x1_37_end_mask_0, x = q_57_cast_fp16)[name = string("x1_37_cast_fp16")];
            tensor<int32, [4]> x2_37_begin_0 = const()[name = string("x2_37_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_37_end_0 = const()[name = string("x2_37_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_37_end_mask_0 = const()[name = string("x2_37_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_37_cast_fp16 = slice_by_index(begin = x2_37_begin_0, end = x2_37_end_0, end_mask = x2_37_end_mask_0, x = q_57_cast_fp16)[name = string("x2_37_cast_fp16")];
            fp16 const_152_promoted_to_fp16 = const()[name = string("const_152_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_3327_cast_fp16 = mul(x = x2_37_cast_fp16, y = const_152_promoted_to_fp16)[name = string("op_3327_cast_fp16")];
            int32 var_3329 = const()[name = string("op_3329"), val = int32(-1)];
            bool var_3330_interleave_0 = const()[name = string("op_3330_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_3330_cast_fp16 = concat(axis = var_3329, interleave = var_3330_interleave_0, values = (var_3327_cast_fp16, x1_37_cast_fp16))[name = string("op_3330_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_3333_cast_fp16 = mul(x = var_3330_cast_fp16, y = var_1118_to_fp16)[name = string("op_3333_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_59_cast_fp16 = add(x = var_3306_cast_fp16, y = var_3333_cast_fp16)[name = string("q_59_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_57_cast_fp16 = transpose(perm = k_57_perm_0, x = k_55_cast_fp16)[name = string("transpose_124")];
            tensor<fp16, [1, 16, 291, 64]> var_3338_cast_fp16 = mul(x = k_57_cast_fp16, y = var_1091_to_fp16)[name = string("op_3338_cast_fp16")];
            tensor<int32, [4]> x1_39_begin_0 = const()[name = string("x1_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_39_end_0 = const()[name = string("x1_39_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_39_end_mask_0 = const()[name = string("x1_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_39_cast_fp16 = slice_by_index(begin = x1_39_begin_0, end = x1_39_end_0, end_mask = x1_39_end_mask_0, x = k_57_cast_fp16)[name = string("x1_39_cast_fp16")];
            tensor<int32, [4]> x2_39_begin_0 = const()[name = string("x2_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_39_end_0 = const()[name = string("x2_39_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_39_end_mask_0 = const()[name = string("x2_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_39_cast_fp16 = slice_by_index(begin = x2_39_begin_0, end = x2_39_end_0, end_mask = x2_39_end_mask_0, x = k_57_cast_fp16)[name = string("x2_39_cast_fp16")];
            fp16 const_155_promoted_to_fp16 = const()[name = string("const_155_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_3359_cast_fp16 = mul(x = x2_39_cast_fp16, y = const_155_promoted_to_fp16)[name = string("op_3359_cast_fp16")];
            int32 var_3361 = const()[name = string("op_3361"), val = int32(-1)];
            bool var_3362_interleave_0 = const()[name = string("op_3362_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_3362_cast_fp16 = concat(axis = var_3361, interleave = var_3362_interleave_0, values = (var_3359_cast_fp16, x1_39_cast_fp16))[name = string("op_3362_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_3365_cast_fp16 = mul(x = var_3362_cast_fp16, y = var_1118_to_fp16)[name = string("op_3365_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_59_cast_fp16 = add(x = var_3338_cast_fp16, y = var_3365_cast_fp16)[name = string("k_59_cast_fp16")];
            bool var_3371_transpose_x_1 = const()[name = string("op_3371_transpose_x_1"), val = bool(false)];
            bool var_3371_transpose_y_1 = const()[name = string("op_3371_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_3371_cast_fp16 = matmul(transpose_x = var_3371_transpose_x_1, transpose_y = var_3371_transpose_y_1, x = q_59_cast_fp16, y = k_59_cast_fp16)[name = string("op_3371_cast_fp16")];
            fp16 _inversed_scores_55_y_0_to_fp16 = const()[name = string("_inversed_scores_55_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_55_cast_fp16 = mul(x = var_3371_cast_fp16, y = _inversed_scores_55_y_0_to_fp16)[name = string("_inversed_scores_55_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_57_cast_fp16 = add(x = _inversed_scores_55_cast_fp16, y = const_21_to_fp16)[name = string("scores_57_cast_fp16")];
            int32 var_3386 = const()[name = string("op_3386"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_3388_cast_fp16 = softmax(axis = var_3386, x = scores_57_cast_fp16)[name = string("op_3388_cast_fp16")];
            bool attn_out_37_transpose_x_0 = const()[name = string("attn_out_37_transpose_x_0"), val = bool(false)];
            bool attn_out_37_transpose_y_0 = const()[name = string("attn_out_37_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_39_cast_fp16 = transpose(perm = v_39_perm_0, x = v_37_cast_fp16)[name = string("transpose_123")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_37_cast_fp16 = matmul(transpose_x = attn_out_37_transpose_x_0, transpose_y = attn_out_37_transpose_y_0, x = var_3388_cast_fp16, y = v_39_cast_fp16)[name = string("attn_out_37_cast_fp16")];
            tensor<int32, [4]> var_3397_perm_0 = const()[name = string("op_3397_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3399 = const()[name = string("op_3399"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_3397_cast_fp16 = transpose(perm = var_3397_perm_0, x = attn_out_37_cast_fp16)[name = string("transpose_122")];
            tensor<fp16, [1, 291, 1024]> input_117_cast_fp16 = reshape(shape = var_3399, x = var_3397_cast_fp16)[name = string("input_117_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_9_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_9_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(310548544)))];
            tensor<fp16, [1, 291, 1024]> linear_66_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_9_self_attn_o_proj_weight_to_fp16, x = input_117_cast_fp16)[name = string("linear_66_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_151_cast_fp16 = add(x = hidden_states_143_cast_fp16, y = linear_66_cast_fp16)[name = string("hidden_states_151_cast_fp16")];
            fp16 var_3409_promoted_to_fp16 = const()[name = string("op_3409_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3415_cast_fp16 = pow(x = hidden_states_151_cast_fp16, y = var_3409_promoted_to_fp16)[name = string("op_3415_cast_fp16")];
            tensor<int32, [1]> variance_39_axes_0 = const()[name = string("variance_39_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_39_keep_dims_0 = const()[name = string("variance_39_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_39_cast_fp16 = reduce_mean(axes = variance_39_axes_0, keep_dims = variance_39_keep_dims_0, x = var_3415_cast_fp16)[name = string("variance_39_cast_fp16")];
            fp16 var_3418_to_fp16 = const()[name = string("op_3418_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3419_cast_fp16 = add(x = variance_39_cast_fp16, y = var_3418_to_fp16)[name = string("op_3419_cast_fp16")];
            fp32 var_3420_epsilon_0 = const()[name = string("op_3420_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3420_cast_fp16 = rsqrt(epsilon = var_3420_epsilon_0, x = var_3419_cast_fp16)[name = string("op_3420_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_155_cast_fp16 = mul(x = hidden_states_151_cast_fp16, y = var_3420_cast_fp16)[name = string("hidden_states_155_cast_fp16")];
            tensor<fp16, [1024]> layers_9_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_9_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312645760)))];
            tensor<fp16, [1, 291, 1024]> input_119_cast_fp16 = mul(x = layers_9_post_attention_layernorm_weight_to_fp16, y = hidden_states_155_cast_fp16)[name = string("input_119_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_9_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_9_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312647872)))];
            tensor<fp16, [1, 291, 4096]> linear_67_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_9_mlp_gate_proj_weight_to_fp16, x = input_119_cast_fp16)[name = string("linear_67_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_3433_cast_fp16 = silu(x = linear_67_cast_fp16)[name = string("op_3433_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_9_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_9_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(321036544)))];
            tensor<fp16, [1, 291, 4096]> linear_68_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_9_mlp_up_proj_weight_to_fp16, x = input_119_cast_fp16)[name = string("linear_68_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_123_cast_fp16 = mul(x = var_3433_cast_fp16, y = linear_68_cast_fp16)[name = string("input_123_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_9_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_9_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(329425216)))];
            tensor<fp16, [1, 291, 1024]> linear_69_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_9_mlp_down_proj_weight_to_fp16, x = input_123_cast_fp16)[name = string("linear_69_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_159_cast_fp16 = add(x = hidden_states_151_cast_fp16, y = linear_69_cast_fp16)[name = string("hidden_states_159_cast_fp16")];
            tensor<int32, [4]> var_3445 = const()[name = string("op_3445"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_3457 = const()[name = string("op_3457"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_3446_cast_fp16 = transpose(perm = var_3445, x = k_59_cast_fp16)[name = string("transpose_121")];
            tensor<fp16, [1, 1024, 1, 291]> input_125_cast_fp16 = reshape(shape = var_3457, x = var_3446_cast_fp16)[name = string("input_125_cast_fp16")];
            tensor<int32, [8]> var_3463_pad_0 = const()[name = string("op_3463_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_3463_mode_0 = const()[name = string("op_3463_mode_0"), val = string("constant")];
            fp16 const_159_to_fp16 = const()[name = string("const_159_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_3463_cast_fp16 = pad(constant_val = const_159_to_fp16, mode = var_3463_mode_0, pad = var_3463_pad_0, x = input_125_cast_fp16)[name = string("op_3463_cast_fp16")];
            tensor<int32, [4]> var_3468 = const()[name = string("op_3468"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_3480 = const()[name = string("op_3480"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_3469_cast_fp16 = transpose(perm = var_3468, x = v_39_cast_fp16)[name = string("transpose_120")];
            tensor<fp16, [1, 1024, 1, 291]> input_127_cast_fp16 = reshape(shape = var_3480, x = var_3469_cast_fp16)[name = string("input_127_cast_fp16")];
            tensor<int32, [8]> var_3486_pad_0 = const()[name = string("op_3486_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_3486_mode_0 = const()[name = string("op_3486_mode_0"), val = string("constant")];
            fp16 const_162_to_fp16 = const()[name = string("const_162_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_3486_cast_fp16 = pad(constant_val = const_162_to_fp16, mode = var_3486_mode_0, pad = var_3486_pad_0, x = input_127_cast_fp16)[name = string("op_3486_cast_fp16")];
            fp16 var_3490_promoted_to_fp16 = const()[name = string("op_3490_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3496_cast_fp16 = pow(x = hidden_states_159_cast_fp16, y = var_3490_promoted_to_fp16)[name = string("op_3496_cast_fp16")];
            tensor<int32, [1]> variance_41_axes_0 = const()[name = string("variance_41_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_41_keep_dims_0 = const()[name = string("variance_41_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_41_cast_fp16 = reduce_mean(axes = variance_41_axes_0, keep_dims = variance_41_keep_dims_0, x = var_3496_cast_fp16)[name = string("variance_41_cast_fp16")];
            fp16 var_3499_to_fp16 = const()[name = string("op_3499_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3500_cast_fp16 = add(x = variance_41_cast_fp16, y = var_3499_to_fp16)[name = string("op_3500_cast_fp16")];
            fp32 var_3501_epsilon_0 = const()[name = string("op_3501_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3501_cast_fp16 = rsqrt(epsilon = var_3501_epsilon_0, x = var_3500_cast_fp16)[name = string("op_3501_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_163_cast_fp16 = mul(x = hidden_states_159_cast_fp16, y = var_3501_cast_fp16)[name = string("hidden_states_163_cast_fp16")];
            tensor<fp16, [1024]> layers_10_input_layernorm_weight_to_fp16 = const()[name = string("layers_10_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337813888)))];
            tensor<fp16, [1, 291, 1024]> hidden_41_cast_fp16 = mul(x = layers_10_input_layernorm_weight_to_fp16, y = hidden_states_163_cast_fp16)[name = string("hidden_41_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337816000)))];
            tensor<fp16, [1, 291, 1024]> linear_70_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_10_self_attn_q_proj_weight_to_fp16, x = hidden_41_cast_fp16)[name = string("linear_70_cast_fp16")];
            tensor<int32, [4]> var_3525 = const()[name = string("op_3525"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_61_cast_fp16 = reshape(shape = var_3525, x = linear_70_cast_fp16)[name = string("q_61_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339913216)))];
            tensor<fp16, [1, 291, 1024]> linear_71_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_10_self_attn_k_proj_weight_to_fp16, x = hidden_41_cast_fp16)[name = string("linear_71_cast_fp16")];
            tensor<int32, [4]> var_3532 = const()[name = string("op_3532"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_61_cast_fp16 = reshape(shape = var_3532, x = linear_71_cast_fp16)[name = string("k_61_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(342010432)))];
            tensor<fp16, [1, 291, 1024]> linear_72_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_10_self_attn_v_proj_weight_to_fp16, x = hidden_41_cast_fp16)[name = string("linear_72_cast_fp16")];
            tensor<int32, [4]> var_3539 = const()[name = string("op_3539"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_41_cast_fp16 = reshape(shape = var_3539, x = linear_72_cast_fp16)[name = string("v_41_cast_fp16")];
            tensor<int32, [4]> q_63_perm_0 = const()[name = string("q_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_63_perm_0 = const()[name = string("k_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_43_perm_0 = const()[name = string("v_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_63_cast_fp16 = transpose(perm = q_63_perm_0, x = q_61_cast_fp16)[name = string("transpose_119")];
            tensor<fp16, [1, 16, 291, 64]> var_3552_cast_fp16 = mul(x = q_63_cast_fp16, y = var_1091_to_fp16)[name = string("op_3552_cast_fp16")];
            tensor<int32, [4]> x1_41_begin_0 = const()[name = string("x1_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_41_end_0 = const()[name = string("x1_41_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_41_end_mask_0 = const()[name = string("x1_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_41_cast_fp16 = slice_by_index(begin = x1_41_begin_0, end = x1_41_end_0, end_mask = x1_41_end_mask_0, x = q_63_cast_fp16)[name = string("x1_41_cast_fp16")];
            tensor<int32, [4]> x2_41_begin_0 = const()[name = string("x2_41_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_41_end_0 = const()[name = string("x2_41_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_41_end_mask_0 = const()[name = string("x2_41_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_41_cast_fp16 = slice_by_index(begin = x2_41_begin_0, end = x2_41_end_0, end_mask = x2_41_end_mask_0, x = q_63_cast_fp16)[name = string("x2_41_cast_fp16")];
            fp16 const_167_promoted_to_fp16 = const()[name = string("const_167_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_3573_cast_fp16 = mul(x = x2_41_cast_fp16, y = const_167_promoted_to_fp16)[name = string("op_3573_cast_fp16")];
            int32 var_3575 = const()[name = string("op_3575"), val = int32(-1)];
            bool var_3576_interleave_0 = const()[name = string("op_3576_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_3576_cast_fp16 = concat(axis = var_3575, interleave = var_3576_interleave_0, values = (var_3573_cast_fp16, x1_41_cast_fp16))[name = string("op_3576_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_3579_cast_fp16 = mul(x = var_3576_cast_fp16, y = var_1118_to_fp16)[name = string("op_3579_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_65_cast_fp16 = add(x = var_3552_cast_fp16, y = var_3579_cast_fp16)[name = string("q_65_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_63_cast_fp16 = transpose(perm = k_63_perm_0, x = k_61_cast_fp16)[name = string("transpose_118")];
            tensor<fp16, [1, 16, 291, 64]> var_3584_cast_fp16 = mul(x = k_63_cast_fp16, y = var_1091_to_fp16)[name = string("op_3584_cast_fp16")];
            tensor<int32, [4]> x1_43_begin_0 = const()[name = string("x1_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_43_end_0 = const()[name = string("x1_43_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_43_end_mask_0 = const()[name = string("x1_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_43_cast_fp16 = slice_by_index(begin = x1_43_begin_0, end = x1_43_end_0, end_mask = x1_43_end_mask_0, x = k_63_cast_fp16)[name = string("x1_43_cast_fp16")];
            tensor<int32, [4]> x2_43_begin_0 = const()[name = string("x2_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_43_end_0 = const()[name = string("x2_43_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_43_end_mask_0 = const()[name = string("x2_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_43_cast_fp16 = slice_by_index(begin = x2_43_begin_0, end = x2_43_end_0, end_mask = x2_43_end_mask_0, x = k_63_cast_fp16)[name = string("x2_43_cast_fp16")];
            fp16 const_170_promoted_to_fp16 = const()[name = string("const_170_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_3605_cast_fp16 = mul(x = x2_43_cast_fp16, y = const_170_promoted_to_fp16)[name = string("op_3605_cast_fp16")];
            int32 var_3607 = const()[name = string("op_3607"), val = int32(-1)];
            bool var_3608_interleave_0 = const()[name = string("op_3608_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_3608_cast_fp16 = concat(axis = var_3607, interleave = var_3608_interleave_0, values = (var_3605_cast_fp16, x1_43_cast_fp16))[name = string("op_3608_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_3611_cast_fp16 = mul(x = var_3608_cast_fp16, y = var_1118_to_fp16)[name = string("op_3611_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_65_cast_fp16 = add(x = var_3584_cast_fp16, y = var_3611_cast_fp16)[name = string("k_65_cast_fp16")];
            bool var_3617_transpose_x_1 = const()[name = string("op_3617_transpose_x_1"), val = bool(false)];
            bool var_3617_transpose_y_1 = const()[name = string("op_3617_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_3617_cast_fp16 = matmul(transpose_x = var_3617_transpose_x_1, transpose_y = var_3617_transpose_y_1, x = q_65_cast_fp16, y = k_65_cast_fp16)[name = string("op_3617_cast_fp16")];
            fp16 _inversed_scores_61_y_0_to_fp16 = const()[name = string("_inversed_scores_61_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_61_cast_fp16 = mul(x = var_3617_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = string("_inversed_scores_61_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_63_cast_fp16 = add(x = _inversed_scores_61_cast_fp16, y = const_21_to_fp16)[name = string("scores_63_cast_fp16")];
            int32 var_3632 = const()[name = string("op_3632"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_3634_cast_fp16 = softmax(axis = var_3632, x = scores_63_cast_fp16)[name = string("op_3634_cast_fp16")];
            bool attn_out_41_transpose_x_0 = const()[name = string("attn_out_41_transpose_x_0"), val = bool(false)];
            bool attn_out_41_transpose_y_0 = const()[name = string("attn_out_41_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_43_cast_fp16 = transpose(perm = v_43_perm_0, x = v_41_cast_fp16)[name = string("transpose_117")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_41_cast_fp16 = matmul(transpose_x = attn_out_41_transpose_x_0, transpose_y = attn_out_41_transpose_y_0, x = var_3634_cast_fp16, y = v_43_cast_fp16)[name = string("attn_out_41_cast_fp16")];
            tensor<int32, [4]> var_3643_perm_0 = const()[name = string("op_3643_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3645 = const()[name = string("op_3645"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_3643_cast_fp16 = transpose(perm = var_3643_perm_0, x = attn_out_41_cast_fp16)[name = string("transpose_116")];
            tensor<fp16, [1, 291, 1024]> input_129_cast_fp16 = reshape(shape = var_3645, x = var_3643_cast_fp16)[name = string("input_129_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_10_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_10_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(344107648)))];
            tensor<fp16, [1, 291, 1024]> linear_73_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_10_self_attn_o_proj_weight_to_fp16, x = input_129_cast_fp16)[name = string("linear_73_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_167_cast_fp16 = add(x = hidden_states_159_cast_fp16, y = linear_73_cast_fp16)[name = string("hidden_states_167_cast_fp16")];
            fp16 var_3655_promoted_to_fp16 = const()[name = string("op_3655_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3661_cast_fp16 = pow(x = hidden_states_167_cast_fp16, y = var_3655_promoted_to_fp16)[name = string("op_3661_cast_fp16")];
            tensor<int32, [1]> variance_43_axes_0 = const()[name = string("variance_43_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_43_keep_dims_0 = const()[name = string("variance_43_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_43_cast_fp16 = reduce_mean(axes = variance_43_axes_0, keep_dims = variance_43_keep_dims_0, x = var_3661_cast_fp16)[name = string("variance_43_cast_fp16")];
            fp16 var_3664_to_fp16 = const()[name = string("op_3664_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3665_cast_fp16 = add(x = variance_43_cast_fp16, y = var_3664_to_fp16)[name = string("op_3665_cast_fp16")];
            fp32 var_3666_epsilon_0 = const()[name = string("op_3666_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3666_cast_fp16 = rsqrt(epsilon = var_3666_epsilon_0, x = var_3665_cast_fp16)[name = string("op_3666_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_171_cast_fp16 = mul(x = hidden_states_167_cast_fp16, y = var_3666_cast_fp16)[name = string("hidden_states_171_cast_fp16")];
            tensor<fp16, [1024]> layers_10_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_10_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(346204864)))];
            tensor<fp16, [1, 291, 1024]> input_131_cast_fp16 = mul(x = layers_10_post_attention_layernorm_weight_to_fp16, y = hidden_states_171_cast_fp16)[name = string("input_131_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_10_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_10_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(346206976)))];
            tensor<fp16, [1, 291, 4096]> linear_74_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_10_mlp_gate_proj_weight_to_fp16, x = input_131_cast_fp16)[name = string("linear_74_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_3679_cast_fp16 = silu(x = linear_74_cast_fp16)[name = string("op_3679_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_10_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_10_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354595648)))];
            tensor<fp16, [1, 291, 4096]> linear_75_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_10_mlp_up_proj_weight_to_fp16, x = input_131_cast_fp16)[name = string("linear_75_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_135_cast_fp16 = mul(x = var_3679_cast_fp16, y = linear_75_cast_fp16)[name = string("input_135_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_10_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_10_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362984320)))];
            tensor<fp16, [1, 291, 1024]> linear_76_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_10_mlp_down_proj_weight_to_fp16, x = input_135_cast_fp16)[name = string("linear_76_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_175_cast_fp16 = add(x = hidden_states_167_cast_fp16, y = linear_76_cast_fp16)[name = string("hidden_states_175_cast_fp16")];
            tensor<int32, [4]> var_3691 = const()[name = string("op_3691"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_3703 = const()[name = string("op_3703"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_3692_cast_fp16 = transpose(perm = var_3691, x = k_65_cast_fp16)[name = string("transpose_115")];
            tensor<fp16, [1, 1024, 1, 291]> input_137_cast_fp16 = reshape(shape = var_3703, x = var_3692_cast_fp16)[name = string("input_137_cast_fp16")];
            tensor<int32, [8]> var_3709_pad_0 = const()[name = string("op_3709_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_3709_mode_0 = const()[name = string("op_3709_mode_0"), val = string("constant")];
            fp16 const_174_to_fp16 = const()[name = string("const_174_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_3709_cast_fp16 = pad(constant_val = const_174_to_fp16, mode = var_3709_mode_0, pad = var_3709_pad_0, x = input_137_cast_fp16)[name = string("op_3709_cast_fp16")];
            tensor<int32, [4]> var_3714 = const()[name = string("op_3714"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_3726 = const()[name = string("op_3726"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_3715_cast_fp16 = transpose(perm = var_3714, x = v_43_cast_fp16)[name = string("transpose_114")];
            tensor<fp16, [1, 1024, 1, 291]> input_139_cast_fp16 = reshape(shape = var_3726, x = var_3715_cast_fp16)[name = string("input_139_cast_fp16")];
            tensor<int32, [8]> var_3732_pad_0 = const()[name = string("op_3732_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_3732_mode_0 = const()[name = string("op_3732_mode_0"), val = string("constant")];
            fp16 const_177_to_fp16 = const()[name = string("const_177_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_3732_cast_fp16 = pad(constant_val = const_177_to_fp16, mode = var_3732_mode_0, pad = var_3732_pad_0, x = input_139_cast_fp16)[name = string("op_3732_cast_fp16")];
            fp16 var_3736_promoted_to_fp16 = const()[name = string("op_3736_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3742_cast_fp16 = pow(x = hidden_states_175_cast_fp16, y = var_3736_promoted_to_fp16)[name = string("op_3742_cast_fp16")];
            tensor<int32, [1]> variance_45_axes_0 = const()[name = string("variance_45_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_45_keep_dims_0 = const()[name = string("variance_45_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_45_cast_fp16 = reduce_mean(axes = variance_45_axes_0, keep_dims = variance_45_keep_dims_0, x = var_3742_cast_fp16)[name = string("variance_45_cast_fp16")];
            fp16 var_3745_to_fp16 = const()[name = string("op_3745_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3746_cast_fp16 = add(x = variance_45_cast_fp16, y = var_3745_to_fp16)[name = string("op_3746_cast_fp16")];
            fp32 var_3747_epsilon_0 = const()[name = string("op_3747_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3747_cast_fp16 = rsqrt(epsilon = var_3747_epsilon_0, x = var_3746_cast_fp16)[name = string("op_3747_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_179_cast_fp16 = mul(x = hidden_states_175_cast_fp16, y = var_3747_cast_fp16)[name = string("hidden_states_179_cast_fp16")];
            tensor<fp16, [1024]> layers_11_input_layernorm_weight_to_fp16 = const()[name = string("layers_11_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371372992)))];
            tensor<fp16, [1, 291, 1024]> hidden_45_cast_fp16 = mul(x = layers_11_input_layernorm_weight_to_fp16, y = hidden_states_179_cast_fp16)[name = string("hidden_45_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(371375104)))];
            tensor<fp16, [1, 291, 1024]> linear_77_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_11_self_attn_q_proj_weight_to_fp16, x = hidden_45_cast_fp16)[name = string("linear_77_cast_fp16")];
            tensor<int32, [4]> var_3771 = const()[name = string("op_3771"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_67_cast_fp16 = reshape(shape = var_3771, x = linear_77_cast_fp16)[name = string("q_67_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(373472320)))];
            tensor<fp16, [1, 291, 1024]> linear_78_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_11_self_attn_k_proj_weight_to_fp16, x = hidden_45_cast_fp16)[name = string("linear_78_cast_fp16")];
            tensor<int32, [4]> var_3778 = const()[name = string("op_3778"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_67_cast_fp16 = reshape(shape = var_3778, x = linear_78_cast_fp16)[name = string("k_67_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(375569536)))];
            tensor<fp16, [1, 291, 1024]> linear_79_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_11_self_attn_v_proj_weight_to_fp16, x = hidden_45_cast_fp16)[name = string("linear_79_cast_fp16")];
            tensor<int32, [4]> var_3785 = const()[name = string("op_3785"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_45_cast_fp16 = reshape(shape = var_3785, x = linear_79_cast_fp16)[name = string("v_45_cast_fp16")];
            tensor<int32, [4]> q_69_perm_0 = const()[name = string("q_69_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_69_perm_0 = const()[name = string("k_69_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_47_perm_0 = const()[name = string("v_47_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_69_cast_fp16 = transpose(perm = q_69_perm_0, x = q_67_cast_fp16)[name = string("transpose_113")];
            tensor<fp16, [1, 16, 291, 64]> var_3798_cast_fp16 = mul(x = q_69_cast_fp16, y = var_1091_to_fp16)[name = string("op_3798_cast_fp16")];
            tensor<int32, [4]> x1_45_begin_0 = const()[name = string("x1_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_45_end_0 = const()[name = string("x1_45_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_45_end_mask_0 = const()[name = string("x1_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_45_cast_fp16 = slice_by_index(begin = x1_45_begin_0, end = x1_45_end_0, end_mask = x1_45_end_mask_0, x = q_69_cast_fp16)[name = string("x1_45_cast_fp16")];
            tensor<int32, [4]> x2_45_begin_0 = const()[name = string("x2_45_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_45_end_0 = const()[name = string("x2_45_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_45_end_mask_0 = const()[name = string("x2_45_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_45_cast_fp16 = slice_by_index(begin = x2_45_begin_0, end = x2_45_end_0, end_mask = x2_45_end_mask_0, x = q_69_cast_fp16)[name = string("x2_45_cast_fp16")];
            fp16 const_182_promoted_to_fp16 = const()[name = string("const_182_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_3819_cast_fp16 = mul(x = x2_45_cast_fp16, y = const_182_promoted_to_fp16)[name = string("op_3819_cast_fp16")];
            int32 var_3821 = const()[name = string("op_3821"), val = int32(-1)];
            bool var_3822_interleave_0 = const()[name = string("op_3822_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_3822_cast_fp16 = concat(axis = var_3821, interleave = var_3822_interleave_0, values = (var_3819_cast_fp16, x1_45_cast_fp16))[name = string("op_3822_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_3825_cast_fp16 = mul(x = var_3822_cast_fp16, y = var_1118_to_fp16)[name = string("op_3825_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_71_cast_fp16 = add(x = var_3798_cast_fp16, y = var_3825_cast_fp16)[name = string("q_71_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_69_cast_fp16 = transpose(perm = k_69_perm_0, x = k_67_cast_fp16)[name = string("transpose_112")];
            tensor<fp16, [1, 16, 291, 64]> var_3830_cast_fp16 = mul(x = k_69_cast_fp16, y = var_1091_to_fp16)[name = string("op_3830_cast_fp16")];
            tensor<int32, [4]> x1_47_begin_0 = const()[name = string("x1_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_47_end_0 = const()[name = string("x1_47_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_47_end_mask_0 = const()[name = string("x1_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_47_cast_fp16 = slice_by_index(begin = x1_47_begin_0, end = x1_47_end_0, end_mask = x1_47_end_mask_0, x = k_69_cast_fp16)[name = string("x1_47_cast_fp16")];
            tensor<int32, [4]> x2_47_begin_0 = const()[name = string("x2_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_47_end_0 = const()[name = string("x2_47_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_47_end_mask_0 = const()[name = string("x2_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_47_cast_fp16 = slice_by_index(begin = x2_47_begin_0, end = x2_47_end_0, end_mask = x2_47_end_mask_0, x = k_69_cast_fp16)[name = string("x2_47_cast_fp16")];
            fp16 const_185_promoted_to_fp16 = const()[name = string("const_185_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_3851_cast_fp16 = mul(x = x2_47_cast_fp16, y = const_185_promoted_to_fp16)[name = string("op_3851_cast_fp16")];
            int32 var_3853 = const()[name = string("op_3853"), val = int32(-1)];
            bool var_3854_interleave_0 = const()[name = string("op_3854_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_3854_cast_fp16 = concat(axis = var_3853, interleave = var_3854_interleave_0, values = (var_3851_cast_fp16, x1_47_cast_fp16))[name = string("op_3854_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_3857_cast_fp16 = mul(x = var_3854_cast_fp16, y = var_1118_to_fp16)[name = string("op_3857_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_71_cast_fp16 = add(x = var_3830_cast_fp16, y = var_3857_cast_fp16)[name = string("k_71_cast_fp16")];
            bool var_3863_transpose_x_1 = const()[name = string("op_3863_transpose_x_1"), val = bool(false)];
            bool var_3863_transpose_y_1 = const()[name = string("op_3863_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_3863_cast_fp16 = matmul(transpose_x = var_3863_transpose_x_1, transpose_y = var_3863_transpose_y_1, x = q_71_cast_fp16, y = k_71_cast_fp16)[name = string("op_3863_cast_fp16")];
            fp16 _inversed_scores_67_y_0_to_fp16 = const()[name = string("_inversed_scores_67_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_67_cast_fp16 = mul(x = var_3863_cast_fp16, y = _inversed_scores_67_y_0_to_fp16)[name = string("_inversed_scores_67_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_69_cast_fp16 = add(x = _inversed_scores_67_cast_fp16, y = const_21_to_fp16)[name = string("scores_69_cast_fp16")];
            int32 var_3878 = const()[name = string("op_3878"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_3880_cast_fp16 = softmax(axis = var_3878, x = scores_69_cast_fp16)[name = string("op_3880_cast_fp16")];
            bool attn_out_45_transpose_x_0 = const()[name = string("attn_out_45_transpose_x_0"), val = bool(false)];
            bool attn_out_45_transpose_y_0 = const()[name = string("attn_out_45_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_47_cast_fp16 = transpose(perm = v_47_perm_0, x = v_45_cast_fp16)[name = string("transpose_111")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_45_cast_fp16 = matmul(transpose_x = attn_out_45_transpose_x_0, transpose_y = attn_out_45_transpose_y_0, x = var_3880_cast_fp16, y = v_47_cast_fp16)[name = string("attn_out_45_cast_fp16")];
            tensor<int32, [4]> var_3889_perm_0 = const()[name = string("op_3889_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3891 = const()[name = string("op_3891"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_3889_cast_fp16 = transpose(perm = var_3889_perm_0, x = attn_out_45_cast_fp16)[name = string("transpose_110")];
            tensor<fp16, [1, 291, 1024]> input_141_cast_fp16 = reshape(shape = var_3891, x = var_3889_cast_fp16)[name = string("input_141_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_11_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_11_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(377666752)))];
            tensor<fp16, [1, 291, 1024]> linear_80_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_11_self_attn_o_proj_weight_to_fp16, x = input_141_cast_fp16)[name = string("linear_80_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_183_cast_fp16 = add(x = hidden_states_175_cast_fp16, y = linear_80_cast_fp16)[name = string("hidden_states_183_cast_fp16")];
            fp16 var_3901_promoted_to_fp16 = const()[name = string("op_3901_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3907_cast_fp16 = pow(x = hidden_states_183_cast_fp16, y = var_3901_promoted_to_fp16)[name = string("op_3907_cast_fp16")];
            tensor<int32, [1]> variance_47_axes_0 = const()[name = string("variance_47_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_47_keep_dims_0 = const()[name = string("variance_47_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_47_cast_fp16 = reduce_mean(axes = variance_47_axes_0, keep_dims = variance_47_keep_dims_0, x = var_3907_cast_fp16)[name = string("variance_47_cast_fp16")];
            fp16 var_3910_to_fp16 = const()[name = string("op_3910_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3911_cast_fp16 = add(x = variance_47_cast_fp16, y = var_3910_to_fp16)[name = string("op_3911_cast_fp16")];
            fp32 var_3912_epsilon_0 = const()[name = string("op_3912_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3912_cast_fp16 = rsqrt(epsilon = var_3912_epsilon_0, x = var_3911_cast_fp16)[name = string("op_3912_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_187_cast_fp16 = mul(x = hidden_states_183_cast_fp16, y = var_3912_cast_fp16)[name = string("hidden_states_187_cast_fp16")];
            tensor<fp16, [1024]> layers_11_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_11_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379763968)))];
            tensor<fp16, [1, 291, 1024]> input_143_cast_fp16 = mul(x = layers_11_post_attention_layernorm_weight_to_fp16, y = hidden_states_187_cast_fp16)[name = string("input_143_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_11_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_11_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379766080)))];
            tensor<fp16, [1, 291, 4096]> linear_81_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_11_mlp_gate_proj_weight_to_fp16, x = input_143_cast_fp16)[name = string("linear_81_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_3925_cast_fp16 = silu(x = linear_81_cast_fp16)[name = string("op_3925_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_11_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_11_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(388154752)))];
            tensor<fp16, [1, 291, 4096]> linear_82_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_11_mlp_up_proj_weight_to_fp16, x = input_143_cast_fp16)[name = string("linear_82_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_147_cast_fp16 = mul(x = var_3925_cast_fp16, y = linear_82_cast_fp16)[name = string("input_147_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_11_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_11_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396543424)))];
            tensor<fp16, [1, 291, 1024]> linear_83_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_11_mlp_down_proj_weight_to_fp16, x = input_147_cast_fp16)[name = string("linear_83_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_191_cast_fp16 = add(x = hidden_states_183_cast_fp16, y = linear_83_cast_fp16)[name = string("hidden_states_191_cast_fp16")];
            tensor<int32, [4]> var_3937 = const()[name = string("op_3937"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_3949 = const()[name = string("op_3949"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_3938_cast_fp16 = transpose(perm = var_3937, x = k_71_cast_fp16)[name = string("transpose_109")];
            tensor<fp16, [1, 1024, 1, 291]> input_149_cast_fp16 = reshape(shape = var_3949, x = var_3938_cast_fp16)[name = string("input_149_cast_fp16")];
            tensor<int32, [8]> var_3955_pad_0 = const()[name = string("op_3955_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_3955_mode_0 = const()[name = string("op_3955_mode_0"), val = string("constant")];
            fp16 const_189_to_fp16 = const()[name = string("const_189_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_3955_cast_fp16 = pad(constant_val = const_189_to_fp16, mode = var_3955_mode_0, pad = var_3955_pad_0, x = input_149_cast_fp16)[name = string("op_3955_cast_fp16")];
            tensor<int32, [4]> var_3960 = const()[name = string("op_3960"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_3972 = const()[name = string("op_3972"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_3961_cast_fp16 = transpose(perm = var_3960, x = v_47_cast_fp16)[name = string("transpose_108")];
            tensor<fp16, [1, 1024, 1, 291]> input_151_cast_fp16 = reshape(shape = var_3972, x = var_3961_cast_fp16)[name = string("input_151_cast_fp16")];
            tensor<int32, [8]> var_3978_pad_0 = const()[name = string("op_3978_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_3978_mode_0 = const()[name = string("op_3978_mode_0"), val = string("constant")];
            fp16 const_192_to_fp16 = const()[name = string("const_192_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_3978_cast_fp16 = pad(constant_val = const_192_to_fp16, mode = var_3978_mode_0, pad = var_3978_pad_0, x = input_151_cast_fp16)[name = string("op_3978_cast_fp16")];
            fp16 var_3982_promoted_to_fp16 = const()[name = string("op_3982_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_3988_cast_fp16 = pow(x = hidden_states_191_cast_fp16, y = var_3982_promoted_to_fp16)[name = string("op_3988_cast_fp16")];
            tensor<int32, [1]> variance_49_axes_0 = const()[name = string("variance_49_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_49_keep_dims_0 = const()[name = string("variance_49_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_49_cast_fp16 = reduce_mean(axes = variance_49_axes_0, keep_dims = variance_49_keep_dims_0, x = var_3988_cast_fp16)[name = string("variance_49_cast_fp16")];
            fp16 var_3991_to_fp16 = const()[name = string("op_3991_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_3992_cast_fp16 = add(x = variance_49_cast_fp16, y = var_3991_to_fp16)[name = string("op_3992_cast_fp16")];
            fp32 var_3993_epsilon_0 = const()[name = string("op_3993_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_3993_cast_fp16 = rsqrt(epsilon = var_3993_epsilon_0, x = var_3992_cast_fp16)[name = string("op_3993_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_195_cast_fp16 = mul(x = hidden_states_191_cast_fp16, y = var_3993_cast_fp16)[name = string("hidden_states_195_cast_fp16")];
            tensor<fp16, [1024]> layers_12_input_layernorm_weight_to_fp16 = const()[name = string("layers_12_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404932096)))];
            tensor<fp16, [1, 291, 1024]> hidden_49_cast_fp16 = mul(x = layers_12_input_layernorm_weight_to_fp16, y = hidden_states_195_cast_fp16)[name = string("hidden_49_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404934208)))];
            tensor<fp16, [1, 291, 1024]> linear_84_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_12_self_attn_q_proj_weight_to_fp16, x = hidden_49_cast_fp16)[name = string("linear_84_cast_fp16")];
            tensor<int32, [4]> var_4017 = const()[name = string("op_4017"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_73_cast_fp16 = reshape(shape = var_4017, x = linear_84_cast_fp16)[name = string("q_73_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(407031424)))];
            tensor<fp16, [1, 291, 1024]> linear_85_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_12_self_attn_k_proj_weight_to_fp16, x = hidden_49_cast_fp16)[name = string("linear_85_cast_fp16")];
            tensor<int32, [4]> var_4024 = const()[name = string("op_4024"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_73_cast_fp16 = reshape(shape = var_4024, x = linear_85_cast_fp16)[name = string("k_73_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(409128640)))];
            tensor<fp16, [1, 291, 1024]> linear_86_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_12_self_attn_v_proj_weight_to_fp16, x = hidden_49_cast_fp16)[name = string("linear_86_cast_fp16")];
            tensor<int32, [4]> var_4031 = const()[name = string("op_4031"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_49_cast_fp16 = reshape(shape = var_4031, x = linear_86_cast_fp16)[name = string("v_49_cast_fp16")];
            tensor<int32, [4]> q_75_perm_0 = const()[name = string("q_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_75_perm_0 = const()[name = string("k_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_51_perm_0 = const()[name = string("v_51_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_75_cast_fp16 = transpose(perm = q_75_perm_0, x = q_73_cast_fp16)[name = string("transpose_107")];
            tensor<fp16, [1, 16, 291, 64]> var_4044_cast_fp16 = mul(x = q_75_cast_fp16, y = var_1091_to_fp16)[name = string("op_4044_cast_fp16")];
            tensor<int32, [4]> x1_49_begin_0 = const()[name = string("x1_49_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_49_end_0 = const()[name = string("x1_49_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_49_end_mask_0 = const()[name = string("x1_49_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_49_cast_fp16 = slice_by_index(begin = x1_49_begin_0, end = x1_49_end_0, end_mask = x1_49_end_mask_0, x = q_75_cast_fp16)[name = string("x1_49_cast_fp16")];
            tensor<int32, [4]> x2_49_begin_0 = const()[name = string("x2_49_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_49_end_0 = const()[name = string("x2_49_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_49_end_mask_0 = const()[name = string("x2_49_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_49_cast_fp16 = slice_by_index(begin = x2_49_begin_0, end = x2_49_end_0, end_mask = x2_49_end_mask_0, x = q_75_cast_fp16)[name = string("x2_49_cast_fp16")];
            fp16 const_197_promoted_to_fp16 = const()[name = string("const_197_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_4065_cast_fp16 = mul(x = x2_49_cast_fp16, y = const_197_promoted_to_fp16)[name = string("op_4065_cast_fp16")];
            int32 var_4067 = const()[name = string("op_4067"), val = int32(-1)];
            bool var_4068_interleave_0 = const()[name = string("op_4068_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_4068_cast_fp16 = concat(axis = var_4067, interleave = var_4068_interleave_0, values = (var_4065_cast_fp16, x1_49_cast_fp16))[name = string("op_4068_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_4071_cast_fp16 = mul(x = var_4068_cast_fp16, y = var_1118_to_fp16)[name = string("op_4071_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_77_cast_fp16 = add(x = var_4044_cast_fp16, y = var_4071_cast_fp16)[name = string("q_77_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_75_cast_fp16 = transpose(perm = k_75_perm_0, x = k_73_cast_fp16)[name = string("transpose_106")];
            tensor<fp16, [1, 16, 291, 64]> var_4076_cast_fp16 = mul(x = k_75_cast_fp16, y = var_1091_to_fp16)[name = string("op_4076_cast_fp16")];
            tensor<int32, [4]> x1_51_begin_0 = const()[name = string("x1_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_51_end_0 = const()[name = string("x1_51_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_51_end_mask_0 = const()[name = string("x1_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_51_cast_fp16 = slice_by_index(begin = x1_51_begin_0, end = x1_51_end_0, end_mask = x1_51_end_mask_0, x = k_75_cast_fp16)[name = string("x1_51_cast_fp16")];
            tensor<int32, [4]> x2_51_begin_0 = const()[name = string("x2_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_51_end_0 = const()[name = string("x2_51_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_51_end_mask_0 = const()[name = string("x2_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_51_cast_fp16 = slice_by_index(begin = x2_51_begin_0, end = x2_51_end_0, end_mask = x2_51_end_mask_0, x = k_75_cast_fp16)[name = string("x2_51_cast_fp16")];
            fp16 const_200_promoted_to_fp16 = const()[name = string("const_200_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_4097_cast_fp16 = mul(x = x2_51_cast_fp16, y = const_200_promoted_to_fp16)[name = string("op_4097_cast_fp16")];
            int32 var_4099 = const()[name = string("op_4099"), val = int32(-1)];
            bool var_4100_interleave_0 = const()[name = string("op_4100_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_4100_cast_fp16 = concat(axis = var_4099, interleave = var_4100_interleave_0, values = (var_4097_cast_fp16, x1_51_cast_fp16))[name = string("op_4100_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_4103_cast_fp16 = mul(x = var_4100_cast_fp16, y = var_1118_to_fp16)[name = string("op_4103_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_77_cast_fp16 = add(x = var_4076_cast_fp16, y = var_4103_cast_fp16)[name = string("k_77_cast_fp16")];
            bool var_4109_transpose_x_1 = const()[name = string("op_4109_transpose_x_1"), val = bool(false)];
            bool var_4109_transpose_y_1 = const()[name = string("op_4109_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_4109_cast_fp16 = matmul(transpose_x = var_4109_transpose_x_1, transpose_y = var_4109_transpose_y_1, x = q_77_cast_fp16, y = k_77_cast_fp16)[name = string("op_4109_cast_fp16")];
            fp16 _inversed_scores_73_y_0_to_fp16 = const()[name = string("_inversed_scores_73_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_73_cast_fp16 = mul(x = var_4109_cast_fp16, y = _inversed_scores_73_y_0_to_fp16)[name = string("_inversed_scores_73_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_75_cast_fp16 = add(x = _inversed_scores_73_cast_fp16, y = const_21_to_fp16)[name = string("scores_75_cast_fp16")];
            int32 var_4124 = const()[name = string("op_4124"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_4126_cast_fp16 = softmax(axis = var_4124, x = scores_75_cast_fp16)[name = string("op_4126_cast_fp16")];
            bool attn_out_49_transpose_x_0 = const()[name = string("attn_out_49_transpose_x_0"), val = bool(false)];
            bool attn_out_49_transpose_y_0 = const()[name = string("attn_out_49_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_51_cast_fp16 = transpose(perm = v_51_perm_0, x = v_49_cast_fp16)[name = string("transpose_105")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_49_cast_fp16 = matmul(transpose_x = attn_out_49_transpose_x_0, transpose_y = attn_out_49_transpose_y_0, x = var_4126_cast_fp16, y = v_51_cast_fp16)[name = string("attn_out_49_cast_fp16")];
            tensor<int32, [4]> var_4135_perm_0 = const()[name = string("op_4135_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4137 = const()[name = string("op_4137"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_4135_cast_fp16 = transpose(perm = var_4135_perm_0, x = attn_out_49_cast_fp16)[name = string("transpose_104")];
            tensor<fp16, [1, 291, 1024]> input_153_cast_fp16 = reshape(shape = var_4137, x = var_4135_cast_fp16)[name = string("input_153_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_12_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_12_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(411225856)))];
            tensor<fp16, [1, 291, 1024]> linear_87_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_12_self_attn_o_proj_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_87_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_199_cast_fp16 = add(x = hidden_states_191_cast_fp16, y = linear_87_cast_fp16)[name = string("hidden_states_199_cast_fp16")];
            fp16 var_4147_promoted_to_fp16 = const()[name = string("op_4147_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_4153_cast_fp16 = pow(x = hidden_states_199_cast_fp16, y = var_4147_promoted_to_fp16)[name = string("op_4153_cast_fp16")];
            tensor<int32, [1]> variance_51_axes_0 = const()[name = string("variance_51_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_51_keep_dims_0 = const()[name = string("variance_51_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_51_cast_fp16 = reduce_mean(axes = variance_51_axes_0, keep_dims = variance_51_keep_dims_0, x = var_4153_cast_fp16)[name = string("variance_51_cast_fp16")];
            fp16 var_4156_to_fp16 = const()[name = string("op_4156_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_4157_cast_fp16 = add(x = variance_51_cast_fp16, y = var_4156_to_fp16)[name = string("op_4157_cast_fp16")];
            fp32 var_4158_epsilon_0 = const()[name = string("op_4158_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_4158_cast_fp16 = rsqrt(epsilon = var_4158_epsilon_0, x = var_4157_cast_fp16)[name = string("op_4158_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_203_cast_fp16 = mul(x = hidden_states_199_cast_fp16, y = var_4158_cast_fp16)[name = string("hidden_states_203_cast_fp16")];
            tensor<fp16, [1024]> layers_12_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_12_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413323072)))];
            tensor<fp16, [1, 291, 1024]> input_155_cast_fp16 = mul(x = layers_12_post_attention_layernorm_weight_to_fp16, y = hidden_states_203_cast_fp16)[name = string("input_155_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_12_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_12_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(413325184)))];
            tensor<fp16, [1, 291, 4096]> linear_88_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_12_mlp_gate_proj_weight_to_fp16, x = input_155_cast_fp16)[name = string("linear_88_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_4171_cast_fp16 = silu(x = linear_88_cast_fp16)[name = string("op_4171_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_12_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_12_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421713856)))];
            tensor<fp16, [1, 291, 4096]> linear_89_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_12_mlp_up_proj_weight_to_fp16, x = input_155_cast_fp16)[name = string("linear_89_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_159_cast_fp16 = mul(x = var_4171_cast_fp16, y = linear_89_cast_fp16)[name = string("input_159_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_12_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_12_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(430102528)))];
            tensor<fp16, [1, 291, 1024]> linear_90_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_12_mlp_down_proj_weight_to_fp16, x = input_159_cast_fp16)[name = string("linear_90_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_207_cast_fp16 = add(x = hidden_states_199_cast_fp16, y = linear_90_cast_fp16)[name = string("hidden_states_207_cast_fp16")];
            tensor<int32, [4]> var_4183 = const()[name = string("op_4183"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_4195 = const()[name = string("op_4195"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_4184_cast_fp16 = transpose(perm = var_4183, x = k_77_cast_fp16)[name = string("transpose_103")];
            tensor<fp16, [1, 1024, 1, 291]> input_161_cast_fp16 = reshape(shape = var_4195, x = var_4184_cast_fp16)[name = string("input_161_cast_fp16")];
            tensor<int32, [8]> var_4201_pad_0 = const()[name = string("op_4201_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_4201_mode_0 = const()[name = string("op_4201_mode_0"), val = string("constant")];
            fp16 const_204_to_fp16 = const()[name = string("const_204_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_4201_cast_fp16 = pad(constant_val = const_204_to_fp16, mode = var_4201_mode_0, pad = var_4201_pad_0, x = input_161_cast_fp16)[name = string("op_4201_cast_fp16")];
            tensor<int32, [4]> var_4206 = const()[name = string("op_4206"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_4218 = const()[name = string("op_4218"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_4207_cast_fp16 = transpose(perm = var_4206, x = v_51_cast_fp16)[name = string("transpose_102")];
            tensor<fp16, [1, 1024, 1, 291]> input_163_cast_fp16 = reshape(shape = var_4218, x = var_4207_cast_fp16)[name = string("input_163_cast_fp16")];
            tensor<int32, [8]> var_4224_pad_0 = const()[name = string("op_4224_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_4224_mode_0 = const()[name = string("op_4224_mode_0"), val = string("constant")];
            fp16 const_207_to_fp16 = const()[name = string("const_207_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_4224_cast_fp16 = pad(constant_val = const_207_to_fp16, mode = var_4224_mode_0, pad = var_4224_pad_0, x = input_163_cast_fp16)[name = string("op_4224_cast_fp16")];
            fp16 var_4228_promoted_to_fp16 = const()[name = string("op_4228_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_4234_cast_fp16 = pow(x = hidden_states_207_cast_fp16, y = var_4228_promoted_to_fp16)[name = string("op_4234_cast_fp16")];
            tensor<int32, [1]> variance_53_axes_0 = const()[name = string("variance_53_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_53_keep_dims_0 = const()[name = string("variance_53_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_53_cast_fp16 = reduce_mean(axes = variance_53_axes_0, keep_dims = variance_53_keep_dims_0, x = var_4234_cast_fp16)[name = string("variance_53_cast_fp16")];
            fp16 var_4237_to_fp16 = const()[name = string("op_4237_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_4238_cast_fp16 = add(x = variance_53_cast_fp16, y = var_4237_to_fp16)[name = string("op_4238_cast_fp16")];
            fp32 var_4239_epsilon_0 = const()[name = string("op_4239_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_4239_cast_fp16 = rsqrt(epsilon = var_4239_epsilon_0, x = var_4238_cast_fp16)[name = string("op_4239_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_211_cast_fp16 = mul(x = hidden_states_207_cast_fp16, y = var_4239_cast_fp16)[name = string("hidden_states_211_cast_fp16")];
            tensor<fp16, [1024]> layers_13_input_layernorm_weight_to_fp16 = const()[name = string("layers_13_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438491200)))];
            tensor<fp16, [1, 291, 1024]> hidden_53_cast_fp16 = mul(x = layers_13_input_layernorm_weight_to_fp16, y = hidden_states_211_cast_fp16)[name = string("hidden_53_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438493312)))];
            tensor<fp16, [1, 291, 1024]> linear_91_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_13_self_attn_q_proj_weight_to_fp16, x = hidden_53_cast_fp16)[name = string("linear_91_cast_fp16")];
            tensor<int32, [4]> var_4263 = const()[name = string("op_4263"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_79_cast_fp16 = reshape(shape = var_4263, x = linear_91_cast_fp16)[name = string("q_79_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440590528)))];
            tensor<fp16, [1, 291, 1024]> linear_92_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_13_self_attn_k_proj_weight_to_fp16, x = hidden_53_cast_fp16)[name = string("linear_92_cast_fp16")];
            tensor<int32, [4]> var_4270 = const()[name = string("op_4270"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_79_cast_fp16 = reshape(shape = var_4270, x = linear_92_cast_fp16)[name = string("k_79_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(442687744)))];
            tensor<fp16, [1, 291, 1024]> linear_93_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_13_self_attn_v_proj_weight_to_fp16, x = hidden_53_cast_fp16)[name = string("linear_93_cast_fp16")];
            tensor<int32, [4]> var_4277 = const()[name = string("op_4277"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_53_cast_fp16 = reshape(shape = var_4277, x = linear_93_cast_fp16)[name = string("v_53_cast_fp16")];
            tensor<int32, [4]> q_81_perm_0 = const()[name = string("q_81_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_81_perm_0 = const()[name = string("k_81_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_55_perm_0 = const()[name = string("v_55_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_81_cast_fp16 = transpose(perm = q_81_perm_0, x = q_79_cast_fp16)[name = string("transpose_101")];
            tensor<fp16, [1, 16, 291, 64]> var_4290_cast_fp16 = mul(x = q_81_cast_fp16, y = var_1091_to_fp16)[name = string("op_4290_cast_fp16")];
            tensor<int32, [4]> x1_53_begin_0 = const()[name = string("x1_53_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_53_end_0 = const()[name = string("x1_53_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_53_end_mask_0 = const()[name = string("x1_53_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_53_cast_fp16 = slice_by_index(begin = x1_53_begin_0, end = x1_53_end_0, end_mask = x1_53_end_mask_0, x = q_81_cast_fp16)[name = string("x1_53_cast_fp16")];
            tensor<int32, [4]> x2_53_begin_0 = const()[name = string("x2_53_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_53_end_0 = const()[name = string("x2_53_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_53_end_mask_0 = const()[name = string("x2_53_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_53_cast_fp16 = slice_by_index(begin = x2_53_begin_0, end = x2_53_end_0, end_mask = x2_53_end_mask_0, x = q_81_cast_fp16)[name = string("x2_53_cast_fp16")];
            fp16 const_212_promoted_to_fp16 = const()[name = string("const_212_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_4311_cast_fp16 = mul(x = x2_53_cast_fp16, y = const_212_promoted_to_fp16)[name = string("op_4311_cast_fp16")];
            int32 var_4313 = const()[name = string("op_4313"), val = int32(-1)];
            bool var_4314_interleave_0 = const()[name = string("op_4314_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_4314_cast_fp16 = concat(axis = var_4313, interleave = var_4314_interleave_0, values = (var_4311_cast_fp16, x1_53_cast_fp16))[name = string("op_4314_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_4317_cast_fp16 = mul(x = var_4314_cast_fp16, y = var_1118_to_fp16)[name = string("op_4317_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_83_cast_fp16 = add(x = var_4290_cast_fp16, y = var_4317_cast_fp16)[name = string("q_83_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_81_cast_fp16 = transpose(perm = k_81_perm_0, x = k_79_cast_fp16)[name = string("transpose_100")];
            tensor<fp16, [1, 16, 291, 64]> var_4322_cast_fp16 = mul(x = k_81_cast_fp16, y = var_1091_to_fp16)[name = string("op_4322_cast_fp16")];
            tensor<int32, [4]> x1_55_begin_0 = const()[name = string("x1_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_55_end_0 = const()[name = string("x1_55_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_55_end_mask_0 = const()[name = string("x1_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_55_cast_fp16 = slice_by_index(begin = x1_55_begin_0, end = x1_55_end_0, end_mask = x1_55_end_mask_0, x = k_81_cast_fp16)[name = string("x1_55_cast_fp16")];
            tensor<int32, [4]> x2_55_begin_0 = const()[name = string("x2_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_55_end_0 = const()[name = string("x2_55_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_55_end_mask_0 = const()[name = string("x2_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_55_cast_fp16 = slice_by_index(begin = x2_55_begin_0, end = x2_55_end_0, end_mask = x2_55_end_mask_0, x = k_81_cast_fp16)[name = string("x2_55_cast_fp16")];
            fp16 const_215_promoted_to_fp16 = const()[name = string("const_215_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_4343_cast_fp16 = mul(x = x2_55_cast_fp16, y = const_215_promoted_to_fp16)[name = string("op_4343_cast_fp16")];
            int32 var_4345 = const()[name = string("op_4345"), val = int32(-1)];
            bool var_4346_interleave_0 = const()[name = string("op_4346_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_4346_cast_fp16 = concat(axis = var_4345, interleave = var_4346_interleave_0, values = (var_4343_cast_fp16, x1_55_cast_fp16))[name = string("op_4346_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_4349_cast_fp16 = mul(x = var_4346_cast_fp16, y = var_1118_to_fp16)[name = string("op_4349_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_83_cast_fp16 = add(x = var_4322_cast_fp16, y = var_4349_cast_fp16)[name = string("k_83_cast_fp16")];
            bool var_4355_transpose_x_1 = const()[name = string("op_4355_transpose_x_1"), val = bool(false)];
            bool var_4355_transpose_y_1 = const()[name = string("op_4355_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_4355_cast_fp16 = matmul(transpose_x = var_4355_transpose_x_1, transpose_y = var_4355_transpose_y_1, x = q_83_cast_fp16, y = k_83_cast_fp16)[name = string("op_4355_cast_fp16")];
            fp16 _inversed_scores_79_y_0_to_fp16 = const()[name = string("_inversed_scores_79_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_79_cast_fp16 = mul(x = var_4355_cast_fp16, y = _inversed_scores_79_y_0_to_fp16)[name = string("_inversed_scores_79_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_81_cast_fp16 = add(x = _inversed_scores_79_cast_fp16, y = const_21_to_fp16)[name = string("scores_81_cast_fp16")];
            int32 var_4370 = const()[name = string("op_4370"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_4372_cast_fp16 = softmax(axis = var_4370, x = scores_81_cast_fp16)[name = string("op_4372_cast_fp16")];
            bool attn_out_53_transpose_x_0 = const()[name = string("attn_out_53_transpose_x_0"), val = bool(false)];
            bool attn_out_53_transpose_y_0 = const()[name = string("attn_out_53_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_55_cast_fp16 = transpose(perm = v_55_perm_0, x = v_53_cast_fp16)[name = string("transpose_99")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_53_cast_fp16 = matmul(transpose_x = attn_out_53_transpose_x_0, transpose_y = attn_out_53_transpose_y_0, x = var_4372_cast_fp16, y = v_55_cast_fp16)[name = string("attn_out_53_cast_fp16")];
            tensor<int32, [4]> var_4381_perm_0 = const()[name = string("op_4381_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4383 = const()[name = string("op_4383"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_4381_cast_fp16 = transpose(perm = var_4381_perm_0, x = attn_out_53_cast_fp16)[name = string("transpose_98")];
            tensor<fp16, [1, 291, 1024]> input_165_cast_fp16 = reshape(shape = var_4383, x = var_4381_cast_fp16)[name = string("input_165_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_13_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_13_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444784960)))];
            tensor<fp16, [1, 291, 1024]> linear_94_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_13_self_attn_o_proj_weight_to_fp16, x = input_165_cast_fp16)[name = string("linear_94_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_215_cast_fp16 = add(x = hidden_states_207_cast_fp16, y = linear_94_cast_fp16)[name = string("hidden_states_215_cast_fp16")];
            fp16 var_4393_promoted_to_fp16 = const()[name = string("op_4393_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_4399_cast_fp16 = pow(x = hidden_states_215_cast_fp16, y = var_4393_promoted_to_fp16)[name = string("op_4399_cast_fp16")];
            tensor<int32, [1]> variance_55_axes_0 = const()[name = string("variance_55_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_55_keep_dims_0 = const()[name = string("variance_55_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_55_cast_fp16 = reduce_mean(axes = variance_55_axes_0, keep_dims = variance_55_keep_dims_0, x = var_4399_cast_fp16)[name = string("variance_55_cast_fp16")];
            fp16 var_4402_to_fp16 = const()[name = string("op_4402_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_4403_cast_fp16 = add(x = variance_55_cast_fp16, y = var_4402_to_fp16)[name = string("op_4403_cast_fp16")];
            fp32 var_4404_epsilon_0 = const()[name = string("op_4404_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_4404_cast_fp16 = rsqrt(epsilon = var_4404_epsilon_0, x = var_4403_cast_fp16)[name = string("op_4404_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_219_cast_fp16 = mul(x = hidden_states_215_cast_fp16, y = var_4404_cast_fp16)[name = string("hidden_states_219_cast_fp16")];
            tensor<fp16, [1024]> layers_13_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_13_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446882176)))];
            tensor<fp16, [1, 291, 1024]> input_167_cast_fp16 = mul(x = layers_13_post_attention_layernorm_weight_to_fp16, y = hidden_states_219_cast_fp16)[name = string("input_167_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_13_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_13_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446884288)))];
            tensor<fp16, [1, 291, 4096]> linear_95_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_13_mlp_gate_proj_weight_to_fp16, x = input_167_cast_fp16)[name = string("linear_95_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_4417_cast_fp16 = silu(x = linear_95_cast_fp16)[name = string("op_4417_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_13_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_13_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(455272960)))];
            tensor<fp16, [1, 291, 4096]> linear_96_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_13_mlp_up_proj_weight_to_fp16, x = input_167_cast_fp16)[name = string("linear_96_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_171_cast_fp16 = mul(x = var_4417_cast_fp16, y = linear_96_cast_fp16)[name = string("input_171_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_13_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_13_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463661632)))];
            tensor<fp16, [1, 291, 1024]> linear_97_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_13_mlp_down_proj_weight_to_fp16, x = input_171_cast_fp16)[name = string("linear_97_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_223_cast_fp16 = add(x = hidden_states_215_cast_fp16, y = linear_97_cast_fp16)[name = string("hidden_states_223_cast_fp16")];
            tensor<int32, [4]> var_4429 = const()[name = string("op_4429"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_4441 = const()[name = string("op_4441"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_4430_cast_fp16 = transpose(perm = var_4429, x = k_83_cast_fp16)[name = string("transpose_97")];
            tensor<fp16, [1, 1024, 1, 291]> input_173_cast_fp16 = reshape(shape = var_4441, x = var_4430_cast_fp16)[name = string("input_173_cast_fp16")];
            tensor<int32, [8]> var_4447_pad_0 = const()[name = string("op_4447_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_4447_mode_0 = const()[name = string("op_4447_mode_0"), val = string("constant")];
            fp16 const_219_to_fp16 = const()[name = string("const_219_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_4447_cast_fp16 = pad(constant_val = const_219_to_fp16, mode = var_4447_mode_0, pad = var_4447_pad_0, x = input_173_cast_fp16)[name = string("op_4447_cast_fp16")];
            tensor<int32, [4]> var_4452 = const()[name = string("op_4452"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_4464 = const()[name = string("op_4464"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_4453_cast_fp16 = transpose(perm = var_4452, x = v_55_cast_fp16)[name = string("transpose_96")];
            tensor<fp16, [1, 1024, 1, 291]> input_175_cast_fp16 = reshape(shape = var_4464, x = var_4453_cast_fp16)[name = string("input_175_cast_fp16")];
            tensor<int32, [8]> var_4470_pad_0 = const()[name = string("op_4470_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_4470_mode_0 = const()[name = string("op_4470_mode_0"), val = string("constant")];
            fp16 const_222_to_fp16 = const()[name = string("const_222_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_4470_cast_fp16 = pad(constant_val = const_222_to_fp16, mode = var_4470_mode_0, pad = var_4470_pad_0, x = input_175_cast_fp16)[name = string("op_4470_cast_fp16")];
            fp16 var_4474_promoted_to_fp16 = const()[name = string("op_4474_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_4480_cast_fp16 = pow(x = hidden_states_223_cast_fp16, y = var_4474_promoted_to_fp16)[name = string("op_4480_cast_fp16")];
            tensor<int32, [1]> variance_57_axes_0 = const()[name = string("variance_57_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_57_keep_dims_0 = const()[name = string("variance_57_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_57_cast_fp16 = reduce_mean(axes = variance_57_axes_0, keep_dims = variance_57_keep_dims_0, x = var_4480_cast_fp16)[name = string("variance_57_cast_fp16")];
            fp16 var_4483_to_fp16 = const()[name = string("op_4483_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_4484_cast_fp16 = add(x = variance_57_cast_fp16, y = var_4483_to_fp16)[name = string("op_4484_cast_fp16")];
            fp32 var_4485_epsilon_0 = const()[name = string("op_4485_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_4485_cast_fp16 = rsqrt(epsilon = var_4485_epsilon_0, x = var_4484_cast_fp16)[name = string("op_4485_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_227_cast_fp16 = mul(x = hidden_states_223_cast_fp16, y = var_4485_cast_fp16)[name = string("hidden_states_227_cast_fp16")];
            tensor<fp16, [1024]> layers_14_input_layernorm_weight_to_fp16 = const()[name = string("layers_14_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(472050304)))];
            tensor<fp16, [1, 291, 1024]> hidden_57_cast_fp16 = mul(x = layers_14_input_layernorm_weight_to_fp16, y = hidden_states_227_cast_fp16)[name = string("hidden_57_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(472052416)))];
            tensor<fp16, [1, 291, 1024]> linear_98_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_14_self_attn_q_proj_weight_to_fp16, x = hidden_57_cast_fp16)[name = string("linear_98_cast_fp16")];
            tensor<int32, [4]> var_4509 = const()[name = string("op_4509"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_85_cast_fp16 = reshape(shape = var_4509, x = linear_98_cast_fp16)[name = string("q_85_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(474149632)))];
            tensor<fp16, [1, 291, 1024]> linear_99_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_14_self_attn_k_proj_weight_to_fp16, x = hidden_57_cast_fp16)[name = string("linear_99_cast_fp16")];
            tensor<int32, [4]> var_4516 = const()[name = string("op_4516"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_85_cast_fp16 = reshape(shape = var_4516, x = linear_99_cast_fp16)[name = string("k_85_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(476246848)))];
            tensor<fp16, [1, 291, 1024]> linear_100_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_14_self_attn_v_proj_weight_to_fp16, x = hidden_57_cast_fp16)[name = string("linear_100_cast_fp16")];
            tensor<int32, [4]> var_4523 = const()[name = string("op_4523"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_57_cast_fp16 = reshape(shape = var_4523, x = linear_100_cast_fp16)[name = string("v_57_cast_fp16")];
            tensor<int32, [4]> q_87_perm_0 = const()[name = string("q_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_87_perm_0 = const()[name = string("k_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_59_perm_0 = const()[name = string("v_59_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_87_cast_fp16 = transpose(perm = q_87_perm_0, x = q_85_cast_fp16)[name = string("transpose_95")];
            tensor<fp16, [1, 16, 291, 64]> var_4536_cast_fp16 = mul(x = q_87_cast_fp16, y = var_1091_to_fp16)[name = string("op_4536_cast_fp16")];
            tensor<int32, [4]> x1_57_begin_0 = const()[name = string("x1_57_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_57_end_0 = const()[name = string("x1_57_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_57_end_mask_0 = const()[name = string("x1_57_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_57_cast_fp16 = slice_by_index(begin = x1_57_begin_0, end = x1_57_end_0, end_mask = x1_57_end_mask_0, x = q_87_cast_fp16)[name = string("x1_57_cast_fp16")];
            tensor<int32, [4]> x2_57_begin_0 = const()[name = string("x2_57_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_57_end_0 = const()[name = string("x2_57_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_57_end_mask_0 = const()[name = string("x2_57_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_57_cast_fp16 = slice_by_index(begin = x2_57_begin_0, end = x2_57_end_0, end_mask = x2_57_end_mask_0, x = q_87_cast_fp16)[name = string("x2_57_cast_fp16")];
            fp16 const_227_promoted_to_fp16 = const()[name = string("const_227_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_4557_cast_fp16 = mul(x = x2_57_cast_fp16, y = const_227_promoted_to_fp16)[name = string("op_4557_cast_fp16")];
            int32 var_4559 = const()[name = string("op_4559"), val = int32(-1)];
            bool var_4560_interleave_0 = const()[name = string("op_4560_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_4560_cast_fp16 = concat(axis = var_4559, interleave = var_4560_interleave_0, values = (var_4557_cast_fp16, x1_57_cast_fp16))[name = string("op_4560_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_4563_cast_fp16 = mul(x = var_4560_cast_fp16, y = var_1118_to_fp16)[name = string("op_4563_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_89_cast_fp16 = add(x = var_4536_cast_fp16, y = var_4563_cast_fp16)[name = string("q_89_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_87_cast_fp16 = transpose(perm = k_87_perm_0, x = k_85_cast_fp16)[name = string("transpose_94")];
            tensor<fp16, [1, 16, 291, 64]> var_4568_cast_fp16 = mul(x = k_87_cast_fp16, y = var_1091_to_fp16)[name = string("op_4568_cast_fp16")];
            tensor<int32, [4]> x1_59_begin_0 = const()[name = string("x1_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_59_end_0 = const()[name = string("x1_59_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_59_end_mask_0 = const()[name = string("x1_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_59_cast_fp16 = slice_by_index(begin = x1_59_begin_0, end = x1_59_end_0, end_mask = x1_59_end_mask_0, x = k_87_cast_fp16)[name = string("x1_59_cast_fp16")];
            tensor<int32, [4]> x2_59_begin_0 = const()[name = string("x2_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_59_end_0 = const()[name = string("x2_59_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_59_end_mask_0 = const()[name = string("x2_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_59_cast_fp16 = slice_by_index(begin = x2_59_begin_0, end = x2_59_end_0, end_mask = x2_59_end_mask_0, x = k_87_cast_fp16)[name = string("x2_59_cast_fp16")];
            fp16 const_230_promoted_to_fp16 = const()[name = string("const_230_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_4589_cast_fp16 = mul(x = x2_59_cast_fp16, y = const_230_promoted_to_fp16)[name = string("op_4589_cast_fp16")];
            int32 var_4591 = const()[name = string("op_4591"), val = int32(-1)];
            bool var_4592_interleave_0 = const()[name = string("op_4592_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_4592_cast_fp16 = concat(axis = var_4591, interleave = var_4592_interleave_0, values = (var_4589_cast_fp16, x1_59_cast_fp16))[name = string("op_4592_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_4595_cast_fp16 = mul(x = var_4592_cast_fp16, y = var_1118_to_fp16)[name = string("op_4595_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_89_cast_fp16 = add(x = var_4568_cast_fp16, y = var_4595_cast_fp16)[name = string("k_89_cast_fp16")];
            bool var_4601_transpose_x_1 = const()[name = string("op_4601_transpose_x_1"), val = bool(false)];
            bool var_4601_transpose_y_1 = const()[name = string("op_4601_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_4601_cast_fp16 = matmul(transpose_x = var_4601_transpose_x_1, transpose_y = var_4601_transpose_y_1, x = q_89_cast_fp16, y = k_89_cast_fp16)[name = string("op_4601_cast_fp16")];
            fp16 _inversed_scores_85_y_0_to_fp16 = const()[name = string("_inversed_scores_85_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_85_cast_fp16 = mul(x = var_4601_cast_fp16, y = _inversed_scores_85_y_0_to_fp16)[name = string("_inversed_scores_85_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_87_cast_fp16 = add(x = _inversed_scores_85_cast_fp16, y = const_21_to_fp16)[name = string("scores_87_cast_fp16")];
            int32 var_4616 = const()[name = string("op_4616"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_4618_cast_fp16 = softmax(axis = var_4616, x = scores_87_cast_fp16)[name = string("op_4618_cast_fp16")];
            bool attn_out_57_transpose_x_0 = const()[name = string("attn_out_57_transpose_x_0"), val = bool(false)];
            bool attn_out_57_transpose_y_0 = const()[name = string("attn_out_57_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_59_cast_fp16 = transpose(perm = v_59_perm_0, x = v_57_cast_fp16)[name = string("transpose_93")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_57_cast_fp16 = matmul(transpose_x = attn_out_57_transpose_x_0, transpose_y = attn_out_57_transpose_y_0, x = var_4618_cast_fp16, y = v_59_cast_fp16)[name = string("attn_out_57_cast_fp16")];
            tensor<int32, [4]> var_4627_perm_0 = const()[name = string("op_4627_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4629 = const()[name = string("op_4629"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_4627_cast_fp16 = transpose(perm = var_4627_perm_0, x = attn_out_57_cast_fp16)[name = string("transpose_92")];
            tensor<fp16, [1, 291, 1024]> input_177_cast_fp16 = reshape(shape = var_4629, x = var_4627_cast_fp16)[name = string("input_177_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_14_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_14_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(478344064)))];
            tensor<fp16, [1, 291, 1024]> linear_101_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_14_self_attn_o_proj_weight_to_fp16, x = input_177_cast_fp16)[name = string("linear_101_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_231_cast_fp16 = add(x = hidden_states_223_cast_fp16, y = linear_101_cast_fp16)[name = string("hidden_states_231_cast_fp16")];
            fp16 var_4639_promoted_to_fp16 = const()[name = string("op_4639_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_4645_cast_fp16 = pow(x = hidden_states_231_cast_fp16, y = var_4639_promoted_to_fp16)[name = string("op_4645_cast_fp16")];
            tensor<int32, [1]> variance_59_axes_0 = const()[name = string("variance_59_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_59_keep_dims_0 = const()[name = string("variance_59_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_59_cast_fp16 = reduce_mean(axes = variance_59_axes_0, keep_dims = variance_59_keep_dims_0, x = var_4645_cast_fp16)[name = string("variance_59_cast_fp16")];
            fp16 var_4648_to_fp16 = const()[name = string("op_4648_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_4649_cast_fp16 = add(x = variance_59_cast_fp16, y = var_4648_to_fp16)[name = string("op_4649_cast_fp16")];
            fp32 var_4650_epsilon_0 = const()[name = string("op_4650_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_4650_cast_fp16 = rsqrt(epsilon = var_4650_epsilon_0, x = var_4649_cast_fp16)[name = string("op_4650_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_235_cast_fp16 = mul(x = hidden_states_231_cast_fp16, y = var_4650_cast_fp16)[name = string("hidden_states_235_cast_fp16")];
            tensor<fp16, [1024]> layers_14_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_14_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480441280)))];
            tensor<fp16, [1, 291, 1024]> input_179_cast_fp16 = mul(x = layers_14_post_attention_layernorm_weight_to_fp16, y = hidden_states_235_cast_fp16)[name = string("input_179_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_14_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_14_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480443392)))];
            tensor<fp16, [1, 291, 4096]> linear_102_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_14_mlp_gate_proj_weight_to_fp16, x = input_179_cast_fp16)[name = string("linear_102_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_4663_cast_fp16 = silu(x = linear_102_cast_fp16)[name = string("op_4663_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_14_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_14_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488832064)))];
            tensor<fp16, [1, 291, 4096]> linear_103_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_14_mlp_up_proj_weight_to_fp16, x = input_179_cast_fp16)[name = string("linear_103_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_183_cast_fp16 = mul(x = var_4663_cast_fp16, y = linear_103_cast_fp16)[name = string("input_183_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_14_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_14_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(497220736)))];
            tensor<fp16, [1, 291, 1024]> linear_104_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_14_mlp_down_proj_weight_to_fp16, x = input_183_cast_fp16)[name = string("linear_104_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_239_cast_fp16 = add(x = hidden_states_231_cast_fp16, y = linear_104_cast_fp16)[name = string("hidden_states_239_cast_fp16")];
            tensor<int32, [4]> var_4675 = const()[name = string("op_4675"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_4687 = const()[name = string("op_4687"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_4676_cast_fp16 = transpose(perm = var_4675, x = k_89_cast_fp16)[name = string("transpose_91")];
            tensor<fp16, [1, 1024, 1, 291]> input_185_cast_fp16 = reshape(shape = var_4687, x = var_4676_cast_fp16)[name = string("input_185_cast_fp16")];
            tensor<int32, [8]> var_4693_pad_0 = const()[name = string("op_4693_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_4693_mode_0 = const()[name = string("op_4693_mode_0"), val = string("constant")];
            fp16 const_234_to_fp16 = const()[name = string("const_234_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_4693_cast_fp16 = pad(constant_val = const_234_to_fp16, mode = var_4693_mode_0, pad = var_4693_pad_0, x = input_185_cast_fp16)[name = string("op_4693_cast_fp16")];
            tensor<int32, [4]> var_4698 = const()[name = string("op_4698"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_4710 = const()[name = string("op_4710"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_4699_cast_fp16 = transpose(perm = var_4698, x = v_59_cast_fp16)[name = string("transpose_90")];
            tensor<fp16, [1, 1024, 1, 291]> input_187_cast_fp16 = reshape(shape = var_4710, x = var_4699_cast_fp16)[name = string("input_187_cast_fp16")];
            tensor<int32, [8]> var_4716_pad_0 = const()[name = string("op_4716_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_4716_mode_0 = const()[name = string("op_4716_mode_0"), val = string("constant")];
            fp16 const_237_to_fp16 = const()[name = string("const_237_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_4716_cast_fp16 = pad(constant_val = const_237_to_fp16, mode = var_4716_mode_0, pad = var_4716_pad_0, x = input_187_cast_fp16)[name = string("op_4716_cast_fp16")];
            fp16 var_4720_promoted_to_fp16 = const()[name = string("op_4720_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_4726_cast_fp16 = pow(x = hidden_states_239_cast_fp16, y = var_4720_promoted_to_fp16)[name = string("op_4726_cast_fp16")];
            tensor<int32, [1]> variance_61_axes_0 = const()[name = string("variance_61_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_61_keep_dims_0 = const()[name = string("variance_61_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_61_cast_fp16 = reduce_mean(axes = variance_61_axes_0, keep_dims = variance_61_keep_dims_0, x = var_4726_cast_fp16)[name = string("variance_61_cast_fp16")];
            fp16 var_4729_to_fp16 = const()[name = string("op_4729_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_4730_cast_fp16 = add(x = variance_61_cast_fp16, y = var_4729_to_fp16)[name = string("op_4730_cast_fp16")];
            fp32 var_4731_epsilon_0 = const()[name = string("op_4731_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_4731_cast_fp16 = rsqrt(epsilon = var_4731_epsilon_0, x = var_4730_cast_fp16)[name = string("op_4731_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_243_cast_fp16 = mul(x = hidden_states_239_cast_fp16, y = var_4731_cast_fp16)[name = string("hidden_states_243_cast_fp16")];
            tensor<fp16, [1024]> layers_15_input_layernorm_weight_to_fp16 = const()[name = string("layers_15_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(505609408)))];
            tensor<fp16, [1, 291, 1024]> hidden_61_cast_fp16 = mul(x = layers_15_input_layernorm_weight_to_fp16, y = hidden_states_243_cast_fp16)[name = string("hidden_61_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(505611520)))];
            tensor<fp16, [1, 291, 1024]> linear_105_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_15_self_attn_q_proj_weight_to_fp16, x = hidden_61_cast_fp16)[name = string("linear_105_cast_fp16")];
            tensor<int32, [4]> var_4755 = const()[name = string("op_4755"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_91_cast_fp16 = reshape(shape = var_4755, x = linear_105_cast_fp16)[name = string("q_91_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(507708736)))];
            tensor<fp16, [1, 291, 1024]> linear_106_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_15_self_attn_k_proj_weight_to_fp16, x = hidden_61_cast_fp16)[name = string("linear_106_cast_fp16")];
            tensor<int32, [4]> var_4762 = const()[name = string("op_4762"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_91_cast_fp16 = reshape(shape = var_4762, x = linear_106_cast_fp16)[name = string("k_91_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(509805952)))];
            tensor<fp16, [1, 291, 1024]> linear_107_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_15_self_attn_v_proj_weight_to_fp16, x = hidden_61_cast_fp16)[name = string("linear_107_cast_fp16")];
            tensor<int32, [4]> var_4769 = const()[name = string("op_4769"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_61_cast_fp16 = reshape(shape = var_4769, x = linear_107_cast_fp16)[name = string("v_61_cast_fp16")];
            tensor<int32, [4]> q_93_perm_0 = const()[name = string("q_93_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_93_perm_0 = const()[name = string("k_93_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_63_perm_0 = const()[name = string("v_63_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_93_cast_fp16 = transpose(perm = q_93_perm_0, x = q_91_cast_fp16)[name = string("transpose_89")];
            tensor<fp16, [1, 16, 291, 64]> var_4782_cast_fp16 = mul(x = q_93_cast_fp16, y = var_1091_to_fp16)[name = string("op_4782_cast_fp16")];
            tensor<int32, [4]> x1_61_begin_0 = const()[name = string("x1_61_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_61_end_0 = const()[name = string("x1_61_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_61_end_mask_0 = const()[name = string("x1_61_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_61_cast_fp16 = slice_by_index(begin = x1_61_begin_0, end = x1_61_end_0, end_mask = x1_61_end_mask_0, x = q_93_cast_fp16)[name = string("x1_61_cast_fp16")];
            tensor<int32, [4]> x2_61_begin_0 = const()[name = string("x2_61_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_61_end_0 = const()[name = string("x2_61_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_61_end_mask_0 = const()[name = string("x2_61_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_61_cast_fp16 = slice_by_index(begin = x2_61_begin_0, end = x2_61_end_0, end_mask = x2_61_end_mask_0, x = q_93_cast_fp16)[name = string("x2_61_cast_fp16")];
            fp16 const_242_promoted_to_fp16 = const()[name = string("const_242_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_4803_cast_fp16 = mul(x = x2_61_cast_fp16, y = const_242_promoted_to_fp16)[name = string("op_4803_cast_fp16")];
            int32 var_4805 = const()[name = string("op_4805"), val = int32(-1)];
            bool var_4806_interleave_0 = const()[name = string("op_4806_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_4806_cast_fp16 = concat(axis = var_4805, interleave = var_4806_interleave_0, values = (var_4803_cast_fp16, x1_61_cast_fp16))[name = string("op_4806_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_4809_cast_fp16 = mul(x = var_4806_cast_fp16, y = var_1118_to_fp16)[name = string("op_4809_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_95_cast_fp16 = add(x = var_4782_cast_fp16, y = var_4809_cast_fp16)[name = string("q_95_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_93_cast_fp16 = transpose(perm = k_93_perm_0, x = k_91_cast_fp16)[name = string("transpose_88")];
            tensor<fp16, [1, 16, 291, 64]> var_4814_cast_fp16 = mul(x = k_93_cast_fp16, y = var_1091_to_fp16)[name = string("op_4814_cast_fp16")];
            tensor<int32, [4]> x1_63_begin_0 = const()[name = string("x1_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_63_end_0 = const()[name = string("x1_63_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_63_end_mask_0 = const()[name = string("x1_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_63_cast_fp16 = slice_by_index(begin = x1_63_begin_0, end = x1_63_end_0, end_mask = x1_63_end_mask_0, x = k_93_cast_fp16)[name = string("x1_63_cast_fp16")];
            tensor<int32, [4]> x2_63_begin_0 = const()[name = string("x2_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_63_end_0 = const()[name = string("x2_63_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_63_end_mask_0 = const()[name = string("x2_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_63_cast_fp16 = slice_by_index(begin = x2_63_begin_0, end = x2_63_end_0, end_mask = x2_63_end_mask_0, x = k_93_cast_fp16)[name = string("x2_63_cast_fp16")];
            fp16 const_245_promoted_to_fp16 = const()[name = string("const_245_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_4835_cast_fp16 = mul(x = x2_63_cast_fp16, y = const_245_promoted_to_fp16)[name = string("op_4835_cast_fp16")];
            int32 var_4837 = const()[name = string("op_4837"), val = int32(-1)];
            bool var_4838_interleave_0 = const()[name = string("op_4838_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_4838_cast_fp16 = concat(axis = var_4837, interleave = var_4838_interleave_0, values = (var_4835_cast_fp16, x1_63_cast_fp16))[name = string("op_4838_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_4841_cast_fp16 = mul(x = var_4838_cast_fp16, y = var_1118_to_fp16)[name = string("op_4841_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_95_cast_fp16 = add(x = var_4814_cast_fp16, y = var_4841_cast_fp16)[name = string("k_95_cast_fp16")];
            bool var_4847_transpose_x_1 = const()[name = string("op_4847_transpose_x_1"), val = bool(false)];
            bool var_4847_transpose_y_1 = const()[name = string("op_4847_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_4847_cast_fp16 = matmul(transpose_x = var_4847_transpose_x_1, transpose_y = var_4847_transpose_y_1, x = q_95_cast_fp16, y = k_95_cast_fp16)[name = string("op_4847_cast_fp16")];
            fp16 _inversed_scores_91_y_0_to_fp16 = const()[name = string("_inversed_scores_91_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_91_cast_fp16 = mul(x = var_4847_cast_fp16, y = _inversed_scores_91_y_0_to_fp16)[name = string("_inversed_scores_91_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_93_cast_fp16 = add(x = _inversed_scores_91_cast_fp16, y = const_21_to_fp16)[name = string("scores_93_cast_fp16")];
            int32 var_4862 = const()[name = string("op_4862"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_4864_cast_fp16 = softmax(axis = var_4862, x = scores_93_cast_fp16)[name = string("op_4864_cast_fp16")];
            bool attn_out_61_transpose_x_0 = const()[name = string("attn_out_61_transpose_x_0"), val = bool(false)];
            bool attn_out_61_transpose_y_0 = const()[name = string("attn_out_61_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_63_cast_fp16 = transpose(perm = v_63_perm_0, x = v_61_cast_fp16)[name = string("transpose_87")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_61_cast_fp16 = matmul(transpose_x = attn_out_61_transpose_x_0, transpose_y = attn_out_61_transpose_y_0, x = var_4864_cast_fp16, y = v_63_cast_fp16)[name = string("attn_out_61_cast_fp16")];
            tensor<int32, [4]> var_4873_perm_0 = const()[name = string("op_4873_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_4875 = const()[name = string("op_4875"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_4873_cast_fp16 = transpose(perm = var_4873_perm_0, x = attn_out_61_cast_fp16)[name = string("transpose_86")];
            tensor<fp16, [1, 291, 1024]> input_189_cast_fp16 = reshape(shape = var_4875, x = var_4873_cast_fp16)[name = string("input_189_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_15_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_15_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(511903168)))];
            tensor<fp16, [1, 291, 1024]> linear_108_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_15_self_attn_o_proj_weight_to_fp16, x = input_189_cast_fp16)[name = string("linear_108_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_247_cast_fp16 = add(x = hidden_states_239_cast_fp16, y = linear_108_cast_fp16)[name = string("hidden_states_247_cast_fp16")];
            fp16 var_4885_promoted_to_fp16 = const()[name = string("op_4885_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_4891_cast_fp16 = pow(x = hidden_states_247_cast_fp16, y = var_4885_promoted_to_fp16)[name = string("op_4891_cast_fp16")];
            tensor<int32, [1]> variance_63_axes_0 = const()[name = string("variance_63_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_63_keep_dims_0 = const()[name = string("variance_63_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_63_cast_fp16 = reduce_mean(axes = variance_63_axes_0, keep_dims = variance_63_keep_dims_0, x = var_4891_cast_fp16)[name = string("variance_63_cast_fp16")];
            fp16 var_4894_to_fp16 = const()[name = string("op_4894_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_4895_cast_fp16 = add(x = variance_63_cast_fp16, y = var_4894_to_fp16)[name = string("op_4895_cast_fp16")];
            fp32 var_4896_epsilon_0 = const()[name = string("op_4896_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_4896_cast_fp16 = rsqrt(epsilon = var_4896_epsilon_0, x = var_4895_cast_fp16)[name = string("op_4896_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_251_cast_fp16 = mul(x = hidden_states_247_cast_fp16, y = var_4896_cast_fp16)[name = string("hidden_states_251_cast_fp16")];
            tensor<fp16, [1024]> layers_15_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_15_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(514000384)))];
            tensor<fp16, [1, 291, 1024]> input_191_cast_fp16 = mul(x = layers_15_post_attention_layernorm_weight_to_fp16, y = hidden_states_251_cast_fp16)[name = string("input_191_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_15_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_15_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(514002496)))];
            tensor<fp16, [1, 291, 4096]> linear_109_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_15_mlp_gate_proj_weight_to_fp16, x = input_191_cast_fp16)[name = string("linear_109_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_4909_cast_fp16 = silu(x = linear_109_cast_fp16)[name = string("op_4909_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_15_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_15_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522391168)))];
            tensor<fp16, [1, 291, 4096]> linear_110_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_15_mlp_up_proj_weight_to_fp16, x = input_191_cast_fp16)[name = string("linear_110_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_195_cast_fp16 = mul(x = var_4909_cast_fp16, y = linear_110_cast_fp16)[name = string("input_195_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_15_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_15_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(530779840)))];
            tensor<fp16, [1, 291, 1024]> linear_111_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_15_mlp_down_proj_weight_to_fp16, x = input_195_cast_fp16)[name = string("linear_111_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_255_cast_fp16 = add(x = hidden_states_247_cast_fp16, y = linear_111_cast_fp16)[name = string("hidden_states_255_cast_fp16")];
            tensor<int32, [4]> var_4921 = const()[name = string("op_4921"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_4933 = const()[name = string("op_4933"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_4922_cast_fp16 = transpose(perm = var_4921, x = k_95_cast_fp16)[name = string("transpose_85")];
            tensor<fp16, [1, 1024, 1, 291]> input_197_cast_fp16 = reshape(shape = var_4933, x = var_4922_cast_fp16)[name = string("input_197_cast_fp16")];
            tensor<int32, [8]> var_4939_pad_0 = const()[name = string("op_4939_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_4939_mode_0 = const()[name = string("op_4939_mode_0"), val = string("constant")];
            fp16 const_249_to_fp16 = const()[name = string("const_249_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_4939_cast_fp16 = pad(constant_val = const_249_to_fp16, mode = var_4939_mode_0, pad = var_4939_pad_0, x = input_197_cast_fp16)[name = string("op_4939_cast_fp16")];
            tensor<int32, [4]> var_4944 = const()[name = string("op_4944"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_4956 = const()[name = string("op_4956"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_4945_cast_fp16 = transpose(perm = var_4944, x = v_63_cast_fp16)[name = string("transpose_84")];
            tensor<fp16, [1, 1024, 1, 291]> input_199_cast_fp16 = reshape(shape = var_4956, x = var_4945_cast_fp16)[name = string("input_199_cast_fp16")];
            tensor<int32, [8]> var_4962_pad_0 = const()[name = string("op_4962_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_4962_mode_0 = const()[name = string("op_4962_mode_0"), val = string("constant")];
            fp16 const_252_to_fp16 = const()[name = string("const_252_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_4962_cast_fp16 = pad(constant_val = const_252_to_fp16, mode = var_4962_mode_0, pad = var_4962_pad_0, x = input_199_cast_fp16)[name = string("op_4962_cast_fp16")];
            fp16 var_4966_promoted_to_fp16 = const()[name = string("op_4966_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_4972_cast_fp16 = pow(x = hidden_states_255_cast_fp16, y = var_4966_promoted_to_fp16)[name = string("op_4972_cast_fp16")];
            tensor<int32, [1]> variance_65_axes_0 = const()[name = string("variance_65_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_65_keep_dims_0 = const()[name = string("variance_65_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_65_cast_fp16 = reduce_mean(axes = variance_65_axes_0, keep_dims = variance_65_keep_dims_0, x = var_4972_cast_fp16)[name = string("variance_65_cast_fp16")];
            fp16 var_4975_to_fp16 = const()[name = string("op_4975_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_4976_cast_fp16 = add(x = variance_65_cast_fp16, y = var_4975_to_fp16)[name = string("op_4976_cast_fp16")];
            fp32 var_4977_epsilon_0 = const()[name = string("op_4977_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_4977_cast_fp16 = rsqrt(epsilon = var_4977_epsilon_0, x = var_4976_cast_fp16)[name = string("op_4977_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_259_cast_fp16 = mul(x = hidden_states_255_cast_fp16, y = var_4977_cast_fp16)[name = string("hidden_states_259_cast_fp16")];
            tensor<fp16, [1024]> layers_16_input_layernorm_weight_to_fp16 = const()[name = string("layers_16_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539168512)))];
            tensor<fp16, [1, 291, 1024]> hidden_65_cast_fp16 = mul(x = layers_16_input_layernorm_weight_to_fp16, y = hidden_states_259_cast_fp16)[name = string("hidden_65_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(539170624)))];
            tensor<fp16, [1, 291, 1024]> linear_112_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_16_self_attn_q_proj_weight_to_fp16, x = hidden_65_cast_fp16)[name = string("linear_112_cast_fp16")];
            tensor<int32, [4]> var_5001 = const()[name = string("op_5001"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_97_cast_fp16 = reshape(shape = var_5001, x = linear_112_cast_fp16)[name = string("q_97_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(541267840)))];
            tensor<fp16, [1, 291, 1024]> linear_113_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_16_self_attn_k_proj_weight_to_fp16, x = hidden_65_cast_fp16)[name = string("linear_113_cast_fp16")];
            tensor<int32, [4]> var_5008 = const()[name = string("op_5008"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_97_cast_fp16 = reshape(shape = var_5008, x = linear_113_cast_fp16)[name = string("k_97_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(543365056)))];
            tensor<fp16, [1, 291, 1024]> linear_114_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_16_self_attn_v_proj_weight_to_fp16, x = hidden_65_cast_fp16)[name = string("linear_114_cast_fp16")];
            tensor<int32, [4]> var_5015 = const()[name = string("op_5015"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_65_cast_fp16 = reshape(shape = var_5015, x = linear_114_cast_fp16)[name = string("v_65_cast_fp16")];
            tensor<int32, [4]> q_99_perm_0 = const()[name = string("q_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_99_perm_0 = const()[name = string("k_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_67_perm_0 = const()[name = string("v_67_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_99_cast_fp16 = transpose(perm = q_99_perm_0, x = q_97_cast_fp16)[name = string("transpose_83")];
            tensor<fp16, [1, 16, 291, 64]> var_5028_cast_fp16 = mul(x = q_99_cast_fp16, y = var_1091_to_fp16)[name = string("op_5028_cast_fp16")];
            tensor<int32, [4]> x1_65_begin_0 = const()[name = string("x1_65_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_65_end_0 = const()[name = string("x1_65_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_65_end_mask_0 = const()[name = string("x1_65_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_65_cast_fp16 = slice_by_index(begin = x1_65_begin_0, end = x1_65_end_0, end_mask = x1_65_end_mask_0, x = q_99_cast_fp16)[name = string("x1_65_cast_fp16")];
            tensor<int32, [4]> x2_65_begin_0 = const()[name = string("x2_65_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_65_end_0 = const()[name = string("x2_65_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_65_end_mask_0 = const()[name = string("x2_65_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_65_cast_fp16 = slice_by_index(begin = x2_65_begin_0, end = x2_65_end_0, end_mask = x2_65_end_mask_0, x = q_99_cast_fp16)[name = string("x2_65_cast_fp16")];
            fp16 const_257_promoted_to_fp16 = const()[name = string("const_257_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_5049_cast_fp16 = mul(x = x2_65_cast_fp16, y = const_257_promoted_to_fp16)[name = string("op_5049_cast_fp16")];
            int32 var_5051 = const()[name = string("op_5051"), val = int32(-1)];
            bool var_5052_interleave_0 = const()[name = string("op_5052_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_5052_cast_fp16 = concat(axis = var_5051, interleave = var_5052_interleave_0, values = (var_5049_cast_fp16, x1_65_cast_fp16))[name = string("op_5052_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_5055_cast_fp16 = mul(x = var_5052_cast_fp16, y = var_1118_to_fp16)[name = string("op_5055_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_101_cast_fp16 = add(x = var_5028_cast_fp16, y = var_5055_cast_fp16)[name = string("q_101_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_99_cast_fp16 = transpose(perm = k_99_perm_0, x = k_97_cast_fp16)[name = string("transpose_82")];
            tensor<fp16, [1, 16, 291, 64]> var_5060_cast_fp16 = mul(x = k_99_cast_fp16, y = var_1091_to_fp16)[name = string("op_5060_cast_fp16")];
            tensor<int32, [4]> x1_67_begin_0 = const()[name = string("x1_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_67_end_0 = const()[name = string("x1_67_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_67_end_mask_0 = const()[name = string("x1_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_67_cast_fp16 = slice_by_index(begin = x1_67_begin_0, end = x1_67_end_0, end_mask = x1_67_end_mask_0, x = k_99_cast_fp16)[name = string("x1_67_cast_fp16")];
            tensor<int32, [4]> x2_67_begin_0 = const()[name = string("x2_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_67_end_0 = const()[name = string("x2_67_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_67_end_mask_0 = const()[name = string("x2_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_67_cast_fp16 = slice_by_index(begin = x2_67_begin_0, end = x2_67_end_0, end_mask = x2_67_end_mask_0, x = k_99_cast_fp16)[name = string("x2_67_cast_fp16")];
            fp16 const_260_promoted_to_fp16 = const()[name = string("const_260_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_5081_cast_fp16 = mul(x = x2_67_cast_fp16, y = const_260_promoted_to_fp16)[name = string("op_5081_cast_fp16")];
            int32 var_5083 = const()[name = string("op_5083"), val = int32(-1)];
            bool var_5084_interleave_0 = const()[name = string("op_5084_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_5084_cast_fp16 = concat(axis = var_5083, interleave = var_5084_interleave_0, values = (var_5081_cast_fp16, x1_67_cast_fp16))[name = string("op_5084_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_5087_cast_fp16 = mul(x = var_5084_cast_fp16, y = var_1118_to_fp16)[name = string("op_5087_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_101_cast_fp16 = add(x = var_5060_cast_fp16, y = var_5087_cast_fp16)[name = string("k_101_cast_fp16")];
            bool var_5093_transpose_x_1 = const()[name = string("op_5093_transpose_x_1"), val = bool(false)];
            bool var_5093_transpose_y_1 = const()[name = string("op_5093_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_5093_cast_fp16 = matmul(transpose_x = var_5093_transpose_x_1, transpose_y = var_5093_transpose_y_1, x = q_101_cast_fp16, y = k_101_cast_fp16)[name = string("op_5093_cast_fp16")];
            fp16 _inversed_scores_97_y_0_to_fp16 = const()[name = string("_inversed_scores_97_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_97_cast_fp16 = mul(x = var_5093_cast_fp16, y = _inversed_scores_97_y_0_to_fp16)[name = string("_inversed_scores_97_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_99_cast_fp16 = add(x = _inversed_scores_97_cast_fp16, y = const_21_to_fp16)[name = string("scores_99_cast_fp16")];
            int32 var_5108 = const()[name = string("op_5108"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_5110_cast_fp16 = softmax(axis = var_5108, x = scores_99_cast_fp16)[name = string("op_5110_cast_fp16")];
            bool attn_out_65_transpose_x_0 = const()[name = string("attn_out_65_transpose_x_0"), val = bool(false)];
            bool attn_out_65_transpose_y_0 = const()[name = string("attn_out_65_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_67_cast_fp16 = transpose(perm = v_67_perm_0, x = v_65_cast_fp16)[name = string("transpose_81")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_65_cast_fp16 = matmul(transpose_x = attn_out_65_transpose_x_0, transpose_y = attn_out_65_transpose_y_0, x = var_5110_cast_fp16, y = v_67_cast_fp16)[name = string("attn_out_65_cast_fp16")];
            tensor<int32, [4]> var_5119_perm_0 = const()[name = string("op_5119_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5121 = const()[name = string("op_5121"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_5119_cast_fp16 = transpose(perm = var_5119_perm_0, x = attn_out_65_cast_fp16)[name = string("transpose_80")];
            tensor<fp16, [1, 291, 1024]> input_201_cast_fp16 = reshape(shape = var_5121, x = var_5119_cast_fp16)[name = string("input_201_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_16_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_16_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(545462272)))];
            tensor<fp16, [1, 291, 1024]> linear_115_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_16_self_attn_o_proj_weight_to_fp16, x = input_201_cast_fp16)[name = string("linear_115_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_263_cast_fp16 = add(x = hidden_states_255_cast_fp16, y = linear_115_cast_fp16)[name = string("hidden_states_263_cast_fp16")];
            fp16 var_5131_promoted_to_fp16 = const()[name = string("op_5131_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_5137_cast_fp16 = pow(x = hidden_states_263_cast_fp16, y = var_5131_promoted_to_fp16)[name = string("op_5137_cast_fp16")];
            tensor<int32, [1]> variance_67_axes_0 = const()[name = string("variance_67_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_67_keep_dims_0 = const()[name = string("variance_67_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_67_cast_fp16 = reduce_mean(axes = variance_67_axes_0, keep_dims = variance_67_keep_dims_0, x = var_5137_cast_fp16)[name = string("variance_67_cast_fp16")];
            fp16 var_5140_to_fp16 = const()[name = string("op_5140_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_5141_cast_fp16 = add(x = variance_67_cast_fp16, y = var_5140_to_fp16)[name = string("op_5141_cast_fp16")];
            fp32 var_5142_epsilon_0 = const()[name = string("op_5142_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_5142_cast_fp16 = rsqrt(epsilon = var_5142_epsilon_0, x = var_5141_cast_fp16)[name = string("op_5142_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_267_cast_fp16 = mul(x = hidden_states_263_cast_fp16, y = var_5142_cast_fp16)[name = string("hidden_states_267_cast_fp16")];
            tensor<fp16, [1024]> layers_16_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_16_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547559488)))];
            tensor<fp16, [1, 291, 1024]> input_203_cast_fp16 = mul(x = layers_16_post_attention_layernorm_weight_to_fp16, y = hidden_states_267_cast_fp16)[name = string("input_203_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_16_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_16_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547561600)))];
            tensor<fp16, [1, 291, 4096]> linear_116_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_16_mlp_gate_proj_weight_to_fp16, x = input_203_cast_fp16)[name = string("linear_116_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_5155_cast_fp16 = silu(x = linear_116_cast_fp16)[name = string("op_5155_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_16_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_16_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(555950272)))];
            tensor<fp16, [1, 291, 4096]> linear_117_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_16_mlp_up_proj_weight_to_fp16, x = input_203_cast_fp16)[name = string("linear_117_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_207_cast_fp16 = mul(x = var_5155_cast_fp16, y = linear_117_cast_fp16)[name = string("input_207_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_16_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_16_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564338944)))];
            tensor<fp16, [1, 291, 1024]> linear_118_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_16_mlp_down_proj_weight_to_fp16, x = input_207_cast_fp16)[name = string("linear_118_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_271_cast_fp16 = add(x = hidden_states_263_cast_fp16, y = linear_118_cast_fp16)[name = string("hidden_states_271_cast_fp16")];
            tensor<int32, [4]> var_5167 = const()[name = string("op_5167"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_5179 = const()[name = string("op_5179"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_5168_cast_fp16 = transpose(perm = var_5167, x = k_101_cast_fp16)[name = string("transpose_79")];
            tensor<fp16, [1, 1024, 1, 291]> input_209_cast_fp16 = reshape(shape = var_5179, x = var_5168_cast_fp16)[name = string("input_209_cast_fp16")];
            tensor<int32, [8]> var_5185_pad_0 = const()[name = string("op_5185_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_5185_mode_0 = const()[name = string("op_5185_mode_0"), val = string("constant")];
            fp16 const_264_to_fp16 = const()[name = string("const_264_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_5185_cast_fp16 = pad(constant_val = const_264_to_fp16, mode = var_5185_mode_0, pad = var_5185_pad_0, x = input_209_cast_fp16)[name = string("op_5185_cast_fp16")];
            tensor<int32, [4]> var_5190 = const()[name = string("op_5190"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_5202 = const()[name = string("op_5202"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_5191_cast_fp16 = transpose(perm = var_5190, x = v_67_cast_fp16)[name = string("transpose_78")];
            tensor<fp16, [1, 1024, 1, 291]> input_211_cast_fp16 = reshape(shape = var_5202, x = var_5191_cast_fp16)[name = string("input_211_cast_fp16")];
            tensor<int32, [8]> var_5208_pad_0 = const()[name = string("op_5208_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_5208_mode_0 = const()[name = string("op_5208_mode_0"), val = string("constant")];
            fp16 const_267_to_fp16 = const()[name = string("const_267_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_5208_cast_fp16 = pad(constant_val = const_267_to_fp16, mode = var_5208_mode_0, pad = var_5208_pad_0, x = input_211_cast_fp16)[name = string("op_5208_cast_fp16")];
            fp16 var_5212_promoted_to_fp16 = const()[name = string("op_5212_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_5218_cast_fp16 = pow(x = hidden_states_271_cast_fp16, y = var_5212_promoted_to_fp16)[name = string("op_5218_cast_fp16")];
            tensor<int32, [1]> variance_69_axes_0 = const()[name = string("variance_69_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_69_keep_dims_0 = const()[name = string("variance_69_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_69_cast_fp16 = reduce_mean(axes = variance_69_axes_0, keep_dims = variance_69_keep_dims_0, x = var_5218_cast_fp16)[name = string("variance_69_cast_fp16")];
            fp16 var_5221_to_fp16 = const()[name = string("op_5221_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_5222_cast_fp16 = add(x = variance_69_cast_fp16, y = var_5221_to_fp16)[name = string("op_5222_cast_fp16")];
            fp32 var_5223_epsilon_0 = const()[name = string("op_5223_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_5223_cast_fp16 = rsqrt(epsilon = var_5223_epsilon_0, x = var_5222_cast_fp16)[name = string("op_5223_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_275_cast_fp16 = mul(x = hidden_states_271_cast_fp16, y = var_5223_cast_fp16)[name = string("hidden_states_275_cast_fp16")];
            tensor<fp16, [1024]> layers_17_input_layernorm_weight_to_fp16 = const()[name = string("layers_17_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572727616)))];
            tensor<fp16, [1, 291, 1024]> hidden_69_cast_fp16 = mul(x = layers_17_input_layernorm_weight_to_fp16, y = hidden_states_275_cast_fp16)[name = string("hidden_69_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572729728)))];
            tensor<fp16, [1, 291, 1024]> linear_119_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_17_self_attn_q_proj_weight_to_fp16, x = hidden_69_cast_fp16)[name = string("linear_119_cast_fp16")];
            tensor<int32, [4]> var_5247 = const()[name = string("op_5247"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_103_cast_fp16 = reshape(shape = var_5247, x = linear_119_cast_fp16)[name = string("q_103_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(574826944)))];
            tensor<fp16, [1, 291, 1024]> linear_120_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_17_self_attn_k_proj_weight_to_fp16, x = hidden_69_cast_fp16)[name = string("linear_120_cast_fp16")];
            tensor<int32, [4]> var_5254 = const()[name = string("op_5254"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_103_cast_fp16 = reshape(shape = var_5254, x = linear_120_cast_fp16)[name = string("k_103_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(576924160)))];
            tensor<fp16, [1, 291, 1024]> linear_121_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_17_self_attn_v_proj_weight_to_fp16, x = hidden_69_cast_fp16)[name = string("linear_121_cast_fp16")];
            tensor<int32, [4]> var_5261 = const()[name = string("op_5261"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_69_cast_fp16 = reshape(shape = var_5261, x = linear_121_cast_fp16)[name = string("v_69_cast_fp16")];
            tensor<int32, [4]> q_105_perm_0 = const()[name = string("q_105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_105_perm_0 = const()[name = string("k_105_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_71_perm_0 = const()[name = string("v_71_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_105_cast_fp16 = transpose(perm = q_105_perm_0, x = q_103_cast_fp16)[name = string("transpose_77")];
            tensor<fp16, [1, 16, 291, 64]> var_5274_cast_fp16 = mul(x = q_105_cast_fp16, y = var_1091_to_fp16)[name = string("op_5274_cast_fp16")];
            tensor<int32, [4]> x1_69_begin_0 = const()[name = string("x1_69_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_69_end_0 = const()[name = string("x1_69_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_69_end_mask_0 = const()[name = string("x1_69_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_69_cast_fp16 = slice_by_index(begin = x1_69_begin_0, end = x1_69_end_0, end_mask = x1_69_end_mask_0, x = q_105_cast_fp16)[name = string("x1_69_cast_fp16")];
            tensor<int32, [4]> x2_69_begin_0 = const()[name = string("x2_69_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_69_end_0 = const()[name = string("x2_69_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_69_end_mask_0 = const()[name = string("x2_69_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_69_cast_fp16 = slice_by_index(begin = x2_69_begin_0, end = x2_69_end_0, end_mask = x2_69_end_mask_0, x = q_105_cast_fp16)[name = string("x2_69_cast_fp16")];
            fp16 const_272_promoted_to_fp16 = const()[name = string("const_272_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_5295_cast_fp16 = mul(x = x2_69_cast_fp16, y = const_272_promoted_to_fp16)[name = string("op_5295_cast_fp16")];
            int32 var_5297 = const()[name = string("op_5297"), val = int32(-1)];
            bool var_5298_interleave_0 = const()[name = string("op_5298_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_5298_cast_fp16 = concat(axis = var_5297, interleave = var_5298_interleave_0, values = (var_5295_cast_fp16, x1_69_cast_fp16))[name = string("op_5298_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_5301_cast_fp16 = mul(x = var_5298_cast_fp16, y = var_1118_to_fp16)[name = string("op_5301_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_107_cast_fp16 = add(x = var_5274_cast_fp16, y = var_5301_cast_fp16)[name = string("q_107_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_105_cast_fp16 = transpose(perm = k_105_perm_0, x = k_103_cast_fp16)[name = string("transpose_76")];
            tensor<fp16, [1, 16, 291, 64]> var_5306_cast_fp16 = mul(x = k_105_cast_fp16, y = var_1091_to_fp16)[name = string("op_5306_cast_fp16")];
            tensor<int32, [4]> x1_71_begin_0 = const()[name = string("x1_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_71_end_0 = const()[name = string("x1_71_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_71_end_mask_0 = const()[name = string("x1_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_71_cast_fp16 = slice_by_index(begin = x1_71_begin_0, end = x1_71_end_0, end_mask = x1_71_end_mask_0, x = k_105_cast_fp16)[name = string("x1_71_cast_fp16")];
            tensor<int32, [4]> x2_71_begin_0 = const()[name = string("x2_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_71_end_0 = const()[name = string("x2_71_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_71_end_mask_0 = const()[name = string("x2_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_71_cast_fp16 = slice_by_index(begin = x2_71_begin_0, end = x2_71_end_0, end_mask = x2_71_end_mask_0, x = k_105_cast_fp16)[name = string("x2_71_cast_fp16")];
            fp16 const_275_promoted_to_fp16 = const()[name = string("const_275_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_5327_cast_fp16 = mul(x = x2_71_cast_fp16, y = const_275_promoted_to_fp16)[name = string("op_5327_cast_fp16")];
            int32 var_5329 = const()[name = string("op_5329"), val = int32(-1)];
            bool var_5330_interleave_0 = const()[name = string("op_5330_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_5330_cast_fp16 = concat(axis = var_5329, interleave = var_5330_interleave_0, values = (var_5327_cast_fp16, x1_71_cast_fp16))[name = string("op_5330_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_5333_cast_fp16 = mul(x = var_5330_cast_fp16, y = var_1118_to_fp16)[name = string("op_5333_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_107_cast_fp16 = add(x = var_5306_cast_fp16, y = var_5333_cast_fp16)[name = string("k_107_cast_fp16")];
            bool var_5339_transpose_x_1 = const()[name = string("op_5339_transpose_x_1"), val = bool(false)];
            bool var_5339_transpose_y_1 = const()[name = string("op_5339_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_5339_cast_fp16 = matmul(transpose_x = var_5339_transpose_x_1, transpose_y = var_5339_transpose_y_1, x = q_107_cast_fp16, y = k_107_cast_fp16)[name = string("op_5339_cast_fp16")];
            fp16 _inversed_scores_103_y_0_to_fp16 = const()[name = string("_inversed_scores_103_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_103_cast_fp16 = mul(x = var_5339_cast_fp16, y = _inversed_scores_103_y_0_to_fp16)[name = string("_inversed_scores_103_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_105_cast_fp16 = add(x = _inversed_scores_103_cast_fp16, y = const_21_to_fp16)[name = string("scores_105_cast_fp16")];
            int32 var_5354 = const()[name = string("op_5354"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_5356_cast_fp16 = softmax(axis = var_5354, x = scores_105_cast_fp16)[name = string("op_5356_cast_fp16")];
            bool attn_out_69_transpose_x_0 = const()[name = string("attn_out_69_transpose_x_0"), val = bool(false)];
            bool attn_out_69_transpose_y_0 = const()[name = string("attn_out_69_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_71_cast_fp16 = transpose(perm = v_71_perm_0, x = v_69_cast_fp16)[name = string("transpose_75")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_69_cast_fp16 = matmul(transpose_x = attn_out_69_transpose_x_0, transpose_y = attn_out_69_transpose_y_0, x = var_5356_cast_fp16, y = v_71_cast_fp16)[name = string("attn_out_69_cast_fp16")];
            tensor<int32, [4]> var_5365_perm_0 = const()[name = string("op_5365_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5367 = const()[name = string("op_5367"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_5365_cast_fp16 = transpose(perm = var_5365_perm_0, x = attn_out_69_cast_fp16)[name = string("transpose_74")];
            tensor<fp16, [1, 291, 1024]> input_213_cast_fp16 = reshape(shape = var_5367, x = var_5365_cast_fp16)[name = string("input_213_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_17_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_17_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(579021376)))];
            tensor<fp16, [1, 291, 1024]> linear_122_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_17_self_attn_o_proj_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_122_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_279_cast_fp16 = add(x = hidden_states_271_cast_fp16, y = linear_122_cast_fp16)[name = string("hidden_states_279_cast_fp16")];
            fp16 var_5377_promoted_to_fp16 = const()[name = string("op_5377_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_5383_cast_fp16 = pow(x = hidden_states_279_cast_fp16, y = var_5377_promoted_to_fp16)[name = string("op_5383_cast_fp16")];
            tensor<int32, [1]> variance_71_axes_0 = const()[name = string("variance_71_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_71_keep_dims_0 = const()[name = string("variance_71_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_71_cast_fp16 = reduce_mean(axes = variance_71_axes_0, keep_dims = variance_71_keep_dims_0, x = var_5383_cast_fp16)[name = string("variance_71_cast_fp16")];
            fp16 var_5386_to_fp16 = const()[name = string("op_5386_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_5387_cast_fp16 = add(x = variance_71_cast_fp16, y = var_5386_to_fp16)[name = string("op_5387_cast_fp16")];
            fp32 var_5388_epsilon_0 = const()[name = string("op_5388_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_5388_cast_fp16 = rsqrt(epsilon = var_5388_epsilon_0, x = var_5387_cast_fp16)[name = string("op_5388_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_283_cast_fp16 = mul(x = hidden_states_279_cast_fp16, y = var_5388_cast_fp16)[name = string("hidden_states_283_cast_fp16")];
            tensor<fp16, [1024]> layers_17_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_17_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(581118592)))];
            tensor<fp16, [1, 291, 1024]> input_215_cast_fp16 = mul(x = layers_17_post_attention_layernorm_weight_to_fp16, y = hidden_states_283_cast_fp16)[name = string("input_215_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_17_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_17_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(581120704)))];
            tensor<fp16, [1, 291, 4096]> linear_123_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_17_mlp_gate_proj_weight_to_fp16, x = input_215_cast_fp16)[name = string("linear_123_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_5401_cast_fp16 = silu(x = linear_123_cast_fp16)[name = string("op_5401_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_17_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_17_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589509376)))];
            tensor<fp16, [1, 291, 4096]> linear_124_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_17_mlp_up_proj_weight_to_fp16, x = input_215_cast_fp16)[name = string("linear_124_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_219_cast_fp16 = mul(x = var_5401_cast_fp16, y = linear_124_cast_fp16)[name = string("input_219_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_17_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_17_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597898048)))];
            tensor<fp16, [1, 291, 1024]> linear_125_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_17_mlp_down_proj_weight_to_fp16, x = input_219_cast_fp16)[name = string("linear_125_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_287_cast_fp16 = add(x = hidden_states_279_cast_fp16, y = linear_125_cast_fp16)[name = string("hidden_states_287_cast_fp16")];
            tensor<int32, [4]> var_5413 = const()[name = string("op_5413"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_5425 = const()[name = string("op_5425"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_5414_cast_fp16 = transpose(perm = var_5413, x = k_107_cast_fp16)[name = string("transpose_73")];
            tensor<fp16, [1, 1024, 1, 291]> input_221_cast_fp16 = reshape(shape = var_5425, x = var_5414_cast_fp16)[name = string("input_221_cast_fp16")];
            tensor<int32, [8]> var_5431_pad_0 = const()[name = string("op_5431_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_5431_mode_0 = const()[name = string("op_5431_mode_0"), val = string("constant")];
            fp16 const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_5431_cast_fp16 = pad(constant_val = const_279_to_fp16, mode = var_5431_mode_0, pad = var_5431_pad_0, x = input_221_cast_fp16)[name = string("op_5431_cast_fp16")];
            tensor<int32, [4]> var_5436 = const()[name = string("op_5436"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_5448 = const()[name = string("op_5448"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_5437_cast_fp16 = transpose(perm = var_5436, x = v_71_cast_fp16)[name = string("transpose_72")];
            tensor<fp16, [1, 1024, 1, 291]> input_223_cast_fp16 = reshape(shape = var_5448, x = var_5437_cast_fp16)[name = string("input_223_cast_fp16")];
            tensor<int32, [8]> var_5454_pad_0 = const()[name = string("op_5454_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_5454_mode_0 = const()[name = string("op_5454_mode_0"), val = string("constant")];
            fp16 const_282_to_fp16 = const()[name = string("const_282_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_5454_cast_fp16 = pad(constant_val = const_282_to_fp16, mode = var_5454_mode_0, pad = var_5454_pad_0, x = input_223_cast_fp16)[name = string("op_5454_cast_fp16")];
            fp16 var_5458_promoted_to_fp16 = const()[name = string("op_5458_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_5464_cast_fp16 = pow(x = hidden_states_287_cast_fp16, y = var_5458_promoted_to_fp16)[name = string("op_5464_cast_fp16")];
            tensor<int32, [1]> variance_73_axes_0 = const()[name = string("variance_73_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_73_keep_dims_0 = const()[name = string("variance_73_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_73_cast_fp16 = reduce_mean(axes = variance_73_axes_0, keep_dims = variance_73_keep_dims_0, x = var_5464_cast_fp16)[name = string("variance_73_cast_fp16")];
            fp16 var_5467_to_fp16 = const()[name = string("op_5467_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_5468_cast_fp16 = add(x = variance_73_cast_fp16, y = var_5467_to_fp16)[name = string("op_5468_cast_fp16")];
            fp32 var_5469_epsilon_0 = const()[name = string("op_5469_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_5469_cast_fp16 = rsqrt(epsilon = var_5469_epsilon_0, x = var_5468_cast_fp16)[name = string("op_5469_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_291_cast_fp16 = mul(x = hidden_states_287_cast_fp16, y = var_5469_cast_fp16)[name = string("hidden_states_291_cast_fp16")];
            tensor<fp16, [1024]> layers_18_input_layernorm_weight_to_fp16 = const()[name = string("layers_18_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(606286720)))];
            tensor<fp16, [1, 291, 1024]> hidden_73_cast_fp16 = mul(x = layers_18_input_layernorm_weight_to_fp16, y = hidden_states_291_cast_fp16)[name = string("hidden_73_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_18_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(606288832)))];
            tensor<fp16, [1, 291, 1024]> linear_126_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_18_self_attn_q_proj_weight_to_fp16, x = hidden_73_cast_fp16)[name = string("linear_126_cast_fp16")];
            tensor<int32, [4]> var_5493 = const()[name = string("op_5493"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_109_cast_fp16 = reshape(shape = var_5493, x = linear_126_cast_fp16)[name = string("q_109_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_18_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(608386048)))];
            tensor<fp16, [1, 291, 1024]> linear_127_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_18_self_attn_k_proj_weight_to_fp16, x = hidden_73_cast_fp16)[name = string("linear_127_cast_fp16")];
            tensor<int32, [4]> var_5500 = const()[name = string("op_5500"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_109_cast_fp16 = reshape(shape = var_5500, x = linear_127_cast_fp16)[name = string("k_109_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_18_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(610483264)))];
            tensor<fp16, [1, 291, 1024]> linear_128_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_18_self_attn_v_proj_weight_to_fp16, x = hidden_73_cast_fp16)[name = string("linear_128_cast_fp16")];
            tensor<int32, [4]> var_5507 = const()[name = string("op_5507"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_73_cast_fp16 = reshape(shape = var_5507, x = linear_128_cast_fp16)[name = string("v_73_cast_fp16")];
            tensor<int32, [4]> q_111_perm_0 = const()[name = string("q_111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_111_perm_0 = const()[name = string("k_111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_75_perm_0 = const()[name = string("v_75_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_111_cast_fp16 = transpose(perm = q_111_perm_0, x = q_109_cast_fp16)[name = string("transpose_71")];
            tensor<fp16, [1, 16, 291, 64]> var_5520_cast_fp16 = mul(x = q_111_cast_fp16, y = var_1091_to_fp16)[name = string("op_5520_cast_fp16")];
            tensor<int32, [4]> x1_73_begin_0 = const()[name = string("x1_73_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_73_end_0 = const()[name = string("x1_73_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_73_end_mask_0 = const()[name = string("x1_73_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_73_cast_fp16 = slice_by_index(begin = x1_73_begin_0, end = x1_73_end_0, end_mask = x1_73_end_mask_0, x = q_111_cast_fp16)[name = string("x1_73_cast_fp16")];
            tensor<int32, [4]> x2_73_begin_0 = const()[name = string("x2_73_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_73_end_0 = const()[name = string("x2_73_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_73_end_mask_0 = const()[name = string("x2_73_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_73_cast_fp16 = slice_by_index(begin = x2_73_begin_0, end = x2_73_end_0, end_mask = x2_73_end_mask_0, x = q_111_cast_fp16)[name = string("x2_73_cast_fp16")];
            fp16 const_287_promoted_to_fp16 = const()[name = string("const_287_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_5541_cast_fp16 = mul(x = x2_73_cast_fp16, y = const_287_promoted_to_fp16)[name = string("op_5541_cast_fp16")];
            int32 var_5543 = const()[name = string("op_5543"), val = int32(-1)];
            bool var_5544_interleave_0 = const()[name = string("op_5544_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_5544_cast_fp16 = concat(axis = var_5543, interleave = var_5544_interleave_0, values = (var_5541_cast_fp16, x1_73_cast_fp16))[name = string("op_5544_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_5547_cast_fp16 = mul(x = var_5544_cast_fp16, y = var_1118_to_fp16)[name = string("op_5547_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_113_cast_fp16 = add(x = var_5520_cast_fp16, y = var_5547_cast_fp16)[name = string("q_113_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_111_cast_fp16 = transpose(perm = k_111_perm_0, x = k_109_cast_fp16)[name = string("transpose_70")];
            tensor<fp16, [1, 16, 291, 64]> var_5552_cast_fp16 = mul(x = k_111_cast_fp16, y = var_1091_to_fp16)[name = string("op_5552_cast_fp16")];
            tensor<int32, [4]> x1_75_begin_0 = const()[name = string("x1_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_75_end_0 = const()[name = string("x1_75_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_75_end_mask_0 = const()[name = string("x1_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_75_cast_fp16 = slice_by_index(begin = x1_75_begin_0, end = x1_75_end_0, end_mask = x1_75_end_mask_0, x = k_111_cast_fp16)[name = string("x1_75_cast_fp16")];
            tensor<int32, [4]> x2_75_begin_0 = const()[name = string("x2_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_75_end_0 = const()[name = string("x2_75_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_75_end_mask_0 = const()[name = string("x2_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_75_cast_fp16 = slice_by_index(begin = x2_75_begin_0, end = x2_75_end_0, end_mask = x2_75_end_mask_0, x = k_111_cast_fp16)[name = string("x2_75_cast_fp16")];
            fp16 const_290_promoted_to_fp16 = const()[name = string("const_290_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_5573_cast_fp16 = mul(x = x2_75_cast_fp16, y = const_290_promoted_to_fp16)[name = string("op_5573_cast_fp16")];
            int32 var_5575 = const()[name = string("op_5575"), val = int32(-1)];
            bool var_5576_interleave_0 = const()[name = string("op_5576_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_5576_cast_fp16 = concat(axis = var_5575, interleave = var_5576_interleave_0, values = (var_5573_cast_fp16, x1_75_cast_fp16))[name = string("op_5576_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_5579_cast_fp16 = mul(x = var_5576_cast_fp16, y = var_1118_to_fp16)[name = string("op_5579_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_113_cast_fp16 = add(x = var_5552_cast_fp16, y = var_5579_cast_fp16)[name = string("k_113_cast_fp16")];
            bool var_5585_transpose_x_1 = const()[name = string("op_5585_transpose_x_1"), val = bool(false)];
            bool var_5585_transpose_y_1 = const()[name = string("op_5585_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_5585_cast_fp16 = matmul(transpose_x = var_5585_transpose_x_1, transpose_y = var_5585_transpose_y_1, x = q_113_cast_fp16, y = k_113_cast_fp16)[name = string("op_5585_cast_fp16")];
            fp16 _inversed_scores_109_y_0_to_fp16 = const()[name = string("_inversed_scores_109_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_109_cast_fp16 = mul(x = var_5585_cast_fp16, y = _inversed_scores_109_y_0_to_fp16)[name = string("_inversed_scores_109_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_111_cast_fp16 = add(x = _inversed_scores_109_cast_fp16, y = const_21_to_fp16)[name = string("scores_111_cast_fp16")];
            int32 var_5600 = const()[name = string("op_5600"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_5602_cast_fp16 = softmax(axis = var_5600, x = scores_111_cast_fp16)[name = string("op_5602_cast_fp16")];
            bool attn_out_73_transpose_x_0 = const()[name = string("attn_out_73_transpose_x_0"), val = bool(false)];
            bool attn_out_73_transpose_y_0 = const()[name = string("attn_out_73_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_75_cast_fp16 = transpose(perm = v_75_perm_0, x = v_73_cast_fp16)[name = string("transpose_69")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_73_cast_fp16 = matmul(transpose_x = attn_out_73_transpose_x_0, transpose_y = attn_out_73_transpose_y_0, x = var_5602_cast_fp16, y = v_75_cast_fp16)[name = string("attn_out_73_cast_fp16")];
            tensor<int32, [4]> var_5611_perm_0 = const()[name = string("op_5611_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5613 = const()[name = string("op_5613"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_5611_cast_fp16 = transpose(perm = var_5611_perm_0, x = attn_out_73_cast_fp16)[name = string("transpose_68")];
            tensor<fp16, [1, 291, 1024]> input_225_cast_fp16 = reshape(shape = var_5613, x = var_5611_cast_fp16)[name = string("input_225_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_18_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_18_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(612580480)))];
            tensor<fp16, [1, 291, 1024]> linear_129_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_18_self_attn_o_proj_weight_to_fp16, x = input_225_cast_fp16)[name = string("linear_129_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_295_cast_fp16 = add(x = hidden_states_287_cast_fp16, y = linear_129_cast_fp16)[name = string("hidden_states_295_cast_fp16")];
            fp16 var_5623_promoted_to_fp16 = const()[name = string("op_5623_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_5629_cast_fp16 = pow(x = hidden_states_295_cast_fp16, y = var_5623_promoted_to_fp16)[name = string("op_5629_cast_fp16")];
            tensor<int32, [1]> variance_75_axes_0 = const()[name = string("variance_75_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_75_keep_dims_0 = const()[name = string("variance_75_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_75_cast_fp16 = reduce_mean(axes = variance_75_axes_0, keep_dims = variance_75_keep_dims_0, x = var_5629_cast_fp16)[name = string("variance_75_cast_fp16")];
            fp16 var_5632_to_fp16 = const()[name = string("op_5632_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_5633_cast_fp16 = add(x = variance_75_cast_fp16, y = var_5632_to_fp16)[name = string("op_5633_cast_fp16")];
            fp32 var_5634_epsilon_0 = const()[name = string("op_5634_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_5634_cast_fp16 = rsqrt(epsilon = var_5634_epsilon_0, x = var_5633_cast_fp16)[name = string("op_5634_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_299_cast_fp16 = mul(x = hidden_states_295_cast_fp16, y = var_5634_cast_fp16)[name = string("hidden_states_299_cast_fp16")];
            tensor<fp16, [1024]> layers_18_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_18_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614677696)))];
            tensor<fp16, [1, 291, 1024]> input_227_cast_fp16 = mul(x = layers_18_post_attention_layernorm_weight_to_fp16, y = hidden_states_299_cast_fp16)[name = string("input_227_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_18_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_18_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614679808)))];
            tensor<fp16, [1, 291, 4096]> linear_130_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_18_mlp_gate_proj_weight_to_fp16, x = input_227_cast_fp16)[name = string("linear_130_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_5647_cast_fp16 = silu(x = linear_130_cast_fp16)[name = string("op_5647_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_18_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_18_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(623068480)))];
            tensor<fp16, [1, 291, 4096]> linear_131_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_18_mlp_up_proj_weight_to_fp16, x = input_227_cast_fp16)[name = string("linear_131_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_231_cast_fp16 = mul(x = var_5647_cast_fp16, y = linear_131_cast_fp16)[name = string("input_231_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_18_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_18_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(631457152)))];
            tensor<fp16, [1, 291, 1024]> linear_132_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_18_mlp_down_proj_weight_to_fp16, x = input_231_cast_fp16)[name = string("linear_132_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_303_cast_fp16 = add(x = hidden_states_295_cast_fp16, y = linear_132_cast_fp16)[name = string("hidden_states_303_cast_fp16")];
            tensor<int32, [4]> var_5659 = const()[name = string("op_5659"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_5671 = const()[name = string("op_5671"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_5660_cast_fp16 = transpose(perm = var_5659, x = k_113_cast_fp16)[name = string("transpose_67")];
            tensor<fp16, [1, 1024, 1, 291]> input_233_cast_fp16 = reshape(shape = var_5671, x = var_5660_cast_fp16)[name = string("input_233_cast_fp16")];
            tensor<int32, [8]> var_5677_pad_0 = const()[name = string("op_5677_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_5677_mode_0 = const()[name = string("op_5677_mode_0"), val = string("constant")];
            fp16 const_294_to_fp16 = const()[name = string("const_294_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_5677_cast_fp16 = pad(constant_val = const_294_to_fp16, mode = var_5677_mode_0, pad = var_5677_pad_0, x = input_233_cast_fp16)[name = string("op_5677_cast_fp16")];
            tensor<int32, [4]> var_5682 = const()[name = string("op_5682"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_5694 = const()[name = string("op_5694"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_5683_cast_fp16 = transpose(perm = var_5682, x = v_75_cast_fp16)[name = string("transpose_66")];
            tensor<fp16, [1, 1024, 1, 291]> input_235_cast_fp16 = reshape(shape = var_5694, x = var_5683_cast_fp16)[name = string("input_235_cast_fp16")];
            tensor<int32, [8]> var_5700_pad_0 = const()[name = string("op_5700_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_5700_mode_0 = const()[name = string("op_5700_mode_0"), val = string("constant")];
            fp16 const_297_to_fp16 = const()[name = string("const_297_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_5700_cast_fp16 = pad(constant_val = const_297_to_fp16, mode = var_5700_mode_0, pad = var_5700_pad_0, x = input_235_cast_fp16)[name = string("op_5700_cast_fp16")];
            fp16 var_5704_promoted_to_fp16 = const()[name = string("op_5704_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_5710_cast_fp16 = pow(x = hidden_states_303_cast_fp16, y = var_5704_promoted_to_fp16)[name = string("op_5710_cast_fp16")];
            tensor<int32, [1]> variance_77_axes_0 = const()[name = string("variance_77_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_77_keep_dims_0 = const()[name = string("variance_77_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_77_cast_fp16 = reduce_mean(axes = variance_77_axes_0, keep_dims = variance_77_keep_dims_0, x = var_5710_cast_fp16)[name = string("variance_77_cast_fp16")];
            fp16 var_5713_to_fp16 = const()[name = string("op_5713_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_5714_cast_fp16 = add(x = variance_77_cast_fp16, y = var_5713_to_fp16)[name = string("op_5714_cast_fp16")];
            fp32 var_5715_epsilon_0 = const()[name = string("op_5715_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_5715_cast_fp16 = rsqrt(epsilon = var_5715_epsilon_0, x = var_5714_cast_fp16)[name = string("op_5715_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_307_cast_fp16 = mul(x = hidden_states_303_cast_fp16, y = var_5715_cast_fp16)[name = string("hidden_states_307_cast_fp16")];
            tensor<fp16, [1024]> layers_19_input_layernorm_weight_to_fp16 = const()[name = string("layers_19_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(639845824)))];
            tensor<fp16, [1, 291, 1024]> hidden_77_cast_fp16 = mul(x = layers_19_input_layernorm_weight_to_fp16, y = hidden_states_307_cast_fp16)[name = string("hidden_77_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_19_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(639847936)))];
            tensor<fp16, [1, 291, 1024]> linear_133_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_19_self_attn_q_proj_weight_to_fp16, x = hidden_77_cast_fp16)[name = string("linear_133_cast_fp16")];
            tensor<int32, [4]> var_5739 = const()[name = string("op_5739"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_115_cast_fp16 = reshape(shape = var_5739, x = linear_133_cast_fp16)[name = string("q_115_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_19_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(641945152)))];
            tensor<fp16, [1, 291, 1024]> linear_134_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_19_self_attn_k_proj_weight_to_fp16, x = hidden_77_cast_fp16)[name = string("linear_134_cast_fp16")];
            tensor<int32, [4]> var_5746 = const()[name = string("op_5746"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_115_cast_fp16 = reshape(shape = var_5746, x = linear_134_cast_fp16)[name = string("k_115_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_19_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(644042368)))];
            tensor<fp16, [1, 291, 1024]> linear_135_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_19_self_attn_v_proj_weight_to_fp16, x = hidden_77_cast_fp16)[name = string("linear_135_cast_fp16")];
            tensor<int32, [4]> var_5753 = const()[name = string("op_5753"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_77_cast_fp16 = reshape(shape = var_5753, x = linear_135_cast_fp16)[name = string("v_77_cast_fp16")];
            tensor<int32, [4]> q_117_perm_0 = const()[name = string("q_117_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_117_perm_0 = const()[name = string("k_117_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_79_perm_0 = const()[name = string("v_79_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_117_cast_fp16 = transpose(perm = q_117_perm_0, x = q_115_cast_fp16)[name = string("transpose_65")];
            tensor<fp16, [1, 16, 291, 64]> var_5766_cast_fp16 = mul(x = q_117_cast_fp16, y = var_1091_to_fp16)[name = string("op_5766_cast_fp16")];
            tensor<int32, [4]> x1_77_begin_0 = const()[name = string("x1_77_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_77_end_0 = const()[name = string("x1_77_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_77_end_mask_0 = const()[name = string("x1_77_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_77_cast_fp16 = slice_by_index(begin = x1_77_begin_0, end = x1_77_end_0, end_mask = x1_77_end_mask_0, x = q_117_cast_fp16)[name = string("x1_77_cast_fp16")];
            tensor<int32, [4]> x2_77_begin_0 = const()[name = string("x2_77_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_77_end_0 = const()[name = string("x2_77_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_77_end_mask_0 = const()[name = string("x2_77_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_77_cast_fp16 = slice_by_index(begin = x2_77_begin_0, end = x2_77_end_0, end_mask = x2_77_end_mask_0, x = q_117_cast_fp16)[name = string("x2_77_cast_fp16")];
            fp16 const_302_promoted_to_fp16 = const()[name = string("const_302_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_5787_cast_fp16 = mul(x = x2_77_cast_fp16, y = const_302_promoted_to_fp16)[name = string("op_5787_cast_fp16")];
            int32 var_5789 = const()[name = string("op_5789"), val = int32(-1)];
            bool var_5790_interleave_0 = const()[name = string("op_5790_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_5790_cast_fp16 = concat(axis = var_5789, interleave = var_5790_interleave_0, values = (var_5787_cast_fp16, x1_77_cast_fp16))[name = string("op_5790_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_5793_cast_fp16 = mul(x = var_5790_cast_fp16, y = var_1118_to_fp16)[name = string("op_5793_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_119_cast_fp16 = add(x = var_5766_cast_fp16, y = var_5793_cast_fp16)[name = string("q_119_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_117_cast_fp16 = transpose(perm = k_117_perm_0, x = k_115_cast_fp16)[name = string("transpose_64")];
            tensor<fp16, [1, 16, 291, 64]> var_5798_cast_fp16 = mul(x = k_117_cast_fp16, y = var_1091_to_fp16)[name = string("op_5798_cast_fp16")];
            tensor<int32, [4]> x1_79_begin_0 = const()[name = string("x1_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_79_end_0 = const()[name = string("x1_79_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_79_end_mask_0 = const()[name = string("x1_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_79_cast_fp16 = slice_by_index(begin = x1_79_begin_0, end = x1_79_end_0, end_mask = x1_79_end_mask_0, x = k_117_cast_fp16)[name = string("x1_79_cast_fp16")];
            tensor<int32, [4]> x2_79_begin_0 = const()[name = string("x2_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_79_end_0 = const()[name = string("x2_79_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_79_end_mask_0 = const()[name = string("x2_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_79_cast_fp16 = slice_by_index(begin = x2_79_begin_0, end = x2_79_end_0, end_mask = x2_79_end_mask_0, x = k_117_cast_fp16)[name = string("x2_79_cast_fp16")];
            fp16 const_305_promoted_to_fp16 = const()[name = string("const_305_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_5819_cast_fp16 = mul(x = x2_79_cast_fp16, y = const_305_promoted_to_fp16)[name = string("op_5819_cast_fp16")];
            int32 var_5821 = const()[name = string("op_5821"), val = int32(-1)];
            bool var_5822_interleave_0 = const()[name = string("op_5822_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_5822_cast_fp16 = concat(axis = var_5821, interleave = var_5822_interleave_0, values = (var_5819_cast_fp16, x1_79_cast_fp16))[name = string("op_5822_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_5825_cast_fp16 = mul(x = var_5822_cast_fp16, y = var_1118_to_fp16)[name = string("op_5825_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_119_cast_fp16 = add(x = var_5798_cast_fp16, y = var_5825_cast_fp16)[name = string("k_119_cast_fp16")];
            bool var_5831_transpose_x_1 = const()[name = string("op_5831_transpose_x_1"), val = bool(false)];
            bool var_5831_transpose_y_1 = const()[name = string("op_5831_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_5831_cast_fp16 = matmul(transpose_x = var_5831_transpose_x_1, transpose_y = var_5831_transpose_y_1, x = q_119_cast_fp16, y = k_119_cast_fp16)[name = string("op_5831_cast_fp16")];
            fp16 _inversed_scores_115_y_0_to_fp16 = const()[name = string("_inversed_scores_115_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_115_cast_fp16 = mul(x = var_5831_cast_fp16, y = _inversed_scores_115_y_0_to_fp16)[name = string("_inversed_scores_115_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_117_cast_fp16 = add(x = _inversed_scores_115_cast_fp16, y = const_21_to_fp16)[name = string("scores_117_cast_fp16")];
            int32 var_5846 = const()[name = string("op_5846"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_5848_cast_fp16 = softmax(axis = var_5846, x = scores_117_cast_fp16)[name = string("op_5848_cast_fp16")];
            bool attn_out_77_transpose_x_0 = const()[name = string("attn_out_77_transpose_x_0"), val = bool(false)];
            bool attn_out_77_transpose_y_0 = const()[name = string("attn_out_77_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_79_cast_fp16 = transpose(perm = v_79_perm_0, x = v_77_cast_fp16)[name = string("transpose_63")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_77_cast_fp16 = matmul(transpose_x = attn_out_77_transpose_x_0, transpose_y = attn_out_77_transpose_y_0, x = var_5848_cast_fp16, y = v_79_cast_fp16)[name = string("attn_out_77_cast_fp16")];
            tensor<int32, [4]> var_5857_perm_0 = const()[name = string("op_5857_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_5859 = const()[name = string("op_5859"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_5857_cast_fp16 = transpose(perm = var_5857_perm_0, x = attn_out_77_cast_fp16)[name = string("transpose_62")];
            tensor<fp16, [1, 291, 1024]> input_237_cast_fp16 = reshape(shape = var_5859, x = var_5857_cast_fp16)[name = string("input_237_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_19_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_19_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(646139584)))];
            tensor<fp16, [1, 291, 1024]> linear_136_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_19_self_attn_o_proj_weight_to_fp16, x = input_237_cast_fp16)[name = string("linear_136_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_311_cast_fp16 = add(x = hidden_states_303_cast_fp16, y = linear_136_cast_fp16)[name = string("hidden_states_311_cast_fp16")];
            fp16 var_5869_promoted_to_fp16 = const()[name = string("op_5869_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_5875_cast_fp16 = pow(x = hidden_states_311_cast_fp16, y = var_5869_promoted_to_fp16)[name = string("op_5875_cast_fp16")];
            tensor<int32, [1]> variance_79_axes_0 = const()[name = string("variance_79_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_79_keep_dims_0 = const()[name = string("variance_79_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_79_cast_fp16 = reduce_mean(axes = variance_79_axes_0, keep_dims = variance_79_keep_dims_0, x = var_5875_cast_fp16)[name = string("variance_79_cast_fp16")];
            fp16 var_5878_to_fp16 = const()[name = string("op_5878_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_5879_cast_fp16 = add(x = variance_79_cast_fp16, y = var_5878_to_fp16)[name = string("op_5879_cast_fp16")];
            fp32 var_5880_epsilon_0 = const()[name = string("op_5880_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_5880_cast_fp16 = rsqrt(epsilon = var_5880_epsilon_0, x = var_5879_cast_fp16)[name = string("op_5880_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_315_cast_fp16 = mul(x = hidden_states_311_cast_fp16, y = var_5880_cast_fp16)[name = string("hidden_states_315_cast_fp16")];
            tensor<fp16, [1024]> layers_19_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_19_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(648236800)))];
            tensor<fp16, [1, 291, 1024]> input_239_cast_fp16 = mul(x = layers_19_post_attention_layernorm_weight_to_fp16, y = hidden_states_315_cast_fp16)[name = string("input_239_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_19_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_19_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(648238912)))];
            tensor<fp16, [1, 291, 4096]> linear_137_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_19_mlp_gate_proj_weight_to_fp16, x = input_239_cast_fp16)[name = string("linear_137_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_5893_cast_fp16 = silu(x = linear_137_cast_fp16)[name = string("op_5893_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_19_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_19_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(656627584)))];
            tensor<fp16, [1, 291, 4096]> linear_138_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_19_mlp_up_proj_weight_to_fp16, x = input_239_cast_fp16)[name = string("linear_138_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_243_cast_fp16 = mul(x = var_5893_cast_fp16, y = linear_138_cast_fp16)[name = string("input_243_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_19_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_19_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(665016256)))];
            tensor<fp16, [1, 291, 1024]> linear_139_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_19_mlp_down_proj_weight_to_fp16, x = input_243_cast_fp16)[name = string("linear_139_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_319_cast_fp16 = add(x = hidden_states_311_cast_fp16, y = linear_139_cast_fp16)[name = string("hidden_states_319_cast_fp16")];
            tensor<int32, [4]> var_5905 = const()[name = string("op_5905"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_5917 = const()[name = string("op_5917"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_5906_cast_fp16 = transpose(perm = var_5905, x = k_119_cast_fp16)[name = string("transpose_61")];
            tensor<fp16, [1, 1024, 1, 291]> input_245_cast_fp16 = reshape(shape = var_5917, x = var_5906_cast_fp16)[name = string("input_245_cast_fp16")];
            tensor<int32, [8]> var_5923_pad_0 = const()[name = string("op_5923_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_5923_mode_0 = const()[name = string("op_5923_mode_0"), val = string("constant")];
            fp16 const_309_to_fp16 = const()[name = string("const_309_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_5923_cast_fp16 = pad(constant_val = const_309_to_fp16, mode = var_5923_mode_0, pad = var_5923_pad_0, x = input_245_cast_fp16)[name = string("op_5923_cast_fp16")];
            tensor<int32, [4]> var_5928 = const()[name = string("op_5928"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_5940 = const()[name = string("op_5940"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_5929_cast_fp16 = transpose(perm = var_5928, x = v_79_cast_fp16)[name = string("transpose_60")];
            tensor<fp16, [1, 1024, 1, 291]> input_247_cast_fp16 = reshape(shape = var_5940, x = var_5929_cast_fp16)[name = string("input_247_cast_fp16")];
            tensor<int32, [8]> var_5946_pad_0 = const()[name = string("op_5946_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_5946_mode_0 = const()[name = string("op_5946_mode_0"), val = string("constant")];
            fp16 const_312_to_fp16 = const()[name = string("const_312_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_5946_cast_fp16 = pad(constant_val = const_312_to_fp16, mode = var_5946_mode_0, pad = var_5946_pad_0, x = input_247_cast_fp16)[name = string("op_5946_cast_fp16")];
            fp16 var_5950_promoted_to_fp16 = const()[name = string("op_5950_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_5956_cast_fp16 = pow(x = hidden_states_319_cast_fp16, y = var_5950_promoted_to_fp16)[name = string("op_5956_cast_fp16")];
            tensor<int32, [1]> variance_81_axes_0 = const()[name = string("variance_81_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_81_keep_dims_0 = const()[name = string("variance_81_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_81_cast_fp16 = reduce_mean(axes = variance_81_axes_0, keep_dims = variance_81_keep_dims_0, x = var_5956_cast_fp16)[name = string("variance_81_cast_fp16")];
            fp16 var_5959_to_fp16 = const()[name = string("op_5959_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_5960_cast_fp16 = add(x = variance_81_cast_fp16, y = var_5959_to_fp16)[name = string("op_5960_cast_fp16")];
            fp32 var_5961_epsilon_0 = const()[name = string("op_5961_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_5961_cast_fp16 = rsqrt(epsilon = var_5961_epsilon_0, x = var_5960_cast_fp16)[name = string("op_5961_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_323_cast_fp16 = mul(x = hidden_states_319_cast_fp16, y = var_5961_cast_fp16)[name = string("hidden_states_323_cast_fp16")];
            tensor<fp16, [1024]> layers_20_input_layernorm_weight_to_fp16 = const()[name = string("layers_20_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(673404928)))];
            tensor<fp16, [1, 291, 1024]> hidden_81_cast_fp16 = mul(x = layers_20_input_layernorm_weight_to_fp16, y = hidden_states_323_cast_fp16)[name = string("hidden_81_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_20_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(673407040)))];
            tensor<fp16, [1, 291, 1024]> linear_140_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_20_self_attn_q_proj_weight_to_fp16, x = hidden_81_cast_fp16)[name = string("linear_140_cast_fp16")];
            tensor<int32, [4]> var_5985 = const()[name = string("op_5985"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_121_cast_fp16 = reshape(shape = var_5985, x = linear_140_cast_fp16)[name = string("q_121_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_20_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(675504256)))];
            tensor<fp16, [1, 291, 1024]> linear_141_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_20_self_attn_k_proj_weight_to_fp16, x = hidden_81_cast_fp16)[name = string("linear_141_cast_fp16")];
            tensor<int32, [4]> var_5992 = const()[name = string("op_5992"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_121_cast_fp16 = reshape(shape = var_5992, x = linear_141_cast_fp16)[name = string("k_121_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_20_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(677601472)))];
            tensor<fp16, [1, 291, 1024]> linear_142_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_20_self_attn_v_proj_weight_to_fp16, x = hidden_81_cast_fp16)[name = string("linear_142_cast_fp16")];
            tensor<int32, [4]> var_5999 = const()[name = string("op_5999"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_81_cast_fp16 = reshape(shape = var_5999, x = linear_142_cast_fp16)[name = string("v_81_cast_fp16")];
            tensor<int32, [4]> q_123_perm_0 = const()[name = string("q_123_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_123_perm_0 = const()[name = string("k_123_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_83_perm_0 = const()[name = string("v_83_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_123_cast_fp16 = transpose(perm = q_123_perm_0, x = q_121_cast_fp16)[name = string("transpose_59")];
            tensor<fp16, [1, 16, 291, 64]> var_6012_cast_fp16 = mul(x = q_123_cast_fp16, y = var_1091_to_fp16)[name = string("op_6012_cast_fp16")];
            tensor<int32, [4]> x1_81_begin_0 = const()[name = string("x1_81_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_81_end_0 = const()[name = string("x1_81_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_81_end_mask_0 = const()[name = string("x1_81_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_81_cast_fp16 = slice_by_index(begin = x1_81_begin_0, end = x1_81_end_0, end_mask = x1_81_end_mask_0, x = q_123_cast_fp16)[name = string("x1_81_cast_fp16")];
            tensor<int32, [4]> x2_81_begin_0 = const()[name = string("x2_81_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_81_end_0 = const()[name = string("x2_81_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_81_end_mask_0 = const()[name = string("x2_81_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_81_cast_fp16 = slice_by_index(begin = x2_81_begin_0, end = x2_81_end_0, end_mask = x2_81_end_mask_0, x = q_123_cast_fp16)[name = string("x2_81_cast_fp16")];
            fp16 const_317_promoted_to_fp16 = const()[name = string("const_317_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_6033_cast_fp16 = mul(x = x2_81_cast_fp16, y = const_317_promoted_to_fp16)[name = string("op_6033_cast_fp16")];
            int32 var_6035 = const()[name = string("op_6035"), val = int32(-1)];
            bool var_6036_interleave_0 = const()[name = string("op_6036_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_6036_cast_fp16 = concat(axis = var_6035, interleave = var_6036_interleave_0, values = (var_6033_cast_fp16, x1_81_cast_fp16))[name = string("op_6036_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_6039_cast_fp16 = mul(x = var_6036_cast_fp16, y = var_1118_to_fp16)[name = string("op_6039_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_125_cast_fp16 = add(x = var_6012_cast_fp16, y = var_6039_cast_fp16)[name = string("q_125_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_123_cast_fp16 = transpose(perm = k_123_perm_0, x = k_121_cast_fp16)[name = string("transpose_58")];
            tensor<fp16, [1, 16, 291, 64]> var_6044_cast_fp16 = mul(x = k_123_cast_fp16, y = var_1091_to_fp16)[name = string("op_6044_cast_fp16")];
            tensor<int32, [4]> x1_83_begin_0 = const()[name = string("x1_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_83_end_0 = const()[name = string("x1_83_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_83_end_mask_0 = const()[name = string("x1_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_83_cast_fp16 = slice_by_index(begin = x1_83_begin_0, end = x1_83_end_0, end_mask = x1_83_end_mask_0, x = k_123_cast_fp16)[name = string("x1_83_cast_fp16")];
            tensor<int32, [4]> x2_83_begin_0 = const()[name = string("x2_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_83_end_0 = const()[name = string("x2_83_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_83_end_mask_0 = const()[name = string("x2_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_83_cast_fp16 = slice_by_index(begin = x2_83_begin_0, end = x2_83_end_0, end_mask = x2_83_end_mask_0, x = k_123_cast_fp16)[name = string("x2_83_cast_fp16")];
            fp16 const_320_promoted_to_fp16 = const()[name = string("const_320_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_6065_cast_fp16 = mul(x = x2_83_cast_fp16, y = const_320_promoted_to_fp16)[name = string("op_6065_cast_fp16")];
            int32 var_6067 = const()[name = string("op_6067"), val = int32(-1)];
            bool var_6068_interleave_0 = const()[name = string("op_6068_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_6068_cast_fp16 = concat(axis = var_6067, interleave = var_6068_interleave_0, values = (var_6065_cast_fp16, x1_83_cast_fp16))[name = string("op_6068_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_6071_cast_fp16 = mul(x = var_6068_cast_fp16, y = var_1118_to_fp16)[name = string("op_6071_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_125_cast_fp16 = add(x = var_6044_cast_fp16, y = var_6071_cast_fp16)[name = string("k_125_cast_fp16")];
            bool var_6077_transpose_x_1 = const()[name = string("op_6077_transpose_x_1"), val = bool(false)];
            bool var_6077_transpose_y_1 = const()[name = string("op_6077_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_6077_cast_fp16 = matmul(transpose_x = var_6077_transpose_x_1, transpose_y = var_6077_transpose_y_1, x = q_125_cast_fp16, y = k_125_cast_fp16)[name = string("op_6077_cast_fp16")];
            fp16 _inversed_scores_121_y_0_to_fp16 = const()[name = string("_inversed_scores_121_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_121_cast_fp16 = mul(x = var_6077_cast_fp16, y = _inversed_scores_121_y_0_to_fp16)[name = string("_inversed_scores_121_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_123_cast_fp16 = add(x = _inversed_scores_121_cast_fp16, y = const_21_to_fp16)[name = string("scores_123_cast_fp16")];
            int32 var_6092 = const()[name = string("op_6092"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_6094_cast_fp16 = softmax(axis = var_6092, x = scores_123_cast_fp16)[name = string("op_6094_cast_fp16")];
            bool attn_out_81_transpose_x_0 = const()[name = string("attn_out_81_transpose_x_0"), val = bool(false)];
            bool attn_out_81_transpose_y_0 = const()[name = string("attn_out_81_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_83_cast_fp16 = transpose(perm = v_83_perm_0, x = v_81_cast_fp16)[name = string("transpose_57")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_81_cast_fp16 = matmul(transpose_x = attn_out_81_transpose_x_0, transpose_y = attn_out_81_transpose_y_0, x = var_6094_cast_fp16, y = v_83_cast_fp16)[name = string("attn_out_81_cast_fp16")];
            tensor<int32, [4]> var_6103_perm_0 = const()[name = string("op_6103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6105 = const()[name = string("op_6105"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_6103_cast_fp16 = transpose(perm = var_6103_perm_0, x = attn_out_81_cast_fp16)[name = string("transpose_56")];
            tensor<fp16, [1, 291, 1024]> input_249_cast_fp16 = reshape(shape = var_6105, x = var_6103_cast_fp16)[name = string("input_249_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_20_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_20_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(679698688)))];
            tensor<fp16, [1, 291, 1024]> linear_143_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_20_self_attn_o_proj_weight_to_fp16, x = input_249_cast_fp16)[name = string("linear_143_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_327_cast_fp16 = add(x = hidden_states_319_cast_fp16, y = linear_143_cast_fp16)[name = string("hidden_states_327_cast_fp16")];
            fp16 var_6115_promoted_to_fp16 = const()[name = string("op_6115_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_6121_cast_fp16 = pow(x = hidden_states_327_cast_fp16, y = var_6115_promoted_to_fp16)[name = string("op_6121_cast_fp16")];
            tensor<int32, [1]> variance_83_axes_0 = const()[name = string("variance_83_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_83_keep_dims_0 = const()[name = string("variance_83_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_83_cast_fp16 = reduce_mean(axes = variance_83_axes_0, keep_dims = variance_83_keep_dims_0, x = var_6121_cast_fp16)[name = string("variance_83_cast_fp16")];
            fp16 var_6124_to_fp16 = const()[name = string("op_6124_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_6125_cast_fp16 = add(x = variance_83_cast_fp16, y = var_6124_to_fp16)[name = string("op_6125_cast_fp16")];
            fp32 var_6126_epsilon_0 = const()[name = string("op_6126_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_6126_cast_fp16 = rsqrt(epsilon = var_6126_epsilon_0, x = var_6125_cast_fp16)[name = string("op_6126_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_331_cast_fp16 = mul(x = hidden_states_327_cast_fp16, y = var_6126_cast_fp16)[name = string("hidden_states_331_cast_fp16")];
            tensor<fp16, [1024]> layers_20_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_20_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(681795904)))];
            tensor<fp16, [1, 291, 1024]> input_251_cast_fp16 = mul(x = layers_20_post_attention_layernorm_weight_to_fp16, y = hidden_states_331_cast_fp16)[name = string("input_251_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_20_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_20_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(681798016)))];
            tensor<fp16, [1, 291, 4096]> linear_144_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_20_mlp_gate_proj_weight_to_fp16, x = input_251_cast_fp16)[name = string("linear_144_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_6139_cast_fp16 = silu(x = linear_144_cast_fp16)[name = string("op_6139_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_20_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_20_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(690186688)))];
            tensor<fp16, [1, 291, 4096]> linear_145_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_20_mlp_up_proj_weight_to_fp16, x = input_251_cast_fp16)[name = string("linear_145_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_255_cast_fp16 = mul(x = var_6139_cast_fp16, y = linear_145_cast_fp16)[name = string("input_255_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_20_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_20_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(698575360)))];
            tensor<fp16, [1, 291, 1024]> linear_146_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_20_mlp_down_proj_weight_to_fp16, x = input_255_cast_fp16)[name = string("linear_146_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_335_cast_fp16 = add(x = hidden_states_327_cast_fp16, y = linear_146_cast_fp16)[name = string("hidden_states_335_cast_fp16")];
            tensor<int32, [4]> var_6151 = const()[name = string("op_6151"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_6163 = const()[name = string("op_6163"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_6152_cast_fp16 = transpose(perm = var_6151, x = k_125_cast_fp16)[name = string("transpose_55")];
            tensor<fp16, [1, 1024, 1, 291]> input_257_cast_fp16 = reshape(shape = var_6163, x = var_6152_cast_fp16)[name = string("input_257_cast_fp16")];
            tensor<int32, [8]> var_6169_pad_0 = const()[name = string("op_6169_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_6169_mode_0 = const()[name = string("op_6169_mode_0"), val = string("constant")];
            fp16 const_324_to_fp16 = const()[name = string("const_324_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_6169_cast_fp16 = pad(constant_val = const_324_to_fp16, mode = var_6169_mode_0, pad = var_6169_pad_0, x = input_257_cast_fp16)[name = string("op_6169_cast_fp16")];
            tensor<int32, [4]> var_6174 = const()[name = string("op_6174"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_6186 = const()[name = string("op_6186"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_6175_cast_fp16 = transpose(perm = var_6174, x = v_83_cast_fp16)[name = string("transpose_54")];
            tensor<fp16, [1, 1024, 1, 291]> input_259_cast_fp16 = reshape(shape = var_6186, x = var_6175_cast_fp16)[name = string("input_259_cast_fp16")];
            tensor<int32, [8]> var_6192_pad_0 = const()[name = string("op_6192_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_6192_mode_0 = const()[name = string("op_6192_mode_0"), val = string("constant")];
            fp16 const_327_to_fp16 = const()[name = string("const_327_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_6192_cast_fp16 = pad(constant_val = const_327_to_fp16, mode = var_6192_mode_0, pad = var_6192_pad_0, x = input_259_cast_fp16)[name = string("op_6192_cast_fp16")];
            fp16 var_6196_promoted_to_fp16 = const()[name = string("op_6196_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_6202_cast_fp16 = pow(x = hidden_states_335_cast_fp16, y = var_6196_promoted_to_fp16)[name = string("op_6202_cast_fp16")];
            tensor<int32, [1]> variance_85_axes_0 = const()[name = string("variance_85_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_85_keep_dims_0 = const()[name = string("variance_85_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_85_cast_fp16 = reduce_mean(axes = variance_85_axes_0, keep_dims = variance_85_keep_dims_0, x = var_6202_cast_fp16)[name = string("variance_85_cast_fp16")];
            fp16 var_6205_to_fp16 = const()[name = string("op_6205_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_6206_cast_fp16 = add(x = variance_85_cast_fp16, y = var_6205_to_fp16)[name = string("op_6206_cast_fp16")];
            fp32 var_6207_epsilon_0 = const()[name = string("op_6207_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_6207_cast_fp16 = rsqrt(epsilon = var_6207_epsilon_0, x = var_6206_cast_fp16)[name = string("op_6207_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_339_cast_fp16 = mul(x = hidden_states_335_cast_fp16, y = var_6207_cast_fp16)[name = string("hidden_states_339_cast_fp16")];
            tensor<fp16, [1024]> layers_21_input_layernorm_weight_to_fp16 = const()[name = string("layers_21_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(706964032)))];
            tensor<fp16, [1, 291, 1024]> hidden_85_cast_fp16 = mul(x = layers_21_input_layernorm_weight_to_fp16, y = hidden_states_339_cast_fp16)[name = string("hidden_85_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_21_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(706966144)))];
            tensor<fp16, [1, 291, 1024]> linear_147_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_21_self_attn_q_proj_weight_to_fp16, x = hidden_85_cast_fp16)[name = string("linear_147_cast_fp16")];
            tensor<int32, [4]> var_6231 = const()[name = string("op_6231"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_127_cast_fp16 = reshape(shape = var_6231, x = linear_147_cast_fp16)[name = string("q_127_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_21_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(709063360)))];
            tensor<fp16, [1, 291, 1024]> linear_148_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_21_self_attn_k_proj_weight_to_fp16, x = hidden_85_cast_fp16)[name = string("linear_148_cast_fp16")];
            tensor<int32, [4]> var_6238 = const()[name = string("op_6238"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_127_cast_fp16 = reshape(shape = var_6238, x = linear_148_cast_fp16)[name = string("k_127_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_21_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(711160576)))];
            tensor<fp16, [1, 291, 1024]> linear_149_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_21_self_attn_v_proj_weight_to_fp16, x = hidden_85_cast_fp16)[name = string("linear_149_cast_fp16")];
            tensor<int32, [4]> var_6245 = const()[name = string("op_6245"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_85_cast_fp16 = reshape(shape = var_6245, x = linear_149_cast_fp16)[name = string("v_85_cast_fp16")];
            tensor<int32, [4]> q_129_perm_0 = const()[name = string("q_129_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_129_perm_0 = const()[name = string("k_129_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_87_perm_0 = const()[name = string("v_87_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_129_cast_fp16 = transpose(perm = q_129_perm_0, x = q_127_cast_fp16)[name = string("transpose_53")];
            tensor<fp16, [1, 16, 291, 64]> var_6258_cast_fp16 = mul(x = q_129_cast_fp16, y = var_1091_to_fp16)[name = string("op_6258_cast_fp16")];
            tensor<int32, [4]> x1_85_begin_0 = const()[name = string("x1_85_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_85_end_0 = const()[name = string("x1_85_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_85_end_mask_0 = const()[name = string("x1_85_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_85_cast_fp16 = slice_by_index(begin = x1_85_begin_0, end = x1_85_end_0, end_mask = x1_85_end_mask_0, x = q_129_cast_fp16)[name = string("x1_85_cast_fp16")];
            tensor<int32, [4]> x2_85_begin_0 = const()[name = string("x2_85_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_85_end_0 = const()[name = string("x2_85_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_85_end_mask_0 = const()[name = string("x2_85_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_85_cast_fp16 = slice_by_index(begin = x2_85_begin_0, end = x2_85_end_0, end_mask = x2_85_end_mask_0, x = q_129_cast_fp16)[name = string("x2_85_cast_fp16")];
            fp16 const_332_promoted_to_fp16 = const()[name = string("const_332_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_6279_cast_fp16 = mul(x = x2_85_cast_fp16, y = const_332_promoted_to_fp16)[name = string("op_6279_cast_fp16")];
            int32 var_6281 = const()[name = string("op_6281"), val = int32(-1)];
            bool var_6282_interleave_0 = const()[name = string("op_6282_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_6282_cast_fp16 = concat(axis = var_6281, interleave = var_6282_interleave_0, values = (var_6279_cast_fp16, x1_85_cast_fp16))[name = string("op_6282_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_6285_cast_fp16 = mul(x = var_6282_cast_fp16, y = var_1118_to_fp16)[name = string("op_6285_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_131_cast_fp16 = add(x = var_6258_cast_fp16, y = var_6285_cast_fp16)[name = string("q_131_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_129_cast_fp16 = transpose(perm = k_129_perm_0, x = k_127_cast_fp16)[name = string("transpose_52")];
            tensor<fp16, [1, 16, 291, 64]> var_6290_cast_fp16 = mul(x = k_129_cast_fp16, y = var_1091_to_fp16)[name = string("op_6290_cast_fp16")];
            tensor<int32, [4]> x1_87_begin_0 = const()[name = string("x1_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_87_end_0 = const()[name = string("x1_87_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_87_end_mask_0 = const()[name = string("x1_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_87_cast_fp16 = slice_by_index(begin = x1_87_begin_0, end = x1_87_end_0, end_mask = x1_87_end_mask_0, x = k_129_cast_fp16)[name = string("x1_87_cast_fp16")];
            tensor<int32, [4]> x2_87_begin_0 = const()[name = string("x2_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_87_end_0 = const()[name = string("x2_87_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_87_end_mask_0 = const()[name = string("x2_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_87_cast_fp16 = slice_by_index(begin = x2_87_begin_0, end = x2_87_end_0, end_mask = x2_87_end_mask_0, x = k_129_cast_fp16)[name = string("x2_87_cast_fp16")];
            fp16 const_335_promoted_to_fp16 = const()[name = string("const_335_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_6311_cast_fp16 = mul(x = x2_87_cast_fp16, y = const_335_promoted_to_fp16)[name = string("op_6311_cast_fp16")];
            int32 var_6313 = const()[name = string("op_6313"), val = int32(-1)];
            bool var_6314_interleave_0 = const()[name = string("op_6314_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_6314_cast_fp16 = concat(axis = var_6313, interleave = var_6314_interleave_0, values = (var_6311_cast_fp16, x1_87_cast_fp16))[name = string("op_6314_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_6317_cast_fp16 = mul(x = var_6314_cast_fp16, y = var_1118_to_fp16)[name = string("op_6317_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_131_cast_fp16 = add(x = var_6290_cast_fp16, y = var_6317_cast_fp16)[name = string("k_131_cast_fp16")];
            bool var_6323_transpose_x_1 = const()[name = string("op_6323_transpose_x_1"), val = bool(false)];
            bool var_6323_transpose_y_1 = const()[name = string("op_6323_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_6323_cast_fp16 = matmul(transpose_x = var_6323_transpose_x_1, transpose_y = var_6323_transpose_y_1, x = q_131_cast_fp16, y = k_131_cast_fp16)[name = string("op_6323_cast_fp16")];
            fp16 _inversed_scores_127_y_0_to_fp16 = const()[name = string("_inversed_scores_127_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_127_cast_fp16 = mul(x = var_6323_cast_fp16, y = _inversed_scores_127_y_0_to_fp16)[name = string("_inversed_scores_127_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_129_cast_fp16 = add(x = _inversed_scores_127_cast_fp16, y = const_21_to_fp16)[name = string("scores_129_cast_fp16")];
            int32 var_6338 = const()[name = string("op_6338"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_6340_cast_fp16 = softmax(axis = var_6338, x = scores_129_cast_fp16)[name = string("op_6340_cast_fp16")];
            bool attn_out_85_transpose_x_0 = const()[name = string("attn_out_85_transpose_x_0"), val = bool(false)];
            bool attn_out_85_transpose_y_0 = const()[name = string("attn_out_85_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_87_cast_fp16 = transpose(perm = v_87_perm_0, x = v_85_cast_fp16)[name = string("transpose_51")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_85_cast_fp16 = matmul(transpose_x = attn_out_85_transpose_x_0, transpose_y = attn_out_85_transpose_y_0, x = var_6340_cast_fp16, y = v_87_cast_fp16)[name = string("attn_out_85_cast_fp16")];
            tensor<int32, [4]> var_6349_perm_0 = const()[name = string("op_6349_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6351 = const()[name = string("op_6351"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_6349_cast_fp16 = transpose(perm = var_6349_perm_0, x = attn_out_85_cast_fp16)[name = string("transpose_50")];
            tensor<fp16, [1, 291, 1024]> input_261_cast_fp16 = reshape(shape = var_6351, x = var_6349_cast_fp16)[name = string("input_261_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_21_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_21_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(713257792)))];
            tensor<fp16, [1, 291, 1024]> linear_150_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_21_self_attn_o_proj_weight_to_fp16, x = input_261_cast_fp16)[name = string("linear_150_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_343_cast_fp16 = add(x = hidden_states_335_cast_fp16, y = linear_150_cast_fp16)[name = string("hidden_states_343_cast_fp16")];
            fp16 var_6361_promoted_to_fp16 = const()[name = string("op_6361_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_6367_cast_fp16 = pow(x = hidden_states_343_cast_fp16, y = var_6361_promoted_to_fp16)[name = string("op_6367_cast_fp16")];
            tensor<int32, [1]> variance_87_axes_0 = const()[name = string("variance_87_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_87_keep_dims_0 = const()[name = string("variance_87_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_87_cast_fp16 = reduce_mean(axes = variance_87_axes_0, keep_dims = variance_87_keep_dims_0, x = var_6367_cast_fp16)[name = string("variance_87_cast_fp16")];
            fp16 var_6370_to_fp16 = const()[name = string("op_6370_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_6371_cast_fp16 = add(x = variance_87_cast_fp16, y = var_6370_to_fp16)[name = string("op_6371_cast_fp16")];
            fp32 var_6372_epsilon_0 = const()[name = string("op_6372_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_6372_cast_fp16 = rsqrt(epsilon = var_6372_epsilon_0, x = var_6371_cast_fp16)[name = string("op_6372_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_347_cast_fp16 = mul(x = hidden_states_343_cast_fp16, y = var_6372_cast_fp16)[name = string("hidden_states_347_cast_fp16")];
            tensor<fp16, [1024]> layers_21_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_21_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(715355008)))];
            tensor<fp16, [1, 291, 1024]> input_263_cast_fp16 = mul(x = layers_21_post_attention_layernorm_weight_to_fp16, y = hidden_states_347_cast_fp16)[name = string("input_263_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_21_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_21_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(715357120)))];
            tensor<fp16, [1, 291, 4096]> linear_151_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_21_mlp_gate_proj_weight_to_fp16, x = input_263_cast_fp16)[name = string("linear_151_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_6385_cast_fp16 = silu(x = linear_151_cast_fp16)[name = string("op_6385_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_21_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_21_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(723745792)))];
            tensor<fp16, [1, 291, 4096]> linear_152_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_21_mlp_up_proj_weight_to_fp16, x = input_263_cast_fp16)[name = string("linear_152_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_267_cast_fp16 = mul(x = var_6385_cast_fp16, y = linear_152_cast_fp16)[name = string("input_267_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_21_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_21_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(732134464)))];
            tensor<fp16, [1, 291, 1024]> linear_153_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_21_mlp_down_proj_weight_to_fp16, x = input_267_cast_fp16)[name = string("linear_153_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_351_cast_fp16 = add(x = hidden_states_343_cast_fp16, y = linear_153_cast_fp16)[name = string("hidden_states_351_cast_fp16")];
            tensor<int32, [4]> var_6397 = const()[name = string("op_6397"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_6409 = const()[name = string("op_6409"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_6398_cast_fp16 = transpose(perm = var_6397, x = k_131_cast_fp16)[name = string("transpose_49")];
            tensor<fp16, [1, 1024, 1, 291]> input_269_cast_fp16 = reshape(shape = var_6409, x = var_6398_cast_fp16)[name = string("input_269_cast_fp16")];
            tensor<int32, [8]> var_6415_pad_0 = const()[name = string("op_6415_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_6415_mode_0 = const()[name = string("op_6415_mode_0"), val = string("constant")];
            fp16 const_339_to_fp16 = const()[name = string("const_339_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_6415_cast_fp16 = pad(constant_val = const_339_to_fp16, mode = var_6415_mode_0, pad = var_6415_pad_0, x = input_269_cast_fp16)[name = string("op_6415_cast_fp16")];
            tensor<int32, [4]> var_6420 = const()[name = string("op_6420"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_6432 = const()[name = string("op_6432"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_6421_cast_fp16 = transpose(perm = var_6420, x = v_87_cast_fp16)[name = string("transpose_48")];
            tensor<fp16, [1, 1024, 1, 291]> input_271_cast_fp16 = reshape(shape = var_6432, x = var_6421_cast_fp16)[name = string("input_271_cast_fp16")];
            tensor<int32, [8]> var_6438_pad_0 = const()[name = string("op_6438_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_6438_mode_0 = const()[name = string("op_6438_mode_0"), val = string("constant")];
            fp16 const_342_to_fp16 = const()[name = string("const_342_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_6438_cast_fp16 = pad(constant_val = const_342_to_fp16, mode = var_6438_mode_0, pad = var_6438_pad_0, x = input_271_cast_fp16)[name = string("op_6438_cast_fp16")];
            fp16 var_6442_promoted_to_fp16 = const()[name = string("op_6442_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_6448_cast_fp16 = pow(x = hidden_states_351_cast_fp16, y = var_6442_promoted_to_fp16)[name = string("op_6448_cast_fp16")];
            tensor<int32, [1]> variance_89_axes_0 = const()[name = string("variance_89_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_89_keep_dims_0 = const()[name = string("variance_89_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_89_cast_fp16 = reduce_mean(axes = variance_89_axes_0, keep_dims = variance_89_keep_dims_0, x = var_6448_cast_fp16)[name = string("variance_89_cast_fp16")];
            fp16 var_6451_to_fp16 = const()[name = string("op_6451_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_6452_cast_fp16 = add(x = variance_89_cast_fp16, y = var_6451_to_fp16)[name = string("op_6452_cast_fp16")];
            fp32 var_6453_epsilon_0 = const()[name = string("op_6453_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_6453_cast_fp16 = rsqrt(epsilon = var_6453_epsilon_0, x = var_6452_cast_fp16)[name = string("op_6453_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_355_cast_fp16 = mul(x = hidden_states_351_cast_fp16, y = var_6453_cast_fp16)[name = string("hidden_states_355_cast_fp16")];
            tensor<fp16, [1024]> layers_22_input_layernorm_weight_to_fp16 = const()[name = string("layers_22_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(740523136)))];
            tensor<fp16, [1, 291, 1024]> hidden_89_cast_fp16 = mul(x = layers_22_input_layernorm_weight_to_fp16, y = hidden_states_355_cast_fp16)[name = string("hidden_89_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_22_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(740525248)))];
            tensor<fp16, [1, 291, 1024]> linear_154_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_22_self_attn_q_proj_weight_to_fp16, x = hidden_89_cast_fp16)[name = string("linear_154_cast_fp16")];
            tensor<int32, [4]> var_6477 = const()[name = string("op_6477"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_133_cast_fp16 = reshape(shape = var_6477, x = linear_154_cast_fp16)[name = string("q_133_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_22_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(742622464)))];
            tensor<fp16, [1, 291, 1024]> linear_155_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_22_self_attn_k_proj_weight_to_fp16, x = hidden_89_cast_fp16)[name = string("linear_155_cast_fp16")];
            tensor<int32, [4]> var_6484 = const()[name = string("op_6484"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_133_cast_fp16 = reshape(shape = var_6484, x = linear_155_cast_fp16)[name = string("k_133_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_22_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(744719680)))];
            tensor<fp16, [1, 291, 1024]> linear_156_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_22_self_attn_v_proj_weight_to_fp16, x = hidden_89_cast_fp16)[name = string("linear_156_cast_fp16")];
            tensor<int32, [4]> var_6491 = const()[name = string("op_6491"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_89_cast_fp16 = reshape(shape = var_6491, x = linear_156_cast_fp16)[name = string("v_89_cast_fp16")];
            tensor<int32, [4]> q_135_perm_0 = const()[name = string("q_135_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_135_perm_0 = const()[name = string("k_135_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_91_perm_0 = const()[name = string("v_91_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_135_cast_fp16 = transpose(perm = q_135_perm_0, x = q_133_cast_fp16)[name = string("transpose_47")];
            tensor<fp16, [1, 16, 291, 64]> var_6504_cast_fp16 = mul(x = q_135_cast_fp16, y = var_1091_to_fp16)[name = string("op_6504_cast_fp16")];
            tensor<int32, [4]> x1_89_begin_0 = const()[name = string("x1_89_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_89_end_0 = const()[name = string("x1_89_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_89_end_mask_0 = const()[name = string("x1_89_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_89_cast_fp16 = slice_by_index(begin = x1_89_begin_0, end = x1_89_end_0, end_mask = x1_89_end_mask_0, x = q_135_cast_fp16)[name = string("x1_89_cast_fp16")];
            tensor<int32, [4]> x2_89_begin_0 = const()[name = string("x2_89_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_89_end_0 = const()[name = string("x2_89_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_89_end_mask_0 = const()[name = string("x2_89_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_89_cast_fp16 = slice_by_index(begin = x2_89_begin_0, end = x2_89_end_0, end_mask = x2_89_end_mask_0, x = q_135_cast_fp16)[name = string("x2_89_cast_fp16")];
            fp16 const_347_promoted_to_fp16 = const()[name = string("const_347_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_6525_cast_fp16 = mul(x = x2_89_cast_fp16, y = const_347_promoted_to_fp16)[name = string("op_6525_cast_fp16")];
            int32 var_6527 = const()[name = string("op_6527"), val = int32(-1)];
            bool var_6528_interleave_0 = const()[name = string("op_6528_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_6528_cast_fp16 = concat(axis = var_6527, interleave = var_6528_interleave_0, values = (var_6525_cast_fp16, x1_89_cast_fp16))[name = string("op_6528_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_6531_cast_fp16 = mul(x = var_6528_cast_fp16, y = var_1118_to_fp16)[name = string("op_6531_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_137_cast_fp16 = add(x = var_6504_cast_fp16, y = var_6531_cast_fp16)[name = string("q_137_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_135_cast_fp16 = transpose(perm = k_135_perm_0, x = k_133_cast_fp16)[name = string("transpose_46")];
            tensor<fp16, [1, 16, 291, 64]> var_6536_cast_fp16 = mul(x = k_135_cast_fp16, y = var_1091_to_fp16)[name = string("op_6536_cast_fp16")];
            tensor<int32, [4]> x1_91_begin_0 = const()[name = string("x1_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_91_end_0 = const()[name = string("x1_91_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_91_end_mask_0 = const()[name = string("x1_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_91_cast_fp16 = slice_by_index(begin = x1_91_begin_0, end = x1_91_end_0, end_mask = x1_91_end_mask_0, x = k_135_cast_fp16)[name = string("x1_91_cast_fp16")];
            tensor<int32, [4]> x2_91_begin_0 = const()[name = string("x2_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_91_end_0 = const()[name = string("x2_91_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_91_end_mask_0 = const()[name = string("x2_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_91_cast_fp16 = slice_by_index(begin = x2_91_begin_0, end = x2_91_end_0, end_mask = x2_91_end_mask_0, x = k_135_cast_fp16)[name = string("x2_91_cast_fp16")];
            fp16 const_350_promoted_to_fp16 = const()[name = string("const_350_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_6557_cast_fp16 = mul(x = x2_91_cast_fp16, y = const_350_promoted_to_fp16)[name = string("op_6557_cast_fp16")];
            int32 var_6559 = const()[name = string("op_6559"), val = int32(-1)];
            bool var_6560_interleave_0 = const()[name = string("op_6560_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_6560_cast_fp16 = concat(axis = var_6559, interleave = var_6560_interleave_0, values = (var_6557_cast_fp16, x1_91_cast_fp16))[name = string("op_6560_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_6563_cast_fp16 = mul(x = var_6560_cast_fp16, y = var_1118_to_fp16)[name = string("op_6563_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_137_cast_fp16 = add(x = var_6536_cast_fp16, y = var_6563_cast_fp16)[name = string("k_137_cast_fp16")];
            bool var_6569_transpose_x_1 = const()[name = string("op_6569_transpose_x_1"), val = bool(false)];
            bool var_6569_transpose_y_1 = const()[name = string("op_6569_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_6569_cast_fp16 = matmul(transpose_x = var_6569_transpose_x_1, transpose_y = var_6569_transpose_y_1, x = q_137_cast_fp16, y = k_137_cast_fp16)[name = string("op_6569_cast_fp16")];
            fp16 _inversed_scores_133_y_0_to_fp16 = const()[name = string("_inversed_scores_133_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_133_cast_fp16 = mul(x = var_6569_cast_fp16, y = _inversed_scores_133_y_0_to_fp16)[name = string("_inversed_scores_133_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_135_cast_fp16 = add(x = _inversed_scores_133_cast_fp16, y = const_21_to_fp16)[name = string("scores_135_cast_fp16")];
            int32 var_6584 = const()[name = string("op_6584"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_6586_cast_fp16 = softmax(axis = var_6584, x = scores_135_cast_fp16)[name = string("op_6586_cast_fp16")];
            bool attn_out_89_transpose_x_0 = const()[name = string("attn_out_89_transpose_x_0"), val = bool(false)];
            bool attn_out_89_transpose_y_0 = const()[name = string("attn_out_89_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_91_cast_fp16 = transpose(perm = v_91_perm_0, x = v_89_cast_fp16)[name = string("transpose_45")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_89_cast_fp16 = matmul(transpose_x = attn_out_89_transpose_x_0, transpose_y = attn_out_89_transpose_y_0, x = var_6586_cast_fp16, y = v_91_cast_fp16)[name = string("attn_out_89_cast_fp16")];
            tensor<int32, [4]> var_6595_perm_0 = const()[name = string("op_6595_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6597 = const()[name = string("op_6597"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_6595_cast_fp16 = transpose(perm = var_6595_perm_0, x = attn_out_89_cast_fp16)[name = string("transpose_44")];
            tensor<fp16, [1, 291, 1024]> input_273_cast_fp16 = reshape(shape = var_6597, x = var_6595_cast_fp16)[name = string("input_273_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_22_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_22_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(746816896)))];
            tensor<fp16, [1, 291, 1024]> linear_157_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_22_self_attn_o_proj_weight_to_fp16, x = input_273_cast_fp16)[name = string("linear_157_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_359_cast_fp16 = add(x = hidden_states_351_cast_fp16, y = linear_157_cast_fp16)[name = string("hidden_states_359_cast_fp16")];
            fp16 var_6607_promoted_to_fp16 = const()[name = string("op_6607_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_6613_cast_fp16 = pow(x = hidden_states_359_cast_fp16, y = var_6607_promoted_to_fp16)[name = string("op_6613_cast_fp16")];
            tensor<int32, [1]> variance_91_axes_0 = const()[name = string("variance_91_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_91_keep_dims_0 = const()[name = string("variance_91_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_91_cast_fp16 = reduce_mean(axes = variance_91_axes_0, keep_dims = variance_91_keep_dims_0, x = var_6613_cast_fp16)[name = string("variance_91_cast_fp16")];
            fp16 var_6616_to_fp16 = const()[name = string("op_6616_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_6617_cast_fp16 = add(x = variance_91_cast_fp16, y = var_6616_to_fp16)[name = string("op_6617_cast_fp16")];
            fp32 var_6618_epsilon_0 = const()[name = string("op_6618_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_6618_cast_fp16 = rsqrt(epsilon = var_6618_epsilon_0, x = var_6617_cast_fp16)[name = string("op_6618_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_363_cast_fp16 = mul(x = hidden_states_359_cast_fp16, y = var_6618_cast_fp16)[name = string("hidden_states_363_cast_fp16")];
            tensor<fp16, [1024]> layers_22_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_22_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(748914112)))];
            tensor<fp16, [1, 291, 1024]> input_275_cast_fp16 = mul(x = layers_22_post_attention_layernorm_weight_to_fp16, y = hidden_states_363_cast_fp16)[name = string("input_275_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_22_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_22_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(748916224)))];
            tensor<fp16, [1, 291, 4096]> linear_158_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_22_mlp_gate_proj_weight_to_fp16, x = input_275_cast_fp16)[name = string("linear_158_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_6631_cast_fp16 = silu(x = linear_158_cast_fp16)[name = string("op_6631_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_22_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_22_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(757304896)))];
            tensor<fp16, [1, 291, 4096]> linear_159_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_22_mlp_up_proj_weight_to_fp16, x = input_275_cast_fp16)[name = string("linear_159_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_279_cast_fp16 = mul(x = var_6631_cast_fp16, y = linear_159_cast_fp16)[name = string("input_279_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_22_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_22_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(765693568)))];
            tensor<fp16, [1, 291, 1024]> linear_160_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_22_mlp_down_proj_weight_to_fp16, x = input_279_cast_fp16)[name = string("linear_160_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_367_cast_fp16 = add(x = hidden_states_359_cast_fp16, y = linear_160_cast_fp16)[name = string("hidden_states_367_cast_fp16")];
            tensor<int32, [4]> var_6643 = const()[name = string("op_6643"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_6655 = const()[name = string("op_6655"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_6644_cast_fp16 = transpose(perm = var_6643, x = k_137_cast_fp16)[name = string("transpose_43")];
            tensor<fp16, [1, 1024, 1, 291]> input_281_cast_fp16 = reshape(shape = var_6655, x = var_6644_cast_fp16)[name = string("input_281_cast_fp16")];
            tensor<int32, [8]> var_6661_pad_0 = const()[name = string("op_6661_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_6661_mode_0 = const()[name = string("op_6661_mode_0"), val = string("constant")];
            fp16 const_354_to_fp16 = const()[name = string("const_354_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_6661_cast_fp16 = pad(constant_val = const_354_to_fp16, mode = var_6661_mode_0, pad = var_6661_pad_0, x = input_281_cast_fp16)[name = string("op_6661_cast_fp16")];
            tensor<int32, [4]> var_6666 = const()[name = string("op_6666"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_6678 = const()[name = string("op_6678"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_6667_cast_fp16 = transpose(perm = var_6666, x = v_91_cast_fp16)[name = string("transpose_42")];
            tensor<fp16, [1, 1024, 1, 291]> input_283_cast_fp16 = reshape(shape = var_6678, x = var_6667_cast_fp16)[name = string("input_283_cast_fp16")];
            tensor<int32, [8]> var_6684_pad_0 = const()[name = string("op_6684_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_6684_mode_0 = const()[name = string("op_6684_mode_0"), val = string("constant")];
            fp16 const_357_to_fp16 = const()[name = string("const_357_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_6684_cast_fp16 = pad(constant_val = const_357_to_fp16, mode = var_6684_mode_0, pad = var_6684_pad_0, x = input_283_cast_fp16)[name = string("op_6684_cast_fp16")];
            fp16 var_6688_promoted_to_fp16 = const()[name = string("op_6688_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_6694_cast_fp16 = pow(x = hidden_states_367_cast_fp16, y = var_6688_promoted_to_fp16)[name = string("op_6694_cast_fp16")];
            tensor<int32, [1]> variance_93_axes_0 = const()[name = string("variance_93_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_93_keep_dims_0 = const()[name = string("variance_93_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_93_cast_fp16 = reduce_mean(axes = variance_93_axes_0, keep_dims = variance_93_keep_dims_0, x = var_6694_cast_fp16)[name = string("variance_93_cast_fp16")];
            fp16 var_6697_to_fp16 = const()[name = string("op_6697_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_6698_cast_fp16 = add(x = variance_93_cast_fp16, y = var_6697_to_fp16)[name = string("op_6698_cast_fp16")];
            fp32 var_6699_epsilon_0 = const()[name = string("op_6699_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_6699_cast_fp16 = rsqrt(epsilon = var_6699_epsilon_0, x = var_6698_cast_fp16)[name = string("op_6699_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_371_cast_fp16 = mul(x = hidden_states_367_cast_fp16, y = var_6699_cast_fp16)[name = string("hidden_states_371_cast_fp16")];
            tensor<fp16, [1024]> layers_23_input_layernorm_weight_to_fp16 = const()[name = string("layers_23_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(774082240)))];
            tensor<fp16, [1, 291, 1024]> hidden_93_cast_fp16 = mul(x = layers_23_input_layernorm_weight_to_fp16, y = hidden_states_371_cast_fp16)[name = string("hidden_93_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_23_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(774084352)))];
            tensor<fp16, [1, 291, 1024]> linear_161_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_23_self_attn_q_proj_weight_to_fp16, x = hidden_93_cast_fp16)[name = string("linear_161_cast_fp16")];
            tensor<int32, [4]> var_6723 = const()[name = string("op_6723"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_139_cast_fp16 = reshape(shape = var_6723, x = linear_161_cast_fp16)[name = string("q_139_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_23_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(776181568)))];
            tensor<fp16, [1, 291, 1024]> linear_162_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_23_self_attn_k_proj_weight_to_fp16, x = hidden_93_cast_fp16)[name = string("linear_162_cast_fp16")];
            tensor<int32, [4]> var_6730 = const()[name = string("op_6730"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_139_cast_fp16 = reshape(shape = var_6730, x = linear_162_cast_fp16)[name = string("k_139_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_23_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(778278784)))];
            tensor<fp16, [1, 291, 1024]> linear_163_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_23_self_attn_v_proj_weight_to_fp16, x = hidden_93_cast_fp16)[name = string("linear_163_cast_fp16")];
            tensor<int32, [4]> var_6737 = const()[name = string("op_6737"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_93_cast_fp16 = reshape(shape = var_6737, x = linear_163_cast_fp16)[name = string("v_93_cast_fp16")];
            tensor<int32, [4]> q_141_perm_0 = const()[name = string("q_141_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_141_perm_0 = const()[name = string("k_141_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_95_perm_0 = const()[name = string("v_95_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_141_cast_fp16 = transpose(perm = q_141_perm_0, x = q_139_cast_fp16)[name = string("transpose_41")];
            tensor<fp16, [1, 16, 291, 64]> var_6750_cast_fp16 = mul(x = q_141_cast_fp16, y = var_1091_to_fp16)[name = string("op_6750_cast_fp16")];
            tensor<int32, [4]> x1_93_begin_0 = const()[name = string("x1_93_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_93_end_0 = const()[name = string("x1_93_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_93_end_mask_0 = const()[name = string("x1_93_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_93_cast_fp16 = slice_by_index(begin = x1_93_begin_0, end = x1_93_end_0, end_mask = x1_93_end_mask_0, x = q_141_cast_fp16)[name = string("x1_93_cast_fp16")];
            tensor<int32, [4]> x2_93_begin_0 = const()[name = string("x2_93_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_93_end_0 = const()[name = string("x2_93_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_93_end_mask_0 = const()[name = string("x2_93_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_93_cast_fp16 = slice_by_index(begin = x2_93_begin_0, end = x2_93_end_0, end_mask = x2_93_end_mask_0, x = q_141_cast_fp16)[name = string("x2_93_cast_fp16")];
            fp16 const_362_promoted_to_fp16 = const()[name = string("const_362_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_6771_cast_fp16 = mul(x = x2_93_cast_fp16, y = const_362_promoted_to_fp16)[name = string("op_6771_cast_fp16")];
            int32 var_6773 = const()[name = string("op_6773"), val = int32(-1)];
            bool var_6774_interleave_0 = const()[name = string("op_6774_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_6774_cast_fp16 = concat(axis = var_6773, interleave = var_6774_interleave_0, values = (var_6771_cast_fp16, x1_93_cast_fp16))[name = string("op_6774_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_6777_cast_fp16 = mul(x = var_6774_cast_fp16, y = var_1118_to_fp16)[name = string("op_6777_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_143_cast_fp16 = add(x = var_6750_cast_fp16, y = var_6777_cast_fp16)[name = string("q_143_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_141_cast_fp16 = transpose(perm = k_141_perm_0, x = k_139_cast_fp16)[name = string("transpose_40")];
            tensor<fp16, [1, 16, 291, 64]> var_6782_cast_fp16 = mul(x = k_141_cast_fp16, y = var_1091_to_fp16)[name = string("op_6782_cast_fp16")];
            tensor<int32, [4]> x1_95_begin_0 = const()[name = string("x1_95_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_95_end_0 = const()[name = string("x1_95_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_95_end_mask_0 = const()[name = string("x1_95_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_95_cast_fp16 = slice_by_index(begin = x1_95_begin_0, end = x1_95_end_0, end_mask = x1_95_end_mask_0, x = k_141_cast_fp16)[name = string("x1_95_cast_fp16")];
            tensor<int32, [4]> x2_95_begin_0 = const()[name = string("x2_95_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_95_end_0 = const()[name = string("x2_95_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_95_end_mask_0 = const()[name = string("x2_95_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_95_cast_fp16 = slice_by_index(begin = x2_95_begin_0, end = x2_95_end_0, end_mask = x2_95_end_mask_0, x = k_141_cast_fp16)[name = string("x2_95_cast_fp16")];
            fp16 const_365_promoted_to_fp16 = const()[name = string("const_365_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_6803_cast_fp16 = mul(x = x2_95_cast_fp16, y = const_365_promoted_to_fp16)[name = string("op_6803_cast_fp16")];
            int32 var_6805 = const()[name = string("op_6805"), val = int32(-1)];
            bool var_6806_interleave_0 = const()[name = string("op_6806_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_6806_cast_fp16 = concat(axis = var_6805, interleave = var_6806_interleave_0, values = (var_6803_cast_fp16, x1_95_cast_fp16))[name = string("op_6806_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_6809_cast_fp16 = mul(x = var_6806_cast_fp16, y = var_1118_to_fp16)[name = string("op_6809_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_143_cast_fp16 = add(x = var_6782_cast_fp16, y = var_6809_cast_fp16)[name = string("k_143_cast_fp16")];
            bool var_6815_transpose_x_1 = const()[name = string("op_6815_transpose_x_1"), val = bool(false)];
            bool var_6815_transpose_y_1 = const()[name = string("op_6815_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_6815_cast_fp16 = matmul(transpose_x = var_6815_transpose_x_1, transpose_y = var_6815_transpose_y_1, x = q_143_cast_fp16, y = k_143_cast_fp16)[name = string("op_6815_cast_fp16")];
            fp16 _inversed_scores_139_y_0_to_fp16 = const()[name = string("_inversed_scores_139_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_139_cast_fp16 = mul(x = var_6815_cast_fp16, y = _inversed_scores_139_y_0_to_fp16)[name = string("_inversed_scores_139_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_141_cast_fp16 = add(x = _inversed_scores_139_cast_fp16, y = const_21_to_fp16)[name = string("scores_141_cast_fp16")];
            int32 var_6830 = const()[name = string("op_6830"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_6832_cast_fp16 = softmax(axis = var_6830, x = scores_141_cast_fp16)[name = string("op_6832_cast_fp16")];
            bool attn_out_93_transpose_x_0 = const()[name = string("attn_out_93_transpose_x_0"), val = bool(false)];
            bool attn_out_93_transpose_y_0 = const()[name = string("attn_out_93_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_95_cast_fp16 = transpose(perm = v_95_perm_0, x = v_93_cast_fp16)[name = string("transpose_39")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_93_cast_fp16 = matmul(transpose_x = attn_out_93_transpose_x_0, transpose_y = attn_out_93_transpose_y_0, x = var_6832_cast_fp16, y = v_95_cast_fp16)[name = string("attn_out_93_cast_fp16")];
            tensor<int32, [4]> var_6841_perm_0 = const()[name = string("op_6841_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_6843 = const()[name = string("op_6843"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_6841_cast_fp16 = transpose(perm = var_6841_perm_0, x = attn_out_93_cast_fp16)[name = string("transpose_38")];
            tensor<fp16, [1, 291, 1024]> input_285_cast_fp16 = reshape(shape = var_6843, x = var_6841_cast_fp16)[name = string("input_285_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_23_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_23_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(780376000)))];
            tensor<fp16, [1, 291, 1024]> linear_164_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_23_self_attn_o_proj_weight_to_fp16, x = input_285_cast_fp16)[name = string("linear_164_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_375_cast_fp16 = add(x = hidden_states_367_cast_fp16, y = linear_164_cast_fp16)[name = string("hidden_states_375_cast_fp16")];
            fp16 var_6853_promoted_to_fp16 = const()[name = string("op_6853_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_6859_cast_fp16 = pow(x = hidden_states_375_cast_fp16, y = var_6853_promoted_to_fp16)[name = string("op_6859_cast_fp16")];
            tensor<int32, [1]> variance_95_axes_0 = const()[name = string("variance_95_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_95_keep_dims_0 = const()[name = string("variance_95_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_95_cast_fp16 = reduce_mean(axes = variance_95_axes_0, keep_dims = variance_95_keep_dims_0, x = var_6859_cast_fp16)[name = string("variance_95_cast_fp16")];
            fp16 var_6862_to_fp16 = const()[name = string("op_6862_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_6863_cast_fp16 = add(x = variance_95_cast_fp16, y = var_6862_to_fp16)[name = string("op_6863_cast_fp16")];
            fp32 var_6864_epsilon_0 = const()[name = string("op_6864_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_6864_cast_fp16 = rsqrt(epsilon = var_6864_epsilon_0, x = var_6863_cast_fp16)[name = string("op_6864_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_379_cast_fp16 = mul(x = hidden_states_375_cast_fp16, y = var_6864_cast_fp16)[name = string("hidden_states_379_cast_fp16")];
            tensor<fp16, [1024]> layers_23_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_23_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(782473216)))];
            tensor<fp16, [1, 291, 1024]> input_287_cast_fp16 = mul(x = layers_23_post_attention_layernorm_weight_to_fp16, y = hidden_states_379_cast_fp16)[name = string("input_287_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_23_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_23_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(782475328)))];
            tensor<fp16, [1, 291, 4096]> linear_165_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_23_mlp_gate_proj_weight_to_fp16, x = input_287_cast_fp16)[name = string("linear_165_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_6877_cast_fp16 = silu(x = linear_165_cast_fp16)[name = string("op_6877_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_23_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_23_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(790864000)))];
            tensor<fp16, [1, 291, 4096]> linear_166_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_23_mlp_up_proj_weight_to_fp16, x = input_287_cast_fp16)[name = string("linear_166_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_291_cast_fp16 = mul(x = var_6877_cast_fp16, y = linear_166_cast_fp16)[name = string("input_291_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_23_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_23_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(799252672)))];
            tensor<fp16, [1, 291, 1024]> linear_167_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_23_mlp_down_proj_weight_to_fp16, x = input_291_cast_fp16)[name = string("linear_167_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_383_cast_fp16 = add(x = hidden_states_375_cast_fp16, y = linear_167_cast_fp16)[name = string("hidden_states_383_cast_fp16")];
            tensor<int32, [4]> var_6889 = const()[name = string("op_6889"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_6901 = const()[name = string("op_6901"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_6890_cast_fp16 = transpose(perm = var_6889, x = k_143_cast_fp16)[name = string("transpose_37")];
            tensor<fp16, [1, 1024, 1, 291]> input_293_cast_fp16 = reshape(shape = var_6901, x = var_6890_cast_fp16)[name = string("input_293_cast_fp16")];
            tensor<int32, [8]> var_6907_pad_0 = const()[name = string("op_6907_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_6907_mode_0 = const()[name = string("op_6907_mode_0"), val = string("constant")];
            fp16 const_369_to_fp16 = const()[name = string("const_369_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_6907_cast_fp16 = pad(constant_val = const_369_to_fp16, mode = var_6907_mode_0, pad = var_6907_pad_0, x = input_293_cast_fp16)[name = string("op_6907_cast_fp16")];
            tensor<int32, [4]> var_6912 = const()[name = string("op_6912"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_6924 = const()[name = string("op_6924"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_6913_cast_fp16 = transpose(perm = var_6912, x = v_95_cast_fp16)[name = string("transpose_36")];
            tensor<fp16, [1, 1024, 1, 291]> input_295_cast_fp16 = reshape(shape = var_6924, x = var_6913_cast_fp16)[name = string("input_295_cast_fp16")];
            tensor<int32, [8]> var_6930_pad_0 = const()[name = string("op_6930_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_6930_mode_0 = const()[name = string("op_6930_mode_0"), val = string("constant")];
            fp16 const_372_to_fp16 = const()[name = string("const_372_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_6930_cast_fp16 = pad(constant_val = const_372_to_fp16, mode = var_6930_mode_0, pad = var_6930_pad_0, x = input_295_cast_fp16)[name = string("op_6930_cast_fp16")];
            fp16 var_6934_promoted_to_fp16 = const()[name = string("op_6934_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_6940_cast_fp16 = pow(x = hidden_states_383_cast_fp16, y = var_6934_promoted_to_fp16)[name = string("op_6940_cast_fp16")];
            tensor<int32, [1]> variance_97_axes_0 = const()[name = string("variance_97_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_97_keep_dims_0 = const()[name = string("variance_97_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_97_cast_fp16 = reduce_mean(axes = variance_97_axes_0, keep_dims = variance_97_keep_dims_0, x = var_6940_cast_fp16)[name = string("variance_97_cast_fp16")];
            fp16 var_6943_to_fp16 = const()[name = string("op_6943_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_6944_cast_fp16 = add(x = variance_97_cast_fp16, y = var_6943_to_fp16)[name = string("op_6944_cast_fp16")];
            fp32 var_6945_epsilon_0 = const()[name = string("op_6945_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_6945_cast_fp16 = rsqrt(epsilon = var_6945_epsilon_0, x = var_6944_cast_fp16)[name = string("op_6945_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_387_cast_fp16 = mul(x = hidden_states_383_cast_fp16, y = var_6945_cast_fp16)[name = string("hidden_states_387_cast_fp16")];
            tensor<fp16, [1024]> layers_24_input_layernorm_weight_to_fp16 = const()[name = string("layers_24_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(807641344)))];
            tensor<fp16, [1, 291, 1024]> hidden_97_cast_fp16 = mul(x = layers_24_input_layernorm_weight_to_fp16, y = hidden_states_387_cast_fp16)[name = string("hidden_97_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_24_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(807643456)))];
            tensor<fp16, [1, 291, 1024]> linear_168_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_24_self_attn_q_proj_weight_to_fp16, x = hidden_97_cast_fp16)[name = string("linear_168_cast_fp16")];
            tensor<int32, [4]> var_6969 = const()[name = string("op_6969"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_145_cast_fp16 = reshape(shape = var_6969, x = linear_168_cast_fp16)[name = string("q_145_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_24_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(809740672)))];
            tensor<fp16, [1, 291, 1024]> linear_169_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_24_self_attn_k_proj_weight_to_fp16, x = hidden_97_cast_fp16)[name = string("linear_169_cast_fp16")];
            tensor<int32, [4]> var_6976 = const()[name = string("op_6976"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_145_cast_fp16 = reshape(shape = var_6976, x = linear_169_cast_fp16)[name = string("k_145_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_24_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(811837888)))];
            tensor<fp16, [1, 291, 1024]> linear_170_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_24_self_attn_v_proj_weight_to_fp16, x = hidden_97_cast_fp16)[name = string("linear_170_cast_fp16")];
            tensor<int32, [4]> var_6983 = const()[name = string("op_6983"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_97_cast_fp16 = reshape(shape = var_6983, x = linear_170_cast_fp16)[name = string("v_97_cast_fp16")];
            tensor<int32, [4]> q_147_perm_0 = const()[name = string("q_147_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_147_perm_0 = const()[name = string("k_147_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_99_perm_0 = const()[name = string("v_99_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_147_cast_fp16 = transpose(perm = q_147_perm_0, x = q_145_cast_fp16)[name = string("transpose_35")];
            tensor<fp16, [1, 16, 291, 64]> var_6996_cast_fp16 = mul(x = q_147_cast_fp16, y = var_1091_to_fp16)[name = string("op_6996_cast_fp16")];
            tensor<int32, [4]> x1_97_begin_0 = const()[name = string("x1_97_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_97_end_0 = const()[name = string("x1_97_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_97_end_mask_0 = const()[name = string("x1_97_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_97_cast_fp16 = slice_by_index(begin = x1_97_begin_0, end = x1_97_end_0, end_mask = x1_97_end_mask_0, x = q_147_cast_fp16)[name = string("x1_97_cast_fp16")];
            tensor<int32, [4]> x2_97_begin_0 = const()[name = string("x2_97_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_97_end_0 = const()[name = string("x2_97_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_97_end_mask_0 = const()[name = string("x2_97_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_97_cast_fp16 = slice_by_index(begin = x2_97_begin_0, end = x2_97_end_0, end_mask = x2_97_end_mask_0, x = q_147_cast_fp16)[name = string("x2_97_cast_fp16")];
            fp16 const_377_promoted_to_fp16 = const()[name = string("const_377_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_7017_cast_fp16 = mul(x = x2_97_cast_fp16, y = const_377_promoted_to_fp16)[name = string("op_7017_cast_fp16")];
            int32 var_7019 = const()[name = string("op_7019"), val = int32(-1)];
            bool var_7020_interleave_0 = const()[name = string("op_7020_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_7020_cast_fp16 = concat(axis = var_7019, interleave = var_7020_interleave_0, values = (var_7017_cast_fp16, x1_97_cast_fp16))[name = string("op_7020_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_7023_cast_fp16 = mul(x = var_7020_cast_fp16, y = var_1118_to_fp16)[name = string("op_7023_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_149_cast_fp16 = add(x = var_6996_cast_fp16, y = var_7023_cast_fp16)[name = string("q_149_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_147_cast_fp16 = transpose(perm = k_147_perm_0, x = k_145_cast_fp16)[name = string("transpose_34")];
            tensor<fp16, [1, 16, 291, 64]> var_7028_cast_fp16 = mul(x = k_147_cast_fp16, y = var_1091_to_fp16)[name = string("op_7028_cast_fp16")];
            tensor<int32, [4]> x1_99_begin_0 = const()[name = string("x1_99_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_99_end_0 = const()[name = string("x1_99_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_99_end_mask_0 = const()[name = string("x1_99_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_99_cast_fp16 = slice_by_index(begin = x1_99_begin_0, end = x1_99_end_0, end_mask = x1_99_end_mask_0, x = k_147_cast_fp16)[name = string("x1_99_cast_fp16")];
            tensor<int32, [4]> x2_99_begin_0 = const()[name = string("x2_99_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_99_end_0 = const()[name = string("x2_99_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_99_end_mask_0 = const()[name = string("x2_99_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_99_cast_fp16 = slice_by_index(begin = x2_99_begin_0, end = x2_99_end_0, end_mask = x2_99_end_mask_0, x = k_147_cast_fp16)[name = string("x2_99_cast_fp16")];
            fp16 const_380_promoted_to_fp16 = const()[name = string("const_380_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_7049_cast_fp16 = mul(x = x2_99_cast_fp16, y = const_380_promoted_to_fp16)[name = string("op_7049_cast_fp16")];
            int32 var_7051 = const()[name = string("op_7051"), val = int32(-1)];
            bool var_7052_interleave_0 = const()[name = string("op_7052_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_7052_cast_fp16 = concat(axis = var_7051, interleave = var_7052_interleave_0, values = (var_7049_cast_fp16, x1_99_cast_fp16))[name = string("op_7052_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_7055_cast_fp16 = mul(x = var_7052_cast_fp16, y = var_1118_to_fp16)[name = string("op_7055_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_149_cast_fp16 = add(x = var_7028_cast_fp16, y = var_7055_cast_fp16)[name = string("k_149_cast_fp16")];
            bool var_7061_transpose_x_1 = const()[name = string("op_7061_transpose_x_1"), val = bool(false)];
            bool var_7061_transpose_y_1 = const()[name = string("op_7061_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_7061_cast_fp16 = matmul(transpose_x = var_7061_transpose_x_1, transpose_y = var_7061_transpose_y_1, x = q_149_cast_fp16, y = k_149_cast_fp16)[name = string("op_7061_cast_fp16")];
            fp16 _inversed_scores_145_y_0_to_fp16 = const()[name = string("_inversed_scores_145_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_145_cast_fp16 = mul(x = var_7061_cast_fp16, y = _inversed_scores_145_y_0_to_fp16)[name = string("_inversed_scores_145_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_147_cast_fp16 = add(x = _inversed_scores_145_cast_fp16, y = const_21_to_fp16)[name = string("scores_147_cast_fp16")];
            int32 var_7076 = const()[name = string("op_7076"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_7078_cast_fp16 = softmax(axis = var_7076, x = scores_147_cast_fp16)[name = string("op_7078_cast_fp16")];
            bool attn_out_97_transpose_x_0 = const()[name = string("attn_out_97_transpose_x_0"), val = bool(false)];
            bool attn_out_97_transpose_y_0 = const()[name = string("attn_out_97_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_99_cast_fp16 = transpose(perm = v_99_perm_0, x = v_97_cast_fp16)[name = string("transpose_33")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_97_cast_fp16 = matmul(transpose_x = attn_out_97_transpose_x_0, transpose_y = attn_out_97_transpose_y_0, x = var_7078_cast_fp16, y = v_99_cast_fp16)[name = string("attn_out_97_cast_fp16")];
            tensor<int32, [4]> var_7087_perm_0 = const()[name = string("op_7087_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7089 = const()[name = string("op_7089"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_7087_cast_fp16 = transpose(perm = var_7087_perm_0, x = attn_out_97_cast_fp16)[name = string("transpose_32")];
            tensor<fp16, [1, 291, 1024]> input_297_cast_fp16 = reshape(shape = var_7089, x = var_7087_cast_fp16)[name = string("input_297_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_24_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_24_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(813935104)))];
            tensor<fp16, [1, 291, 1024]> linear_171_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_24_self_attn_o_proj_weight_to_fp16, x = input_297_cast_fp16)[name = string("linear_171_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_391_cast_fp16 = add(x = hidden_states_383_cast_fp16, y = linear_171_cast_fp16)[name = string("hidden_states_391_cast_fp16")];
            fp16 var_7099_promoted_to_fp16 = const()[name = string("op_7099_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_7105_cast_fp16 = pow(x = hidden_states_391_cast_fp16, y = var_7099_promoted_to_fp16)[name = string("op_7105_cast_fp16")];
            tensor<int32, [1]> variance_99_axes_0 = const()[name = string("variance_99_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_99_keep_dims_0 = const()[name = string("variance_99_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_99_cast_fp16 = reduce_mean(axes = variance_99_axes_0, keep_dims = variance_99_keep_dims_0, x = var_7105_cast_fp16)[name = string("variance_99_cast_fp16")];
            fp16 var_7108_to_fp16 = const()[name = string("op_7108_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_7109_cast_fp16 = add(x = variance_99_cast_fp16, y = var_7108_to_fp16)[name = string("op_7109_cast_fp16")];
            fp32 var_7110_epsilon_0 = const()[name = string("op_7110_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_7110_cast_fp16 = rsqrt(epsilon = var_7110_epsilon_0, x = var_7109_cast_fp16)[name = string("op_7110_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_395_cast_fp16 = mul(x = hidden_states_391_cast_fp16, y = var_7110_cast_fp16)[name = string("hidden_states_395_cast_fp16")];
            tensor<fp16, [1024]> layers_24_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_24_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(816032320)))];
            tensor<fp16, [1, 291, 1024]> input_299_cast_fp16 = mul(x = layers_24_post_attention_layernorm_weight_to_fp16, y = hidden_states_395_cast_fp16)[name = string("input_299_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_24_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_24_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(816034432)))];
            tensor<fp16, [1, 291, 4096]> linear_172_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_24_mlp_gate_proj_weight_to_fp16, x = input_299_cast_fp16)[name = string("linear_172_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_7123_cast_fp16 = silu(x = linear_172_cast_fp16)[name = string("op_7123_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_24_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_24_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(824423104)))];
            tensor<fp16, [1, 291, 4096]> linear_173_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_24_mlp_up_proj_weight_to_fp16, x = input_299_cast_fp16)[name = string("linear_173_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_303_cast_fp16 = mul(x = var_7123_cast_fp16, y = linear_173_cast_fp16)[name = string("input_303_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_24_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_24_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(832811776)))];
            tensor<fp16, [1, 291, 1024]> linear_174_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_24_mlp_down_proj_weight_to_fp16, x = input_303_cast_fp16)[name = string("linear_174_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_399_cast_fp16 = add(x = hidden_states_391_cast_fp16, y = linear_174_cast_fp16)[name = string("hidden_states_399_cast_fp16")];
            tensor<int32, [4]> var_7135 = const()[name = string("op_7135"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_7147 = const()[name = string("op_7147"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_7136_cast_fp16 = transpose(perm = var_7135, x = k_149_cast_fp16)[name = string("transpose_31")];
            tensor<fp16, [1, 1024, 1, 291]> input_305_cast_fp16 = reshape(shape = var_7147, x = var_7136_cast_fp16)[name = string("input_305_cast_fp16")];
            tensor<int32, [8]> var_7153_pad_0 = const()[name = string("op_7153_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_7153_mode_0 = const()[name = string("op_7153_mode_0"), val = string("constant")];
            fp16 const_384_to_fp16 = const()[name = string("const_384_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_7153_cast_fp16 = pad(constant_val = const_384_to_fp16, mode = var_7153_mode_0, pad = var_7153_pad_0, x = input_305_cast_fp16)[name = string("op_7153_cast_fp16")];
            tensor<int32, [4]> var_7158 = const()[name = string("op_7158"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_7170 = const()[name = string("op_7170"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_7159_cast_fp16 = transpose(perm = var_7158, x = v_99_cast_fp16)[name = string("transpose_30")];
            tensor<fp16, [1, 1024, 1, 291]> input_307_cast_fp16 = reshape(shape = var_7170, x = var_7159_cast_fp16)[name = string("input_307_cast_fp16")];
            tensor<int32, [8]> var_7176_pad_0 = const()[name = string("op_7176_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_7176_mode_0 = const()[name = string("op_7176_mode_0"), val = string("constant")];
            fp16 const_387_to_fp16 = const()[name = string("const_387_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_7176_cast_fp16 = pad(constant_val = const_387_to_fp16, mode = var_7176_mode_0, pad = var_7176_pad_0, x = input_307_cast_fp16)[name = string("op_7176_cast_fp16")];
            fp16 var_7180_promoted_to_fp16 = const()[name = string("op_7180_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_7186_cast_fp16 = pow(x = hidden_states_399_cast_fp16, y = var_7180_promoted_to_fp16)[name = string("op_7186_cast_fp16")];
            tensor<int32, [1]> variance_101_axes_0 = const()[name = string("variance_101_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_101_keep_dims_0 = const()[name = string("variance_101_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_101_cast_fp16 = reduce_mean(axes = variance_101_axes_0, keep_dims = variance_101_keep_dims_0, x = var_7186_cast_fp16)[name = string("variance_101_cast_fp16")];
            fp16 var_7189_to_fp16 = const()[name = string("op_7189_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_7190_cast_fp16 = add(x = variance_101_cast_fp16, y = var_7189_to_fp16)[name = string("op_7190_cast_fp16")];
            fp32 var_7191_epsilon_0 = const()[name = string("op_7191_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_7191_cast_fp16 = rsqrt(epsilon = var_7191_epsilon_0, x = var_7190_cast_fp16)[name = string("op_7191_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_403_cast_fp16 = mul(x = hidden_states_399_cast_fp16, y = var_7191_cast_fp16)[name = string("hidden_states_403_cast_fp16")];
            tensor<fp16, [1024]> layers_25_input_layernorm_weight_to_fp16 = const()[name = string("layers_25_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841200448)))];
            tensor<fp16, [1, 291, 1024]> hidden_101_cast_fp16 = mul(x = layers_25_input_layernorm_weight_to_fp16, y = hidden_states_403_cast_fp16)[name = string("hidden_101_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_25_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(841202560)))];
            tensor<fp16, [1, 291, 1024]> linear_175_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_25_self_attn_q_proj_weight_to_fp16, x = hidden_101_cast_fp16)[name = string("linear_175_cast_fp16")];
            tensor<int32, [4]> var_7215 = const()[name = string("op_7215"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_151_cast_fp16 = reshape(shape = var_7215, x = linear_175_cast_fp16)[name = string("q_151_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_25_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(843299776)))];
            tensor<fp16, [1, 291, 1024]> linear_176_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_25_self_attn_k_proj_weight_to_fp16, x = hidden_101_cast_fp16)[name = string("linear_176_cast_fp16")];
            tensor<int32, [4]> var_7222 = const()[name = string("op_7222"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_151_cast_fp16 = reshape(shape = var_7222, x = linear_176_cast_fp16)[name = string("k_151_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_25_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(845396992)))];
            tensor<fp16, [1, 291, 1024]> linear_177_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_25_self_attn_v_proj_weight_to_fp16, x = hidden_101_cast_fp16)[name = string("linear_177_cast_fp16")];
            tensor<int32, [4]> var_7229 = const()[name = string("op_7229"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_101_cast_fp16 = reshape(shape = var_7229, x = linear_177_cast_fp16)[name = string("v_101_cast_fp16")];
            tensor<int32, [4]> q_153_perm_0 = const()[name = string("q_153_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_153_perm_0 = const()[name = string("k_153_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_103_perm_0 = const()[name = string("v_103_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_153_cast_fp16 = transpose(perm = q_153_perm_0, x = q_151_cast_fp16)[name = string("transpose_29")];
            tensor<fp16, [1, 16, 291, 64]> var_7242_cast_fp16 = mul(x = q_153_cast_fp16, y = var_1091_to_fp16)[name = string("op_7242_cast_fp16")];
            tensor<int32, [4]> x1_101_begin_0 = const()[name = string("x1_101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_101_end_0 = const()[name = string("x1_101_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_101_end_mask_0 = const()[name = string("x1_101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_101_cast_fp16 = slice_by_index(begin = x1_101_begin_0, end = x1_101_end_0, end_mask = x1_101_end_mask_0, x = q_153_cast_fp16)[name = string("x1_101_cast_fp16")];
            tensor<int32, [4]> x2_101_begin_0 = const()[name = string("x2_101_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_101_end_0 = const()[name = string("x2_101_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_101_end_mask_0 = const()[name = string("x2_101_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_101_cast_fp16 = slice_by_index(begin = x2_101_begin_0, end = x2_101_end_0, end_mask = x2_101_end_mask_0, x = q_153_cast_fp16)[name = string("x2_101_cast_fp16")];
            fp16 const_392_promoted_to_fp16 = const()[name = string("const_392_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_7263_cast_fp16 = mul(x = x2_101_cast_fp16, y = const_392_promoted_to_fp16)[name = string("op_7263_cast_fp16")];
            int32 var_7265 = const()[name = string("op_7265"), val = int32(-1)];
            bool var_7266_interleave_0 = const()[name = string("op_7266_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_7266_cast_fp16 = concat(axis = var_7265, interleave = var_7266_interleave_0, values = (var_7263_cast_fp16, x1_101_cast_fp16))[name = string("op_7266_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_7269_cast_fp16 = mul(x = var_7266_cast_fp16, y = var_1118_to_fp16)[name = string("op_7269_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_155_cast_fp16 = add(x = var_7242_cast_fp16, y = var_7269_cast_fp16)[name = string("q_155_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_153_cast_fp16 = transpose(perm = k_153_perm_0, x = k_151_cast_fp16)[name = string("transpose_28")];
            tensor<fp16, [1, 16, 291, 64]> var_7274_cast_fp16 = mul(x = k_153_cast_fp16, y = var_1091_to_fp16)[name = string("op_7274_cast_fp16")];
            tensor<int32, [4]> x1_103_begin_0 = const()[name = string("x1_103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_103_end_0 = const()[name = string("x1_103_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_103_end_mask_0 = const()[name = string("x1_103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_103_cast_fp16 = slice_by_index(begin = x1_103_begin_0, end = x1_103_end_0, end_mask = x1_103_end_mask_0, x = k_153_cast_fp16)[name = string("x1_103_cast_fp16")];
            tensor<int32, [4]> x2_103_begin_0 = const()[name = string("x2_103_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_103_end_0 = const()[name = string("x2_103_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_103_end_mask_0 = const()[name = string("x2_103_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_103_cast_fp16 = slice_by_index(begin = x2_103_begin_0, end = x2_103_end_0, end_mask = x2_103_end_mask_0, x = k_153_cast_fp16)[name = string("x2_103_cast_fp16")];
            fp16 const_395_promoted_to_fp16 = const()[name = string("const_395_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_7295_cast_fp16 = mul(x = x2_103_cast_fp16, y = const_395_promoted_to_fp16)[name = string("op_7295_cast_fp16")];
            int32 var_7297 = const()[name = string("op_7297"), val = int32(-1)];
            bool var_7298_interleave_0 = const()[name = string("op_7298_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_7298_cast_fp16 = concat(axis = var_7297, interleave = var_7298_interleave_0, values = (var_7295_cast_fp16, x1_103_cast_fp16))[name = string("op_7298_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_7301_cast_fp16 = mul(x = var_7298_cast_fp16, y = var_1118_to_fp16)[name = string("op_7301_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_155_cast_fp16 = add(x = var_7274_cast_fp16, y = var_7301_cast_fp16)[name = string("k_155_cast_fp16")];
            bool var_7307_transpose_x_1 = const()[name = string("op_7307_transpose_x_1"), val = bool(false)];
            bool var_7307_transpose_y_1 = const()[name = string("op_7307_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_7307_cast_fp16 = matmul(transpose_x = var_7307_transpose_x_1, transpose_y = var_7307_transpose_y_1, x = q_155_cast_fp16, y = k_155_cast_fp16)[name = string("op_7307_cast_fp16")];
            fp16 _inversed_scores_151_y_0_to_fp16 = const()[name = string("_inversed_scores_151_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_151_cast_fp16 = mul(x = var_7307_cast_fp16, y = _inversed_scores_151_y_0_to_fp16)[name = string("_inversed_scores_151_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_153_cast_fp16 = add(x = _inversed_scores_151_cast_fp16, y = const_21_to_fp16)[name = string("scores_153_cast_fp16")];
            int32 var_7322 = const()[name = string("op_7322"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_7324_cast_fp16 = softmax(axis = var_7322, x = scores_153_cast_fp16)[name = string("op_7324_cast_fp16")];
            bool attn_out_101_transpose_x_0 = const()[name = string("attn_out_101_transpose_x_0"), val = bool(false)];
            bool attn_out_101_transpose_y_0 = const()[name = string("attn_out_101_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_103_cast_fp16 = transpose(perm = v_103_perm_0, x = v_101_cast_fp16)[name = string("transpose_27")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_101_cast_fp16 = matmul(transpose_x = attn_out_101_transpose_x_0, transpose_y = attn_out_101_transpose_y_0, x = var_7324_cast_fp16, y = v_103_cast_fp16)[name = string("attn_out_101_cast_fp16")];
            tensor<int32, [4]> var_7333_perm_0 = const()[name = string("op_7333_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7335 = const()[name = string("op_7335"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_7333_cast_fp16 = transpose(perm = var_7333_perm_0, x = attn_out_101_cast_fp16)[name = string("transpose_26")];
            tensor<fp16, [1, 291, 1024]> input_309_cast_fp16 = reshape(shape = var_7335, x = var_7333_cast_fp16)[name = string("input_309_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_25_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_25_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(847494208)))];
            tensor<fp16, [1, 291, 1024]> linear_178_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_25_self_attn_o_proj_weight_to_fp16, x = input_309_cast_fp16)[name = string("linear_178_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_407_cast_fp16 = add(x = hidden_states_399_cast_fp16, y = linear_178_cast_fp16)[name = string("hidden_states_407_cast_fp16")];
            fp16 var_7345_promoted_to_fp16 = const()[name = string("op_7345_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_7351_cast_fp16 = pow(x = hidden_states_407_cast_fp16, y = var_7345_promoted_to_fp16)[name = string("op_7351_cast_fp16")];
            tensor<int32, [1]> variance_103_axes_0 = const()[name = string("variance_103_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_103_keep_dims_0 = const()[name = string("variance_103_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_103_cast_fp16 = reduce_mean(axes = variance_103_axes_0, keep_dims = variance_103_keep_dims_0, x = var_7351_cast_fp16)[name = string("variance_103_cast_fp16")];
            fp16 var_7354_to_fp16 = const()[name = string("op_7354_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_7355_cast_fp16 = add(x = variance_103_cast_fp16, y = var_7354_to_fp16)[name = string("op_7355_cast_fp16")];
            fp32 var_7356_epsilon_0 = const()[name = string("op_7356_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_7356_cast_fp16 = rsqrt(epsilon = var_7356_epsilon_0, x = var_7355_cast_fp16)[name = string("op_7356_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_411_cast_fp16 = mul(x = hidden_states_407_cast_fp16, y = var_7356_cast_fp16)[name = string("hidden_states_411_cast_fp16")];
            tensor<fp16, [1024]> layers_25_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_25_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(849591424)))];
            tensor<fp16, [1, 291, 1024]> input_311_cast_fp16 = mul(x = layers_25_post_attention_layernorm_weight_to_fp16, y = hidden_states_411_cast_fp16)[name = string("input_311_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_25_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_25_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(849593536)))];
            tensor<fp16, [1, 291, 4096]> linear_179_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_25_mlp_gate_proj_weight_to_fp16, x = input_311_cast_fp16)[name = string("linear_179_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_7369_cast_fp16 = silu(x = linear_179_cast_fp16)[name = string("op_7369_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_25_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_25_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(857982208)))];
            tensor<fp16, [1, 291, 4096]> linear_180_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_25_mlp_up_proj_weight_to_fp16, x = input_311_cast_fp16)[name = string("linear_180_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_315_cast_fp16 = mul(x = var_7369_cast_fp16, y = linear_180_cast_fp16)[name = string("input_315_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_25_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_25_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(866370880)))];
            tensor<fp16, [1, 291, 1024]> linear_181_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_25_mlp_down_proj_weight_to_fp16, x = input_315_cast_fp16)[name = string("linear_181_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_415_cast_fp16 = add(x = hidden_states_407_cast_fp16, y = linear_181_cast_fp16)[name = string("hidden_states_415_cast_fp16")];
            tensor<int32, [4]> var_7381 = const()[name = string("op_7381"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_7393 = const()[name = string("op_7393"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_7382_cast_fp16 = transpose(perm = var_7381, x = k_155_cast_fp16)[name = string("transpose_25")];
            tensor<fp16, [1, 1024, 1, 291]> input_317_cast_fp16 = reshape(shape = var_7393, x = var_7382_cast_fp16)[name = string("input_317_cast_fp16")];
            tensor<int32, [8]> var_7399_pad_0 = const()[name = string("op_7399_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_7399_mode_0 = const()[name = string("op_7399_mode_0"), val = string("constant")];
            fp16 const_399_to_fp16 = const()[name = string("const_399_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_7399_cast_fp16 = pad(constant_val = const_399_to_fp16, mode = var_7399_mode_0, pad = var_7399_pad_0, x = input_317_cast_fp16)[name = string("op_7399_cast_fp16")];
            tensor<int32, [4]> var_7404 = const()[name = string("op_7404"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_7416 = const()[name = string("op_7416"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_7405_cast_fp16 = transpose(perm = var_7404, x = v_103_cast_fp16)[name = string("transpose_24")];
            tensor<fp16, [1, 1024, 1, 291]> input_319_cast_fp16 = reshape(shape = var_7416, x = var_7405_cast_fp16)[name = string("input_319_cast_fp16")];
            tensor<int32, [8]> var_7422_pad_0 = const()[name = string("op_7422_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_7422_mode_0 = const()[name = string("op_7422_mode_0"), val = string("constant")];
            fp16 const_402_to_fp16 = const()[name = string("const_402_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_7422_cast_fp16 = pad(constant_val = const_402_to_fp16, mode = var_7422_mode_0, pad = var_7422_pad_0, x = input_319_cast_fp16)[name = string("op_7422_cast_fp16")];
            fp16 var_7426_promoted_to_fp16 = const()[name = string("op_7426_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_7432_cast_fp16 = pow(x = hidden_states_415_cast_fp16, y = var_7426_promoted_to_fp16)[name = string("op_7432_cast_fp16")];
            tensor<int32, [1]> variance_105_axes_0 = const()[name = string("variance_105_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_105_keep_dims_0 = const()[name = string("variance_105_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_105_cast_fp16 = reduce_mean(axes = variance_105_axes_0, keep_dims = variance_105_keep_dims_0, x = var_7432_cast_fp16)[name = string("variance_105_cast_fp16")];
            fp16 var_7435_to_fp16 = const()[name = string("op_7435_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_7436_cast_fp16 = add(x = variance_105_cast_fp16, y = var_7435_to_fp16)[name = string("op_7436_cast_fp16")];
            fp32 var_7437_epsilon_0 = const()[name = string("op_7437_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_7437_cast_fp16 = rsqrt(epsilon = var_7437_epsilon_0, x = var_7436_cast_fp16)[name = string("op_7437_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_419_cast_fp16 = mul(x = hidden_states_415_cast_fp16, y = var_7437_cast_fp16)[name = string("hidden_states_419_cast_fp16")];
            tensor<fp16, [1024]> layers_26_input_layernorm_weight_to_fp16 = const()[name = string("layers_26_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(874759552)))];
            tensor<fp16, [1, 291, 1024]> hidden_105_cast_fp16 = mul(x = layers_26_input_layernorm_weight_to_fp16, y = hidden_states_419_cast_fp16)[name = string("hidden_105_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_26_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(874761664)))];
            tensor<fp16, [1, 291, 1024]> linear_182_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_26_self_attn_q_proj_weight_to_fp16, x = hidden_105_cast_fp16)[name = string("linear_182_cast_fp16")];
            tensor<int32, [4]> var_7461 = const()[name = string("op_7461"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_157_cast_fp16 = reshape(shape = var_7461, x = linear_182_cast_fp16)[name = string("q_157_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_26_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(876858880)))];
            tensor<fp16, [1, 291, 1024]> linear_183_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_26_self_attn_k_proj_weight_to_fp16, x = hidden_105_cast_fp16)[name = string("linear_183_cast_fp16")];
            tensor<int32, [4]> var_7468 = const()[name = string("op_7468"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_157_cast_fp16 = reshape(shape = var_7468, x = linear_183_cast_fp16)[name = string("k_157_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_26_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(878956096)))];
            tensor<fp16, [1, 291, 1024]> linear_184_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_26_self_attn_v_proj_weight_to_fp16, x = hidden_105_cast_fp16)[name = string("linear_184_cast_fp16")];
            tensor<int32, [4]> var_7475 = const()[name = string("op_7475"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_105_cast_fp16 = reshape(shape = var_7475, x = linear_184_cast_fp16)[name = string("v_105_cast_fp16")];
            tensor<int32, [4]> q_159_perm_0 = const()[name = string("q_159_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_159_perm_0 = const()[name = string("k_159_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_107_perm_0 = const()[name = string("v_107_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_159_cast_fp16 = transpose(perm = q_159_perm_0, x = q_157_cast_fp16)[name = string("transpose_23")];
            tensor<fp16, [1, 16, 291, 64]> var_7488_cast_fp16 = mul(x = q_159_cast_fp16, y = var_1091_to_fp16)[name = string("op_7488_cast_fp16")];
            tensor<int32, [4]> x1_105_begin_0 = const()[name = string("x1_105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_105_end_0 = const()[name = string("x1_105_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_105_end_mask_0 = const()[name = string("x1_105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_105_cast_fp16 = slice_by_index(begin = x1_105_begin_0, end = x1_105_end_0, end_mask = x1_105_end_mask_0, x = q_159_cast_fp16)[name = string("x1_105_cast_fp16")];
            tensor<int32, [4]> x2_105_begin_0 = const()[name = string("x2_105_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_105_end_0 = const()[name = string("x2_105_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_105_end_mask_0 = const()[name = string("x2_105_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_105_cast_fp16 = slice_by_index(begin = x2_105_begin_0, end = x2_105_end_0, end_mask = x2_105_end_mask_0, x = q_159_cast_fp16)[name = string("x2_105_cast_fp16")];
            fp16 const_407_promoted_to_fp16 = const()[name = string("const_407_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_7509_cast_fp16 = mul(x = x2_105_cast_fp16, y = const_407_promoted_to_fp16)[name = string("op_7509_cast_fp16")];
            int32 var_7511 = const()[name = string("op_7511"), val = int32(-1)];
            bool var_7512_interleave_0 = const()[name = string("op_7512_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_7512_cast_fp16 = concat(axis = var_7511, interleave = var_7512_interleave_0, values = (var_7509_cast_fp16, x1_105_cast_fp16))[name = string("op_7512_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_7515_cast_fp16 = mul(x = var_7512_cast_fp16, y = var_1118_to_fp16)[name = string("op_7515_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_161_cast_fp16 = add(x = var_7488_cast_fp16, y = var_7515_cast_fp16)[name = string("q_161_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_159_cast_fp16 = transpose(perm = k_159_perm_0, x = k_157_cast_fp16)[name = string("transpose_22")];
            tensor<fp16, [1, 16, 291, 64]> var_7520_cast_fp16 = mul(x = k_159_cast_fp16, y = var_1091_to_fp16)[name = string("op_7520_cast_fp16")];
            tensor<int32, [4]> x1_107_begin_0 = const()[name = string("x1_107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_107_end_0 = const()[name = string("x1_107_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_107_end_mask_0 = const()[name = string("x1_107_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_107_cast_fp16 = slice_by_index(begin = x1_107_begin_0, end = x1_107_end_0, end_mask = x1_107_end_mask_0, x = k_159_cast_fp16)[name = string("x1_107_cast_fp16")];
            tensor<int32, [4]> x2_107_begin_0 = const()[name = string("x2_107_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_107_end_0 = const()[name = string("x2_107_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_107_end_mask_0 = const()[name = string("x2_107_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_107_cast_fp16 = slice_by_index(begin = x2_107_begin_0, end = x2_107_end_0, end_mask = x2_107_end_mask_0, x = k_159_cast_fp16)[name = string("x2_107_cast_fp16")];
            fp16 const_410_promoted_to_fp16 = const()[name = string("const_410_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_7541_cast_fp16 = mul(x = x2_107_cast_fp16, y = const_410_promoted_to_fp16)[name = string("op_7541_cast_fp16")];
            int32 var_7543 = const()[name = string("op_7543"), val = int32(-1)];
            bool var_7544_interleave_0 = const()[name = string("op_7544_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_7544_cast_fp16 = concat(axis = var_7543, interleave = var_7544_interleave_0, values = (var_7541_cast_fp16, x1_107_cast_fp16))[name = string("op_7544_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_7547_cast_fp16 = mul(x = var_7544_cast_fp16, y = var_1118_to_fp16)[name = string("op_7547_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_161_cast_fp16 = add(x = var_7520_cast_fp16, y = var_7547_cast_fp16)[name = string("k_161_cast_fp16")];
            bool var_7553_transpose_x_1 = const()[name = string("op_7553_transpose_x_1"), val = bool(false)];
            bool var_7553_transpose_y_1 = const()[name = string("op_7553_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_7553_cast_fp16 = matmul(transpose_x = var_7553_transpose_x_1, transpose_y = var_7553_transpose_y_1, x = q_161_cast_fp16, y = k_161_cast_fp16)[name = string("op_7553_cast_fp16")];
            fp16 _inversed_scores_157_y_0_to_fp16 = const()[name = string("_inversed_scores_157_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_157_cast_fp16 = mul(x = var_7553_cast_fp16, y = _inversed_scores_157_y_0_to_fp16)[name = string("_inversed_scores_157_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_159_cast_fp16 = add(x = _inversed_scores_157_cast_fp16, y = const_21_to_fp16)[name = string("scores_159_cast_fp16")];
            int32 var_7568 = const()[name = string("op_7568"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_7570_cast_fp16 = softmax(axis = var_7568, x = scores_159_cast_fp16)[name = string("op_7570_cast_fp16")];
            bool attn_out_105_transpose_x_0 = const()[name = string("attn_out_105_transpose_x_0"), val = bool(false)];
            bool attn_out_105_transpose_y_0 = const()[name = string("attn_out_105_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_107_cast_fp16 = transpose(perm = v_107_perm_0, x = v_105_cast_fp16)[name = string("transpose_21")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_105_cast_fp16 = matmul(transpose_x = attn_out_105_transpose_x_0, transpose_y = attn_out_105_transpose_y_0, x = var_7570_cast_fp16, y = v_107_cast_fp16)[name = string("attn_out_105_cast_fp16")];
            tensor<int32, [4]> var_7579_perm_0 = const()[name = string("op_7579_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7581 = const()[name = string("op_7581"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_7579_cast_fp16 = transpose(perm = var_7579_perm_0, x = attn_out_105_cast_fp16)[name = string("transpose_20")];
            tensor<fp16, [1, 291, 1024]> input_321_cast_fp16 = reshape(shape = var_7581, x = var_7579_cast_fp16)[name = string("input_321_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_26_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_26_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(881053312)))];
            tensor<fp16, [1, 291, 1024]> linear_185_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_26_self_attn_o_proj_weight_to_fp16, x = input_321_cast_fp16)[name = string("linear_185_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_423_cast_fp16 = add(x = hidden_states_415_cast_fp16, y = linear_185_cast_fp16)[name = string("hidden_states_423_cast_fp16")];
            fp16 var_7591_promoted_to_fp16 = const()[name = string("op_7591_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_7597_cast_fp16 = pow(x = hidden_states_423_cast_fp16, y = var_7591_promoted_to_fp16)[name = string("op_7597_cast_fp16")];
            tensor<int32, [1]> variance_107_axes_0 = const()[name = string("variance_107_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_107_keep_dims_0 = const()[name = string("variance_107_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_107_cast_fp16 = reduce_mean(axes = variance_107_axes_0, keep_dims = variance_107_keep_dims_0, x = var_7597_cast_fp16)[name = string("variance_107_cast_fp16")];
            fp16 var_7600_to_fp16 = const()[name = string("op_7600_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_7601_cast_fp16 = add(x = variance_107_cast_fp16, y = var_7600_to_fp16)[name = string("op_7601_cast_fp16")];
            fp32 var_7602_epsilon_0 = const()[name = string("op_7602_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_7602_cast_fp16 = rsqrt(epsilon = var_7602_epsilon_0, x = var_7601_cast_fp16)[name = string("op_7602_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_427_cast_fp16 = mul(x = hidden_states_423_cast_fp16, y = var_7602_cast_fp16)[name = string("hidden_states_427_cast_fp16")];
            tensor<fp16, [1024]> layers_26_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_26_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883150528)))];
            tensor<fp16, [1, 291, 1024]> input_323_cast_fp16 = mul(x = layers_26_post_attention_layernorm_weight_to_fp16, y = hidden_states_427_cast_fp16)[name = string("input_323_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_26_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_26_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(883152640)))];
            tensor<fp16, [1, 291, 4096]> linear_186_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_26_mlp_gate_proj_weight_to_fp16, x = input_323_cast_fp16)[name = string("linear_186_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_7615_cast_fp16 = silu(x = linear_186_cast_fp16)[name = string("op_7615_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_26_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_26_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(891541312)))];
            tensor<fp16, [1, 291, 4096]> linear_187_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_26_mlp_up_proj_weight_to_fp16, x = input_323_cast_fp16)[name = string("linear_187_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_327_cast_fp16 = mul(x = var_7615_cast_fp16, y = linear_187_cast_fp16)[name = string("input_327_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_26_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_26_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(899929984)))];
            tensor<fp16, [1, 291, 1024]> linear_188_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_26_mlp_down_proj_weight_to_fp16, x = input_327_cast_fp16)[name = string("linear_188_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_431_cast_fp16 = add(x = hidden_states_423_cast_fp16, y = linear_188_cast_fp16)[name = string("hidden_states_431_cast_fp16")];
            tensor<int32, [4]> var_7627 = const()[name = string("op_7627"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_7639 = const()[name = string("op_7639"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_7628_cast_fp16 = transpose(perm = var_7627, x = k_161_cast_fp16)[name = string("transpose_19")];
            tensor<fp16, [1, 1024, 1, 291]> input_329_cast_fp16 = reshape(shape = var_7639, x = var_7628_cast_fp16)[name = string("input_329_cast_fp16")];
            tensor<int32, [8]> var_7645_pad_0 = const()[name = string("op_7645_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_7645_mode_0 = const()[name = string("op_7645_mode_0"), val = string("constant")];
            fp16 const_414_to_fp16 = const()[name = string("const_414_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_7645_cast_fp16 = pad(constant_val = const_414_to_fp16, mode = var_7645_mode_0, pad = var_7645_pad_0, x = input_329_cast_fp16)[name = string("op_7645_cast_fp16")];
            tensor<int32, [4]> var_7650 = const()[name = string("op_7650"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_7662 = const()[name = string("op_7662"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_7651_cast_fp16 = transpose(perm = var_7650, x = v_107_cast_fp16)[name = string("transpose_18")];
            tensor<fp16, [1, 1024, 1, 291]> input_331_cast_fp16 = reshape(shape = var_7662, x = var_7651_cast_fp16)[name = string("input_331_cast_fp16")];
            tensor<int32, [8]> var_7668_pad_0 = const()[name = string("op_7668_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_7668_mode_0 = const()[name = string("op_7668_mode_0"), val = string("constant")];
            fp16 const_417_to_fp16 = const()[name = string("const_417_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_7668_cast_fp16 = pad(constant_val = const_417_to_fp16, mode = var_7668_mode_0, pad = var_7668_pad_0, x = input_331_cast_fp16)[name = string("op_7668_cast_fp16")];
            fp16 var_7672_promoted_to_fp16 = const()[name = string("op_7672_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_7678_cast_fp16 = pow(x = hidden_states_431_cast_fp16, y = var_7672_promoted_to_fp16)[name = string("op_7678_cast_fp16")];
            tensor<int32, [1]> variance_109_axes_0 = const()[name = string("variance_109_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_109_keep_dims_0 = const()[name = string("variance_109_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_109_cast_fp16 = reduce_mean(axes = variance_109_axes_0, keep_dims = variance_109_keep_dims_0, x = var_7678_cast_fp16)[name = string("variance_109_cast_fp16")];
            fp16 var_7681_to_fp16 = const()[name = string("op_7681_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_7682_cast_fp16 = add(x = variance_109_cast_fp16, y = var_7681_to_fp16)[name = string("op_7682_cast_fp16")];
            fp32 var_7683_epsilon_0 = const()[name = string("op_7683_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_7683_cast_fp16 = rsqrt(epsilon = var_7683_epsilon_0, x = var_7682_cast_fp16)[name = string("op_7683_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_435_cast_fp16 = mul(x = hidden_states_431_cast_fp16, y = var_7683_cast_fp16)[name = string("hidden_states_435_cast_fp16")];
            tensor<fp16, [1024]> layers_27_input_layernorm_weight_to_fp16 = const()[name = string("layers_27_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(908318656)))];
            tensor<fp16, [1, 291, 1024]> hidden_109_cast_fp16 = mul(x = layers_27_input_layernorm_weight_to_fp16, y = hidden_states_435_cast_fp16)[name = string("hidden_109_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_27_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(908320768)))];
            tensor<fp16, [1, 291, 1024]> linear_189_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_27_self_attn_q_proj_weight_to_fp16, x = hidden_109_cast_fp16)[name = string("linear_189_cast_fp16")];
            tensor<int32, [4]> var_7707 = const()[name = string("op_7707"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_163_cast_fp16 = reshape(shape = var_7707, x = linear_189_cast_fp16)[name = string("q_163_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_27_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(910417984)))];
            tensor<fp16, [1, 291, 1024]> linear_190_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_27_self_attn_k_proj_weight_to_fp16, x = hidden_109_cast_fp16)[name = string("linear_190_cast_fp16")];
            tensor<int32, [4]> var_7714 = const()[name = string("op_7714"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_163_cast_fp16 = reshape(shape = var_7714, x = linear_190_cast_fp16)[name = string("k_163_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_27_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(912515200)))];
            tensor<fp16, [1, 291, 1024]> linear_191_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_27_self_attn_v_proj_weight_to_fp16, x = hidden_109_cast_fp16)[name = string("linear_191_cast_fp16")];
            tensor<int32, [4]> var_7721 = const()[name = string("op_7721"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_109_cast_fp16 = reshape(shape = var_7721, x = linear_191_cast_fp16)[name = string("v_109_cast_fp16")];
            tensor<int32, [4]> q_165_perm_0 = const()[name = string("q_165_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_165_perm_0 = const()[name = string("k_165_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_111_perm_0 = const()[name = string("v_111_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_165_cast_fp16 = transpose(perm = q_165_perm_0, x = q_163_cast_fp16)[name = string("transpose_17")];
            tensor<fp16, [1, 16, 291, 64]> var_7734_cast_fp16 = mul(x = q_165_cast_fp16, y = var_1091_to_fp16)[name = string("op_7734_cast_fp16")];
            tensor<int32, [4]> x1_109_begin_0 = const()[name = string("x1_109_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_109_end_0 = const()[name = string("x1_109_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_109_end_mask_0 = const()[name = string("x1_109_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_109_cast_fp16 = slice_by_index(begin = x1_109_begin_0, end = x1_109_end_0, end_mask = x1_109_end_mask_0, x = q_165_cast_fp16)[name = string("x1_109_cast_fp16")];
            tensor<int32, [4]> x2_109_begin_0 = const()[name = string("x2_109_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_109_end_0 = const()[name = string("x2_109_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_109_end_mask_0 = const()[name = string("x2_109_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_109_cast_fp16 = slice_by_index(begin = x2_109_begin_0, end = x2_109_end_0, end_mask = x2_109_end_mask_0, x = q_165_cast_fp16)[name = string("x2_109_cast_fp16")];
            fp16 const_422_promoted_to_fp16 = const()[name = string("const_422_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_7755_cast_fp16 = mul(x = x2_109_cast_fp16, y = const_422_promoted_to_fp16)[name = string("op_7755_cast_fp16")];
            int32 var_7757 = const()[name = string("op_7757"), val = int32(-1)];
            bool var_7758_interleave_0 = const()[name = string("op_7758_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_7758_cast_fp16 = concat(axis = var_7757, interleave = var_7758_interleave_0, values = (var_7755_cast_fp16, x1_109_cast_fp16))[name = string("op_7758_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_7761_cast_fp16 = mul(x = var_7758_cast_fp16, y = var_1118_to_fp16)[name = string("op_7761_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_167_cast_fp16 = add(x = var_7734_cast_fp16, y = var_7761_cast_fp16)[name = string("q_167_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_165_cast_fp16 = transpose(perm = k_165_perm_0, x = k_163_cast_fp16)[name = string("transpose_16")];
            tensor<fp16, [1, 16, 291, 64]> var_7766_cast_fp16 = mul(x = k_165_cast_fp16, y = var_1091_to_fp16)[name = string("op_7766_cast_fp16")];
            tensor<int32, [4]> x1_111_begin_0 = const()[name = string("x1_111_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_111_end_0 = const()[name = string("x1_111_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_111_end_mask_0 = const()[name = string("x1_111_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_111_cast_fp16 = slice_by_index(begin = x1_111_begin_0, end = x1_111_end_0, end_mask = x1_111_end_mask_0, x = k_165_cast_fp16)[name = string("x1_111_cast_fp16")];
            tensor<int32, [4]> x2_111_begin_0 = const()[name = string("x2_111_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_111_end_0 = const()[name = string("x2_111_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_111_end_mask_0 = const()[name = string("x2_111_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_111_cast_fp16 = slice_by_index(begin = x2_111_begin_0, end = x2_111_end_0, end_mask = x2_111_end_mask_0, x = k_165_cast_fp16)[name = string("x2_111_cast_fp16")];
            fp16 const_425_promoted_to_fp16 = const()[name = string("const_425_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_7787_cast_fp16 = mul(x = x2_111_cast_fp16, y = const_425_promoted_to_fp16)[name = string("op_7787_cast_fp16")];
            int32 var_7789 = const()[name = string("op_7789"), val = int32(-1)];
            bool var_7790_interleave_0 = const()[name = string("op_7790_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_7790_cast_fp16 = concat(axis = var_7789, interleave = var_7790_interleave_0, values = (var_7787_cast_fp16, x1_111_cast_fp16))[name = string("op_7790_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_7793_cast_fp16 = mul(x = var_7790_cast_fp16, y = var_1118_to_fp16)[name = string("op_7793_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_167_cast_fp16 = add(x = var_7766_cast_fp16, y = var_7793_cast_fp16)[name = string("k_167_cast_fp16")];
            bool var_7799_transpose_x_1 = const()[name = string("op_7799_transpose_x_1"), val = bool(false)];
            bool var_7799_transpose_y_1 = const()[name = string("op_7799_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_7799_cast_fp16 = matmul(transpose_x = var_7799_transpose_x_1, transpose_y = var_7799_transpose_y_1, x = q_167_cast_fp16, y = k_167_cast_fp16)[name = string("op_7799_cast_fp16")];
            fp16 _inversed_scores_163_y_0_to_fp16 = const()[name = string("_inversed_scores_163_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_163_cast_fp16 = mul(x = var_7799_cast_fp16, y = _inversed_scores_163_y_0_to_fp16)[name = string("_inversed_scores_163_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_165_cast_fp16 = add(x = _inversed_scores_163_cast_fp16, y = const_21_to_fp16)[name = string("scores_165_cast_fp16")];
            int32 var_7814 = const()[name = string("op_7814"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_7816_cast_fp16 = softmax(axis = var_7814, x = scores_165_cast_fp16)[name = string("op_7816_cast_fp16")];
            bool attn_out_109_transpose_x_0 = const()[name = string("attn_out_109_transpose_x_0"), val = bool(false)];
            bool attn_out_109_transpose_y_0 = const()[name = string("attn_out_109_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_111_cast_fp16 = transpose(perm = v_111_perm_0, x = v_109_cast_fp16)[name = string("transpose_15")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_109_cast_fp16 = matmul(transpose_x = attn_out_109_transpose_x_0, transpose_y = attn_out_109_transpose_y_0, x = var_7816_cast_fp16, y = v_111_cast_fp16)[name = string("attn_out_109_cast_fp16")];
            tensor<int32, [4]> var_7825_perm_0 = const()[name = string("op_7825_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_7827 = const()[name = string("op_7827"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_7825_cast_fp16 = transpose(perm = var_7825_perm_0, x = attn_out_109_cast_fp16)[name = string("transpose_14")];
            tensor<fp16, [1, 291, 1024]> input_333_cast_fp16 = reshape(shape = var_7827, x = var_7825_cast_fp16)[name = string("input_333_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_27_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_27_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(914612416)))];
            tensor<fp16, [1, 291, 1024]> linear_192_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_27_self_attn_o_proj_weight_to_fp16, x = input_333_cast_fp16)[name = string("linear_192_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_439_cast_fp16 = add(x = hidden_states_431_cast_fp16, y = linear_192_cast_fp16)[name = string("hidden_states_439_cast_fp16")];
            fp16 var_7837_promoted_to_fp16 = const()[name = string("op_7837_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_7843_cast_fp16 = pow(x = hidden_states_439_cast_fp16, y = var_7837_promoted_to_fp16)[name = string("op_7843_cast_fp16")];
            tensor<int32, [1]> variance_111_axes_0 = const()[name = string("variance_111_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_111_keep_dims_0 = const()[name = string("variance_111_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_111_cast_fp16 = reduce_mean(axes = variance_111_axes_0, keep_dims = variance_111_keep_dims_0, x = var_7843_cast_fp16)[name = string("variance_111_cast_fp16")];
            fp16 var_7846_to_fp16 = const()[name = string("op_7846_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_7847_cast_fp16 = add(x = variance_111_cast_fp16, y = var_7846_to_fp16)[name = string("op_7847_cast_fp16")];
            fp32 var_7848_epsilon_0 = const()[name = string("op_7848_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_7848_cast_fp16 = rsqrt(epsilon = var_7848_epsilon_0, x = var_7847_cast_fp16)[name = string("op_7848_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_443_cast_fp16 = mul(x = hidden_states_439_cast_fp16, y = var_7848_cast_fp16)[name = string("hidden_states_443_cast_fp16")];
            tensor<fp16, [1024]> layers_27_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_27_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(916709632)))];
            tensor<fp16, [1, 291, 1024]> input_335_cast_fp16 = mul(x = layers_27_post_attention_layernorm_weight_to_fp16, y = hidden_states_443_cast_fp16)[name = string("input_335_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_27_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_27_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(916711744)))];
            tensor<fp16, [1, 291, 4096]> linear_193_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_27_mlp_gate_proj_weight_to_fp16, x = input_335_cast_fp16)[name = string("linear_193_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_7861_cast_fp16 = silu(x = linear_193_cast_fp16)[name = string("op_7861_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_27_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_27_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(925100416)))];
            tensor<fp16, [1, 291, 4096]> linear_194_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_27_mlp_up_proj_weight_to_fp16, x = input_335_cast_fp16)[name = string("linear_194_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_339_cast_fp16 = mul(x = var_7861_cast_fp16, y = linear_194_cast_fp16)[name = string("input_339_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_27_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_27_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(933489088)))];
            tensor<fp16, [1, 291, 1024]> linear_195_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_27_mlp_down_proj_weight_to_fp16, x = input_339_cast_fp16)[name = string("linear_195_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_447_cast_fp16 = add(x = hidden_states_439_cast_fp16, y = linear_195_cast_fp16)[name = string("hidden_states_447_cast_fp16")];
            tensor<int32, [4]> var_7873 = const()[name = string("op_7873"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_7885 = const()[name = string("op_7885"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_7874_cast_fp16 = transpose(perm = var_7873, x = k_167_cast_fp16)[name = string("transpose_13")];
            tensor<fp16, [1, 1024, 1, 291]> input_341_cast_fp16 = reshape(shape = var_7885, x = var_7874_cast_fp16)[name = string("input_341_cast_fp16")];
            tensor<int32, [8]> var_7891_pad_0 = const()[name = string("op_7891_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_7891_mode_0 = const()[name = string("op_7891_mode_0"), val = string("constant")];
            fp16 const_429_to_fp16 = const()[name = string("const_429_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_7891_cast_fp16 = pad(constant_val = const_429_to_fp16, mode = var_7891_mode_0, pad = var_7891_pad_0, x = input_341_cast_fp16)[name = string("op_7891_cast_fp16")];
            tensor<int32, [4]> var_7896 = const()[name = string("op_7896"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_7908 = const()[name = string("op_7908"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_7897_cast_fp16 = transpose(perm = var_7896, x = v_111_cast_fp16)[name = string("transpose_12")];
            tensor<fp16, [1, 1024, 1, 291]> input_343_cast_fp16 = reshape(shape = var_7908, x = var_7897_cast_fp16)[name = string("input_343_cast_fp16")];
            tensor<int32, [8]> var_7914_pad_0 = const()[name = string("op_7914_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_7914_mode_0 = const()[name = string("op_7914_mode_0"), val = string("constant")];
            fp16 const_432_to_fp16 = const()[name = string("const_432_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_7914_cast_fp16 = pad(constant_val = const_432_to_fp16, mode = var_7914_mode_0, pad = var_7914_pad_0, x = input_343_cast_fp16)[name = string("op_7914_cast_fp16")];
            fp16 var_7918_promoted_to_fp16 = const()[name = string("op_7918_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_7924_cast_fp16 = pow(x = hidden_states_447_cast_fp16, y = var_7918_promoted_to_fp16)[name = string("op_7924_cast_fp16")];
            tensor<int32, [1]> variance_113_axes_0 = const()[name = string("variance_113_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_113_keep_dims_0 = const()[name = string("variance_113_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_113_cast_fp16 = reduce_mean(axes = variance_113_axes_0, keep_dims = variance_113_keep_dims_0, x = var_7924_cast_fp16)[name = string("variance_113_cast_fp16")];
            fp16 var_7927_to_fp16 = const()[name = string("op_7927_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_7928_cast_fp16 = add(x = variance_113_cast_fp16, y = var_7927_to_fp16)[name = string("op_7928_cast_fp16")];
            fp32 var_7929_epsilon_0 = const()[name = string("op_7929_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_7929_cast_fp16 = rsqrt(epsilon = var_7929_epsilon_0, x = var_7928_cast_fp16)[name = string("op_7929_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_451_cast_fp16 = mul(x = hidden_states_447_cast_fp16, y = var_7929_cast_fp16)[name = string("hidden_states_451_cast_fp16")];
            tensor<fp16, [1024]> layers_28_input_layernorm_weight_to_fp16 = const()[name = string("layers_28_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(941877760)))];
            tensor<fp16, [1, 291, 1024]> hidden_113_cast_fp16 = mul(x = layers_28_input_layernorm_weight_to_fp16, y = hidden_states_451_cast_fp16)[name = string("hidden_113_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_28_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(941879872)))];
            tensor<fp16, [1, 291, 1024]> linear_196_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_28_self_attn_q_proj_weight_to_fp16, x = hidden_113_cast_fp16)[name = string("linear_196_cast_fp16")];
            tensor<int32, [4]> var_7953 = const()[name = string("op_7953"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_169_cast_fp16 = reshape(shape = var_7953, x = linear_196_cast_fp16)[name = string("q_169_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_28_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(943977088)))];
            tensor<fp16, [1, 291, 1024]> linear_197_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_28_self_attn_k_proj_weight_to_fp16, x = hidden_113_cast_fp16)[name = string("linear_197_cast_fp16")];
            tensor<int32, [4]> var_7960 = const()[name = string("op_7960"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_169_cast_fp16 = reshape(shape = var_7960, x = linear_197_cast_fp16)[name = string("k_169_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_28_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(946074304)))];
            tensor<fp16, [1, 291, 1024]> linear_198_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_28_self_attn_v_proj_weight_to_fp16, x = hidden_113_cast_fp16)[name = string("linear_198_cast_fp16")];
            tensor<int32, [4]> var_7967 = const()[name = string("op_7967"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_113_cast_fp16 = reshape(shape = var_7967, x = linear_198_cast_fp16)[name = string("v_113_cast_fp16")];
            tensor<int32, [4]> q_171_perm_0 = const()[name = string("q_171_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_171_perm_0 = const()[name = string("k_171_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_115_perm_0 = const()[name = string("v_115_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_171_cast_fp16 = transpose(perm = q_171_perm_0, x = q_169_cast_fp16)[name = string("transpose_11")];
            tensor<fp16, [1, 16, 291, 64]> var_7980_cast_fp16 = mul(x = q_171_cast_fp16, y = var_1091_to_fp16)[name = string("op_7980_cast_fp16")];
            tensor<int32, [4]> x1_113_begin_0 = const()[name = string("x1_113_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_113_end_0 = const()[name = string("x1_113_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_113_end_mask_0 = const()[name = string("x1_113_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_113_cast_fp16 = slice_by_index(begin = x1_113_begin_0, end = x1_113_end_0, end_mask = x1_113_end_mask_0, x = q_171_cast_fp16)[name = string("x1_113_cast_fp16")];
            tensor<int32, [4]> x2_113_begin_0 = const()[name = string("x2_113_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_113_end_0 = const()[name = string("x2_113_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_113_end_mask_0 = const()[name = string("x2_113_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_113_cast_fp16 = slice_by_index(begin = x2_113_begin_0, end = x2_113_end_0, end_mask = x2_113_end_mask_0, x = q_171_cast_fp16)[name = string("x2_113_cast_fp16")];
            fp16 const_437_promoted_to_fp16 = const()[name = string("const_437_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_8001_cast_fp16 = mul(x = x2_113_cast_fp16, y = const_437_promoted_to_fp16)[name = string("op_8001_cast_fp16")];
            int32 var_8003 = const()[name = string("op_8003"), val = int32(-1)];
            bool var_8004_interleave_0 = const()[name = string("op_8004_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_8004_cast_fp16 = concat(axis = var_8003, interleave = var_8004_interleave_0, values = (var_8001_cast_fp16, x1_113_cast_fp16))[name = string("op_8004_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_8007_cast_fp16 = mul(x = var_8004_cast_fp16, y = var_1118_to_fp16)[name = string("op_8007_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_173_cast_fp16 = add(x = var_7980_cast_fp16, y = var_8007_cast_fp16)[name = string("q_173_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_171_cast_fp16 = transpose(perm = k_171_perm_0, x = k_169_cast_fp16)[name = string("transpose_10")];
            tensor<fp16, [1, 16, 291, 64]> var_8012_cast_fp16 = mul(x = k_171_cast_fp16, y = var_1091_to_fp16)[name = string("op_8012_cast_fp16")];
            tensor<int32, [4]> x1_115_begin_0 = const()[name = string("x1_115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_115_end_0 = const()[name = string("x1_115_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_115_end_mask_0 = const()[name = string("x1_115_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_115_cast_fp16 = slice_by_index(begin = x1_115_begin_0, end = x1_115_end_0, end_mask = x1_115_end_mask_0, x = k_171_cast_fp16)[name = string("x1_115_cast_fp16")];
            tensor<int32, [4]> x2_115_begin_0 = const()[name = string("x2_115_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_115_end_0 = const()[name = string("x2_115_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_115_end_mask_0 = const()[name = string("x2_115_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_115_cast_fp16 = slice_by_index(begin = x2_115_begin_0, end = x2_115_end_0, end_mask = x2_115_end_mask_0, x = k_171_cast_fp16)[name = string("x2_115_cast_fp16")];
            fp16 const_440_promoted_to_fp16 = const()[name = string("const_440_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_8033_cast_fp16 = mul(x = x2_115_cast_fp16, y = const_440_promoted_to_fp16)[name = string("op_8033_cast_fp16")];
            int32 var_8035 = const()[name = string("op_8035"), val = int32(-1)];
            bool var_8036_interleave_0 = const()[name = string("op_8036_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_8036_cast_fp16 = concat(axis = var_8035, interleave = var_8036_interleave_0, values = (var_8033_cast_fp16, x1_115_cast_fp16))[name = string("op_8036_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_8039_cast_fp16 = mul(x = var_8036_cast_fp16, y = var_1118_to_fp16)[name = string("op_8039_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_173_cast_fp16 = add(x = var_8012_cast_fp16, y = var_8039_cast_fp16)[name = string("k_173_cast_fp16")];
            bool var_8045_transpose_x_1 = const()[name = string("op_8045_transpose_x_1"), val = bool(false)];
            bool var_8045_transpose_y_1 = const()[name = string("op_8045_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_8045_cast_fp16 = matmul(transpose_x = var_8045_transpose_x_1, transpose_y = var_8045_transpose_y_1, x = q_173_cast_fp16, y = k_173_cast_fp16)[name = string("op_8045_cast_fp16")];
            fp16 _inversed_scores_169_y_0_to_fp16 = const()[name = string("_inversed_scores_169_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_169_cast_fp16 = mul(x = var_8045_cast_fp16, y = _inversed_scores_169_y_0_to_fp16)[name = string("_inversed_scores_169_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_171_cast_fp16 = add(x = _inversed_scores_169_cast_fp16, y = const_21_to_fp16)[name = string("scores_171_cast_fp16")];
            int32 var_8060 = const()[name = string("op_8060"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_8062_cast_fp16 = softmax(axis = var_8060, x = scores_171_cast_fp16)[name = string("op_8062_cast_fp16")];
            bool attn_out_113_transpose_x_0 = const()[name = string("attn_out_113_transpose_x_0"), val = bool(false)];
            bool attn_out_113_transpose_y_0 = const()[name = string("attn_out_113_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_115_cast_fp16 = transpose(perm = v_115_perm_0, x = v_113_cast_fp16)[name = string("transpose_9")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_113_cast_fp16 = matmul(transpose_x = attn_out_113_transpose_x_0, transpose_y = attn_out_113_transpose_y_0, x = var_8062_cast_fp16, y = v_115_cast_fp16)[name = string("attn_out_113_cast_fp16")];
            tensor<int32, [4]> var_8071_perm_0 = const()[name = string("op_8071_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_8073 = const()[name = string("op_8073"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_8071_cast_fp16 = transpose(perm = var_8071_perm_0, x = attn_out_113_cast_fp16)[name = string("transpose_8")];
            tensor<fp16, [1, 291, 1024]> input_345_cast_fp16 = reshape(shape = var_8073, x = var_8071_cast_fp16)[name = string("input_345_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_28_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_28_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(948171520)))];
            tensor<fp16, [1, 291, 1024]> linear_199_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_28_self_attn_o_proj_weight_to_fp16, x = input_345_cast_fp16)[name = string("linear_199_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_455_cast_fp16 = add(x = hidden_states_447_cast_fp16, y = linear_199_cast_fp16)[name = string("hidden_states_455_cast_fp16")];
            fp16 var_8083_promoted_to_fp16 = const()[name = string("op_8083_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_8089_cast_fp16 = pow(x = hidden_states_455_cast_fp16, y = var_8083_promoted_to_fp16)[name = string("op_8089_cast_fp16")];
            tensor<int32, [1]> variance_115_axes_0 = const()[name = string("variance_115_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_115_keep_dims_0 = const()[name = string("variance_115_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_115_cast_fp16 = reduce_mean(axes = variance_115_axes_0, keep_dims = variance_115_keep_dims_0, x = var_8089_cast_fp16)[name = string("variance_115_cast_fp16")];
            fp16 var_8092_to_fp16 = const()[name = string("op_8092_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_8093_cast_fp16 = add(x = variance_115_cast_fp16, y = var_8092_to_fp16)[name = string("op_8093_cast_fp16")];
            fp32 var_8094_epsilon_0 = const()[name = string("op_8094_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_8094_cast_fp16 = rsqrt(epsilon = var_8094_epsilon_0, x = var_8093_cast_fp16)[name = string("op_8094_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_459_cast_fp16 = mul(x = hidden_states_455_cast_fp16, y = var_8094_cast_fp16)[name = string("hidden_states_459_cast_fp16")];
            tensor<fp16, [1024]> layers_28_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_28_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(950268736)))];
            tensor<fp16, [1, 291, 1024]> input_347_cast_fp16 = mul(x = layers_28_post_attention_layernorm_weight_to_fp16, y = hidden_states_459_cast_fp16)[name = string("input_347_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_28_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_28_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(950270848)))];
            tensor<fp16, [1, 291, 4096]> linear_200_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_28_mlp_gate_proj_weight_to_fp16, x = input_347_cast_fp16)[name = string("linear_200_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_8107_cast_fp16 = silu(x = linear_200_cast_fp16)[name = string("op_8107_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_28_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_28_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(958659520)))];
            tensor<fp16, [1, 291, 4096]> linear_201_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_28_mlp_up_proj_weight_to_fp16, x = input_347_cast_fp16)[name = string("linear_201_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_351_cast_fp16 = mul(x = var_8107_cast_fp16, y = linear_201_cast_fp16)[name = string("input_351_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_28_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_28_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(967048192)))];
            tensor<fp16, [1, 291, 1024]> linear_202_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_28_mlp_down_proj_weight_to_fp16, x = input_351_cast_fp16)[name = string("linear_202_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_463_cast_fp16 = add(x = hidden_states_455_cast_fp16, y = linear_202_cast_fp16)[name = string("hidden_states_463_cast_fp16")];
            tensor<int32, [4]> var_8119 = const()[name = string("op_8119"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_8131 = const()[name = string("op_8131"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_8120_cast_fp16 = transpose(perm = var_8119, x = k_173_cast_fp16)[name = string("transpose_7")];
            tensor<fp16, [1, 1024, 1, 291]> input_353_cast_fp16 = reshape(shape = var_8131, x = var_8120_cast_fp16)[name = string("input_353_cast_fp16")];
            tensor<int32, [8]> var_8137_pad_0 = const()[name = string("op_8137_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_8137_mode_0 = const()[name = string("op_8137_mode_0"), val = string("constant")];
            fp16 const_444_to_fp16 = const()[name = string("const_444_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_8137_cast_fp16 = pad(constant_val = const_444_to_fp16, mode = var_8137_mode_0, pad = var_8137_pad_0, x = input_353_cast_fp16)[name = string("op_8137_cast_fp16")];
            tensor<int32, [4]> var_8142 = const()[name = string("op_8142"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_8154 = const()[name = string("op_8154"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_8143_cast_fp16 = transpose(perm = var_8142, x = v_115_cast_fp16)[name = string("transpose_6")];
            tensor<fp16, [1, 1024, 1, 291]> input_355_cast_fp16 = reshape(shape = var_8154, x = var_8143_cast_fp16)[name = string("input_355_cast_fp16")];
            tensor<int32, [8]> var_8160_pad_0 = const()[name = string("op_8160_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_8160_mode_0 = const()[name = string("op_8160_mode_0"), val = string("constant")];
            fp16 const_447_to_fp16 = const()[name = string("const_447_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_8160_cast_fp16 = pad(constant_val = const_447_to_fp16, mode = var_8160_mode_0, pad = var_8160_pad_0, x = input_355_cast_fp16)[name = string("op_8160_cast_fp16")];
            fp16 var_8164_promoted_to_fp16 = const()[name = string("op_8164_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_8170_cast_fp16 = pow(x = hidden_states_463_cast_fp16, y = var_8164_promoted_to_fp16)[name = string("op_8170_cast_fp16")];
            tensor<int32, [1]> variance_117_axes_0 = const()[name = string("variance_117_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_117_keep_dims_0 = const()[name = string("variance_117_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_117_cast_fp16 = reduce_mean(axes = variance_117_axes_0, keep_dims = variance_117_keep_dims_0, x = var_8170_cast_fp16)[name = string("variance_117_cast_fp16")];
            fp16 var_8173_to_fp16 = const()[name = string("op_8173_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_8174_cast_fp16 = add(x = variance_117_cast_fp16, y = var_8173_to_fp16)[name = string("op_8174_cast_fp16")];
            fp32 var_8175_epsilon_0 = const()[name = string("op_8175_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_8175_cast_fp16 = rsqrt(epsilon = var_8175_epsilon_0, x = var_8174_cast_fp16)[name = string("op_8175_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_467_cast_fp16 = mul(x = hidden_states_463_cast_fp16, y = var_8175_cast_fp16)[name = string("hidden_states_467_cast_fp16")];
            tensor<fp16, [1024]> layers_29_input_layernorm_weight_to_fp16 = const()[name = string("layers_29_input_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(975436864)))];
            tensor<fp16, [1, 291, 1024]> hidden_117_cast_fp16 = mul(x = layers_29_input_layernorm_weight_to_fp16, y = hidden_states_467_cast_fp16)[name = string("hidden_117_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_29_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(975438976)))];
            tensor<fp16, [1, 291, 1024]> linear_203_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_29_self_attn_q_proj_weight_to_fp16, x = hidden_117_cast_fp16)[name = string("linear_203_cast_fp16")];
            tensor<int32, [4]> var_8199 = const()[name = string("op_8199"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> q_175_cast_fp16 = reshape(shape = var_8199, x = linear_203_cast_fp16)[name = string("q_175_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_29_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(977536192)))];
            tensor<fp16, [1, 291, 1024]> linear_204_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_29_self_attn_k_proj_weight_to_fp16, x = hidden_117_cast_fp16)[name = string("linear_204_cast_fp16")];
            tensor<int32, [4]> var_8206 = const()[name = string("op_8206"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> k_175_cast_fp16 = reshape(shape = var_8206, x = linear_204_cast_fp16)[name = string("k_175_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_29_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(979633408)))];
            tensor<fp16, [1, 291, 1024]> linear_205_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_29_self_attn_v_proj_weight_to_fp16, x = hidden_117_cast_fp16)[name = string("linear_205_cast_fp16")];
            tensor<int32, [4]> var_8213 = const()[name = string("op_8213"), val = tensor<int32, [4]>([1, 291, 16, 64])];
            tensor<fp16, [1, 291, 16, 64]> v_117_cast_fp16 = reshape(shape = var_8213, x = linear_205_cast_fp16)[name = string("v_117_cast_fp16")];
            tensor<int32, [4]> q_177_perm_0 = const()[name = string("q_177_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> k_177_perm_0 = const()[name = string("k_177_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [4]> v_perm_0 = const()[name = string("v_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<fp16, [1, 16, 291, 64]> q_177_cast_fp16 = transpose(perm = q_177_perm_0, x = q_175_cast_fp16)[name = string("transpose_5")];
            tensor<fp16, [1, 16, 291, 64]> var_8226_cast_fp16 = mul(x = q_177_cast_fp16, y = var_1091_to_fp16)[name = string("op_8226_cast_fp16")];
            tensor<int32, [4]> x1_117_begin_0 = const()[name = string("x1_117_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_117_end_0 = const()[name = string("x1_117_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_117_end_mask_0 = const()[name = string("x1_117_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_117_cast_fp16 = slice_by_index(begin = x1_117_begin_0, end = x1_117_end_0, end_mask = x1_117_end_mask_0, x = q_177_cast_fp16)[name = string("x1_117_cast_fp16")];
            tensor<int32, [4]> x2_117_begin_0 = const()[name = string("x2_117_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_117_end_0 = const()[name = string("x2_117_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_117_end_mask_0 = const()[name = string("x2_117_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_117_cast_fp16 = slice_by_index(begin = x2_117_begin_0, end = x2_117_end_0, end_mask = x2_117_end_mask_0, x = q_177_cast_fp16)[name = string("x2_117_cast_fp16")];
            fp16 const_452_promoted_to_fp16 = const()[name = string("const_452_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_8247_cast_fp16 = mul(x = x2_117_cast_fp16, y = const_452_promoted_to_fp16)[name = string("op_8247_cast_fp16")];
            int32 var_8249 = const()[name = string("op_8249"), val = int32(-1)];
            bool var_8250_interleave_0 = const()[name = string("op_8250_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_8250_cast_fp16 = concat(axis = var_8249, interleave = var_8250_interleave_0, values = (var_8247_cast_fp16, x1_117_cast_fp16))[name = string("op_8250_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_8253_cast_fp16 = mul(x = var_8250_cast_fp16, y = var_1118_to_fp16)[name = string("op_8253_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> q_cast_fp16 = add(x = var_8226_cast_fp16, y = var_8253_cast_fp16)[name = string("q_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_177_cast_fp16 = transpose(perm = k_177_perm_0, x = k_175_cast_fp16)[name = string("transpose_4")];
            tensor<fp16, [1, 16, 291, 64]> var_8258_cast_fp16 = mul(x = k_177_cast_fp16, y = var_1091_to_fp16)[name = string("op_8258_cast_fp16")];
            tensor<int32, [4]> x1_begin_0 = const()[name = string("x1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> x1_end_0 = const()[name = string("x1_end_0"), val = tensor<int32, [4]>([1, 16, 291, 32])];
            tensor<bool, [4]> x1_end_mask_0 = const()[name = string("x1_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 16, 291, 32]> x1_cast_fp16 = slice_by_index(begin = x1_begin_0, end = x1_end_0, end_mask = x1_end_mask_0, x = k_177_cast_fp16)[name = string("x1_cast_fp16")];
            tensor<int32, [4]> x2_begin_0 = const()[name = string("x2_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 32])];
            tensor<int32, [4]> x2_end_0 = const()[name = string("x2_end_0"), val = tensor<int32, [4]>([1, 16, 291, 64])];
            tensor<bool, [4]> x2_end_mask_0 = const()[name = string("x2_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 16, 291, 32]> x2_cast_fp16 = slice_by_index(begin = x2_begin_0, end = x2_end_0, end_mask = x2_end_mask_0, x = k_177_cast_fp16)[name = string("x2_cast_fp16")];
            fp16 const_455_promoted_to_fp16 = const()[name = string("const_455_promoted_to_fp16"), val = fp16(-0x1p+0)];
            tensor<fp16, [1, 16, 291, 32]> var_8279_cast_fp16 = mul(x = x2_cast_fp16, y = const_455_promoted_to_fp16)[name = string("op_8279_cast_fp16")];
            int32 var_8281 = const()[name = string("op_8281"), val = int32(-1)];
            bool var_8282_interleave_0 = const()[name = string("op_8282_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> var_8282_cast_fp16 = concat(axis = var_8281, interleave = var_8282_interleave_0, values = (var_8279_cast_fp16, x1_cast_fp16))[name = string("op_8282_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> var_8285_cast_fp16 = mul(x = var_8282_cast_fp16, y = var_1118_to_fp16)[name = string("op_8285_cast_fp16")];
            tensor<fp16, [1, 16, 291, 64]> k_cast_fp16 = add(x = var_8258_cast_fp16, y = var_8285_cast_fp16)[name = string("k_cast_fp16")];
            bool var_8291_transpose_x_1 = const()[name = string("op_8291_transpose_x_1"), val = bool(false)];
            bool var_8291_transpose_y_1 = const()[name = string("op_8291_transpose_y_1"), val = bool(true)];
            tensor<fp16, [1, 16, 291, 291]> var_8291_cast_fp16 = matmul(transpose_x = var_8291_transpose_x_1, transpose_y = var_8291_transpose_y_1, x = q_cast_fp16, y = k_cast_fp16)[name = string("op_8291_cast_fp16")];
            fp16 _inversed_scores_175_y_0_to_fp16 = const()[name = string("_inversed_scores_175_y_0_to_fp16"), val = fp16(0x1p-3)];
            tensor<fp16, [1, 16, 291, 291]> _inversed_scores_175_cast_fp16 = mul(x = var_8291_cast_fp16, y = _inversed_scores_175_y_0_to_fp16)[name = string("_inversed_scores_175_cast_fp16")];
            tensor<fp16, [1, 16, 291, 291]> scores_177_cast_fp16 = add(x = _inversed_scores_175_cast_fp16, y = const_21_to_fp16)[name = string("scores_177_cast_fp16")];
            int32 var_8306 = const()[name = string("op_8306"), val = int32(-1)];
            tensor<fp16, [1, 16, 291, 291]> var_8308_cast_fp16 = softmax(axis = var_8306, x = scores_177_cast_fp16)[name = string("op_8308_cast_fp16")];
            bool attn_out_117_transpose_x_0 = const()[name = string("attn_out_117_transpose_x_0"), val = bool(false)];
            bool attn_out_117_transpose_y_0 = const()[name = string("attn_out_117_transpose_y_0"), val = bool(false)];
            tensor<fp16, [1, 16, 291, 64]> v_cast_fp16 = transpose(perm = v_perm_0, x = v_117_cast_fp16)[name = string("transpose_3")];
            tensor<fp16, [1, 16, 291, 64]> attn_out_117_cast_fp16 = matmul(transpose_x = attn_out_117_transpose_x_0, transpose_y = attn_out_117_transpose_y_0, x = var_8308_cast_fp16, y = v_cast_fp16)[name = string("attn_out_117_cast_fp16")];
            tensor<int32, [4]> var_8317_perm_0 = const()[name = string("op_8317_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_8319 = const()[name = string("op_8319"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<fp16, [1, 291, 16, 64]> var_8317_cast_fp16 = transpose(perm = var_8317_perm_0, x = attn_out_117_cast_fp16)[name = string("transpose_2")];
            tensor<fp16, [1, 291, 1024]> input_357_cast_fp16 = reshape(shape = var_8319, x = var_8317_cast_fp16)[name = string("input_357_cast_fp16")];
            tensor<fp16, [1024, 1024]> layers_29_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_29_self_attn_o_proj_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(981730624)))];
            tensor<fp16, [1, 291, 1024]> linear_206_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_29_self_attn_o_proj_weight_to_fp16, x = input_357_cast_fp16)[name = string("linear_206_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_471_cast_fp16 = add(x = hidden_states_463_cast_fp16, y = linear_206_cast_fp16)[name = string("hidden_states_471_cast_fp16")];
            fp16 var_8329_promoted_to_fp16 = const()[name = string("op_8329_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_8335_cast_fp16 = pow(x = hidden_states_471_cast_fp16, y = var_8329_promoted_to_fp16)[name = string("op_8335_cast_fp16")];
            tensor<int32, [1]> variance_119_axes_0 = const()[name = string("variance_119_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_119_keep_dims_0 = const()[name = string("variance_119_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_119_cast_fp16 = reduce_mean(axes = variance_119_axes_0, keep_dims = variance_119_keep_dims_0, x = var_8335_cast_fp16)[name = string("variance_119_cast_fp16")];
            fp16 var_8338_to_fp16 = const()[name = string("op_8338_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_8339_cast_fp16 = add(x = variance_119_cast_fp16, y = var_8338_to_fp16)[name = string("op_8339_cast_fp16")];
            fp32 var_8340_epsilon_0 = const()[name = string("op_8340_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_8340_cast_fp16 = rsqrt(epsilon = var_8340_epsilon_0, x = var_8339_cast_fp16)[name = string("op_8340_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_475_cast_fp16 = mul(x = hidden_states_471_cast_fp16, y = var_8340_cast_fp16)[name = string("hidden_states_475_cast_fp16")];
            tensor<fp16, [1024]> layers_29_post_attention_layernorm_weight_to_fp16 = const()[name = string("layers_29_post_attention_layernorm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(983827840)))];
            tensor<fp16, [1, 291, 1024]> input_359_cast_fp16 = mul(x = layers_29_post_attention_layernorm_weight_to_fp16, y = hidden_states_475_cast_fp16)[name = string("input_359_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_29_mlp_gate_proj_weight_to_fp16 = const()[name = string("layers_29_mlp_gate_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(983829952)))];
            tensor<fp16, [1, 291, 4096]> linear_207_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_29_mlp_gate_proj_weight_to_fp16, x = input_359_cast_fp16)[name = string("linear_207_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> var_8353_cast_fp16 = silu(x = linear_207_cast_fp16)[name = string("op_8353_cast_fp16")];
            tensor<fp16, [4096, 1024]> layers_29_mlp_up_proj_weight_to_fp16 = const()[name = string("layers_29_mlp_up_proj_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(992218624)))];
            tensor<fp16, [1, 291, 4096]> linear_208_cast_fp16 = linear(bias = linear_4_bias_0_to_fp16, weight = layers_29_mlp_up_proj_weight_to_fp16, x = input_359_cast_fp16)[name = string("linear_208_cast_fp16")];
            tensor<fp16, [1, 291, 4096]> input_363_cast_fp16 = mul(x = var_8353_cast_fp16, y = linear_208_cast_fp16)[name = string("input_363_cast_fp16")];
            tensor<fp16, [1024, 4096]> layers_29_mlp_down_proj_weight_to_fp16 = const()[name = string("layers_29_mlp_down_proj_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1000607296)))];
            tensor<fp16, [1, 291, 1024]> linear_209_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = layers_29_mlp_down_proj_weight_to_fp16, x = input_363_cast_fp16)[name = string("linear_209_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_479_cast_fp16 = add(x = hidden_states_471_cast_fp16, y = linear_209_cast_fp16)[name = string("hidden_states_479_cast_fp16")];
            tensor<int32, [4]> var_8365 = const()[name = string("op_8365"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_8377 = const()[name = string("op_8377"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_8366_cast_fp16 = transpose(perm = var_8365, x = k_cast_fp16)[name = string("transpose_1")];
            tensor<fp16, [1, 1024, 1, 291]> input_365_cast_fp16 = reshape(shape = var_8377, x = var_8366_cast_fp16)[name = string("input_365_cast_fp16")];
            tensor<int32, [8]> var_8383_pad_0 = const()[name = string("op_8383_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_8383_mode_0 = const()[name = string("op_8383_mode_0"), val = string("constant")];
            fp16 const_459_to_fp16 = const()[name = string("const_459_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_8383_cast_fp16 = pad(constant_val = const_459_to_fp16, mode = var_8383_mode_0, pad = var_8383_pad_0, x = input_365_cast_fp16)[name = string("op_8383_cast_fp16")];
            tensor<int32, [4]> var_8388 = const()[name = string("op_8388"), val = tensor<int32, [4]>([0, 1, 3, 2])];
            tensor<int32, [4]> var_8400 = const()[name = string("op_8400"), val = tensor<int32, [4]>([1, 1024, 1, 291])];
            tensor<fp16, [1, 16, 64, 291]> var_8389_cast_fp16 = transpose(perm = var_8388, x = v_cast_fp16)[name = string("transpose_0")];
            tensor<fp16, [1, 1024, 1, 291]> input_cast_fp16 = reshape(shape = var_8400, x = var_8389_cast_fp16)[name = string("input_cast_fp16")];
            tensor<int32, [8]> var_8406_pad_0 = const()[name = string("op_8406_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 0, 733])];
            string var_8406_mode_0 = const()[name = string("op_8406_mode_0"), val = string("constant")];
            fp16 const_462_to_fp16 = const()[name = string("const_462_to_fp16"), val = fp16(0x0p+0)];
            tensor<fp16, [1, 1024, 1, 1024]> var_8406_cast_fp16 = pad(constant_val = const_462_to_fp16, mode = var_8406_mode_0, pad = var_8406_pad_0, x = input_cast_fp16)[name = string("op_8406_cast_fp16")];
            fp16 var_8410_promoted_to_fp16 = const()[name = string("op_8410_promoted_to_fp16"), val = fp16(0x1p+1)];
            tensor<fp16, [1, 291, 1024]> var_8416_cast_fp16 = pow(x = hidden_states_479_cast_fp16, y = var_8410_promoted_to_fp16)[name = string("op_8416_cast_fp16")];
            tensor<int32, [1]> variance_axes_0 = const()[name = string("variance_axes_0"), val = tensor<int32, [1]>([-1])];
            bool variance_keep_dims_0 = const()[name = string("variance_keep_dims_0"), val = bool(true)];
            tensor<fp16, [1, 291, 1]> variance_cast_fp16 = reduce_mean(axes = variance_axes_0, keep_dims = variance_keep_dims_0, x = var_8416_cast_fp16)[name = string("variance_cast_fp16")];
            fp16 var_8419_to_fp16 = const()[name = string("op_8419_to_fp16"), val = fp16(0x1.5p-17)];
            tensor<fp16, [1, 291, 1]> var_8420_cast_fp16 = add(x = variance_cast_fp16, y = var_8419_to_fp16)[name = string("op_8420_cast_fp16")];
            fp32 var_8421_epsilon_0 = const()[name = string("op_8421_epsilon_0"), val = fp32(0x1.197998p-40)];
            tensor<fp16, [1, 291, 1]> var_8421_cast_fp16 = rsqrt(epsilon = var_8421_epsilon_0, x = var_8420_cast_fp16)[name = string("op_8421_cast_fp16")];
            tensor<fp16, [1, 291, 1024]> hidden_states_483_cast_fp16 = mul(x = hidden_states_479_cast_fp16, y = var_8421_cast_fp16)[name = string("hidden_states_483_cast_fp16")];
            tensor<fp16, [1024]> norm_weight_to_fp16 = const()[name = string("norm_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1008995968)))];
            tensor<fp16, [1, 291, 1024]> hidden_cast_fp16 = mul(x = norm_weight_to_fp16, y = hidden_states_483_cast_fp16)[name = string("hidden_cast_fp16")];
            int32 var_8426 = const()[name = string("op_8426"), val = int32(1)];
            bool var_8427_interleave_0 = const()[name = string("op_8427_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 30720, 1, 1024]> key_cache = concat(axis = var_8426, interleave = var_8427_interleave_0, values = (var_1249_cast_fp16, var_1495_cast_fp16, var_1741_cast_fp16, var_1987_cast_fp16, var_2233_cast_fp16, var_2479_cast_fp16, var_2725_cast_fp16, var_2971_cast_fp16, var_3217_cast_fp16, var_3463_cast_fp16, var_3709_cast_fp16, var_3955_cast_fp16, var_4201_cast_fp16, var_4447_cast_fp16, var_4693_cast_fp16, var_4939_cast_fp16, var_5185_cast_fp16, var_5431_cast_fp16, var_5677_cast_fp16, var_5923_cast_fp16, var_6169_cast_fp16, var_6415_cast_fp16, var_6661_cast_fp16, var_6907_cast_fp16, var_7153_cast_fp16, var_7399_cast_fp16, var_7645_cast_fp16, var_7891_cast_fp16, var_8137_cast_fp16, var_8383_cast_fp16))[name = string("op_8427_cast_fp16")];
            int32 var_8429 = const()[name = string("op_8429"), val = int32(1)];
            bool var_8430_interleave_0 = const()[name = string("op_8430_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 30720, 1, 1024]> value_cache = concat(axis = var_8429, interleave = var_8430_interleave_0, values = (var_1272_cast_fp16, var_1518_cast_fp16, var_1764_cast_fp16, var_2010_cast_fp16, var_2256_cast_fp16, var_2502_cast_fp16, var_2748_cast_fp16, var_2994_cast_fp16, var_3240_cast_fp16, var_3486_cast_fp16, var_3732_cast_fp16, var_3978_cast_fp16, var_4224_cast_fp16, var_4470_cast_fp16, var_4716_cast_fp16, var_4962_cast_fp16, var_5208_cast_fp16, var_5454_cast_fp16, var_5700_cast_fp16, var_5946_cast_fp16, var_6192_cast_fp16, var_6438_cast_fp16, var_6684_cast_fp16, var_6930_cast_fp16, var_7176_cast_fp16, var_7422_cast_fp16, var_7668_cast_fp16, var_7914_cast_fp16, var_8160_cast_fp16, var_8406_cast_fp16))[name = string("op_8430_cast_fp16")];
            tensor<int32, [3]> var_8440_begin_0 = const()[name = string("op_8440_begin_0"), val = tensor<int32, [3]>([0, -1, 0])];
            tensor<int32, [3]> var_8440_end_0 = const()[name = string("op_8440_end_0"), val = tensor<int32, [3]>([1, 291, 1024])];
            tensor<bool, [3]> var_8440_end_mask_0 = const()[name = string("op_8440_end_mask_0"), val = tensor<bool, [3]>([true, true, true])];
            tensor<fp16, [1, 1, 1024]> shift_ctx = slice_by_index(begin = var_8440_begin_0, end = var_8440_end_0, end_mask = var_8440_end_mask_0, x = hidden_cast_fp16)[name = string("op_8440_cast_fp16")];
        } -> (shift_ctx, key_cache, value_cache);
}