alexwengg's picture
Upload 16 files
2e263b4 verified
program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3505.3.2"}, {"coremlc-version", "3505.4.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
func main<ios18>(tensor<fp32, [1, 32]> latent, tensor<fp32, [1, 1]> s, tensor<fp32, [1, 1]> t, tensor<fp32, [1, 1024]> transformer_out) {
tensor<fp32, [512]> flow_net_input_proj_bias = const()[name = string("flow_net_input_proj_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
tensor<fp32, [512, 32]> flow_net_input_proj_weight = const()[name = string("flow_net_input_proj_weight"), val = tensor<fp32, [512, 32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2176)))];
tensor<fp32, [512]> flow_net_time_embed_0_mlp_0_bias = const()[name = string("flow_net_time_embed_0_mlp_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67776)))];
tensor<fp32, [512, 256]> flow_net_time_embed_0_mlp_0_weight = const()[name = string("flow_net_time_embed_0_mlp_0_weight"), val = tensor<fp32, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(69888)))];
tensor<fp32, [512]> flow_net_time_embed_0_mlp_2_bias = const()[name = string("flow_net_time_embed_0_mlp_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(594240)))];
tensor<fp32, [512, 512]> flow_net_time_embed_0_mlp_2_weight = const()[name = string("flow_net_time_embed_0_mlp_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(596352)))];
tensor<fp32, [512]> flow_net_time_embed_1_mlp_0_bias = const()[name = string("flow_net_time_embed_1_mlp_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1644992)))];
tensor<fp32, [512, 256]> flow_net_time_embed_1_mlp_0_weight = const()[name = string("flow_net_time_embed_1_mlp_0_weight"), val = tensor<fp32, [512, 256]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1647104)))];
tensor<fp32, [512]> flow_net_time_embed_1_mlp_2_bias = const()[name = string("flow_net_time_embed_1_mlp_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2171456)))];
tensor<fp32, [512, 512]> flow_net_time_embed_1_mlp_2_weight = const()[name = string("flow_net_time_embed_1_mlp_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(2173568)))];
tensor<fp32, [512]> flow_net_cond_embed_bias = const()[name = string("flow_net_cond_embed_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3222208)))];
tensor<fp32, [512, 1024]> flow_net_cond_embed_weight = const()[name = string("flow_net_cond_embed_weight"), val = tensor<fp32, [512, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3224320)))];
tensor<fp32, [1536]> flow_net_res_blocks_0_adaLN_modulation_1_bias = const()[name = string("flow_net_res_blocks_0_adaLN_modulation_1_bias"), val = tensor<fp32, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5321536)))];
tensor<fp32, [1536, 512]> flow_net_res_blocks_0_adaLN_modulation_1_weight = const()[name = string("flow_net_res_blocks_0_adaLN_modulation_1_weight"), val = tensor<fp32, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5327744)))];
tensor<fp32, [512]> flow_net_res_blocks_0_in_ln_bias = const()[name = string("flow_net_res_blocks_0_in_ln_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8473536)))];
tensor<fp32, [512]> flow_net_res_blocks_0_in_ln_weight = const()[name = string("flow_net_res_blocks_0_in_ln_weight"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8475648)))];
tensor<fp32, [512]> flow_net_res_blocks_0_mlp_0_bias = const()[name = string("flow_net_res_blocks_0_mlp_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8477760)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_0_mlp_0_weight = const()[name = string("flow_net_res_blocks_0_mlp_0_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8479872)))];
tensor<fp32, [512]> flow_net_res_blocks_0_mlp_2_bias = const()[name = string("flow_net_res_blocks_0_mlp_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9528512)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_0_mlp_2_weight = const()[name = string("flow_net_res_blocks_0_mlp_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9530624)))];
tensor<fp32, [1536]> flow_net_res_blocks_1_adaLN_modulation_1_bias = const()[name = string("flow_net_res_blocks_1_adaLN_modulation_1_bias"), val = tensor<fp32, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10579264)))];
tensor<fp32, [1536, 512]> flow_net_res_blocks_1_adaLN_modulation_1_weight = const()[name = string("flow_net_res_blocks_1_adaLN_modulation_1_weight"), val = tensor<fp32, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10585472)))];
tensor<fp32, [512]> flow_net_res_blocks_1_in_ln_bias = const()[name = string("flow_net_res_blocks_1_in_ln_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13731264)))];
tensor<fp32, [512]> flow_net_res_blocks_1_in_ln_weight = const()[name = string("flow_net_res_blocks_1_in_ln_weight"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13733376)))];
tensor<fp32, [512]> flow_net_res_blocks_1_mlp_0_bias = const()[name = string("flow_net_res_blocks_1_mlp_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13735488)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_1_mlp_0_weight = const()[name = string("flow_net_res_blocks_1_mlp_0_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13737600)))];
tensor<fp32, [512]> flow_net_res_blocks_1_mlp_2_bias = const()[name = string("flow_net_res_blocks_1_mlp_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14786240)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_1_mlp_2_weight = const()[name = string("flow_net_res_blocks_1_mlp_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14788352)))];
tensor<fp32, [1536]> flow_net_res_blocks_2_adaLN_modulation_1_bias = const()[name = string("flow_net_res_blocks_2_adaLN_modulation_1_bias"), val = tensor<fp32, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15836992)))];
tensor<fp32, [1536, 512]> flow_net_res_blocks_2_adaLN_modulation_1_weight = const()[name = string("flow_net_res_blocks_2_adaLN_modulation_1_weight"), val = tensor<fp32, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15843200)))];
tensor<fp32, [512]> flow_net_res_blocks_2_in_ln_bias = const()[name = string("flow_net_res_blocks_2_in_ln_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18988992)))];
tensor<fp32, [512]> flow_net_res_blocks_2_in_ln_weight = const()[name = string("flow_net_res_blocks_2_in_ln_weight"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18991104)))];
tensor<fp32, [512]> flow_net_res_blocks_2_mlp_0_bias = const()[name = string("flow_net_res_blocks_2_mlp_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18993216)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_2_mlp_0_weight = const()[name = string("flow_net_res_blocks_2_mlp_0_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18995328)))];
tensor<fp32, [512]> flow_net_res_blocks_2_mlp_2_bias = const()[name = string("flow_net_res_blocks_2_mlp_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20043968)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_2_mlp_2_weight = const()[name = string("flow_net_res_blocks_2_mlp_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20046080)))];
tensor<fp32, [1536]> flow_net_res_blocks_3_adaLN_modulation_1_bias = const()[name = string("flow_net_res_blocks_3_adaLN_modulation_1_bias"), val = tensor<fp32, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21094720)))];
tensor<fp32, [1536, 512]> flow_net_res_blocks_3_adaLN_modulation_1_weight = const()[name = string("flow_net_res_blocks_3_adaLN_modulation_1_weight"), val = tensor<fp32, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21100928)))];
tensor<fp32, [512]> flow_net_res_blocks_3_in_ln_bias = const()[name = string("flow_net_res_blocks_3_in_ln_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24246720)))];
tensor<fp32, [512]> flow_net_res_blocks_3_in_ln_weight = const()[name = string("flow_net_res_blocks_3_in_ln_weight"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24248832)))];
tensor<fp32, [512]> flow_net_res_blocks_3_mlp_0_bias = const()[name = string("flow_net_res_blocks_3_mlp_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24250944)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_3_mlp_0_weight = const()[name = string("flow_net_res_blocks_3_mlp_0_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24253056)))];
tensor<fp32, [512]> flow_net_res_blocks_3_mlp_2_bias = const()[name = string("flow_net_res_blocks_3_mlp_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25301696)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_3_mlp_2_weight = const()[name = string("flow_net_res_blocks_3_mlp_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25303808)))];
tensor<fp32, [1536]> flow_net_res_blocks_4_adaLN_modulation_1_bias = const()[name = string("flow_net_res_blocks_4_adaLN_modulation_1_bias"), val = tensor<fp32, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26352448)))];
tensor<fp32, [1536, 512]> flow_net_res_blocks_4_adaLN_modulation_1_weight = const()[name = string("flow_net_res_blocks_4_adaLN_modulation_1_weight"), val = tensor<fp32, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26358656)))];
tensor<fp32, [512]> flow_net_res_blocks_4_in_ln_bias = const()[name = string("flow_net_res_blocks_4_in_ln_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29504448)))];
tensor<fp32, [512]> flow_net_res_blocks_4_in_ln_weight = const()[name = string("flow_net_res_blocks_4_in_ln_weight"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29506560)))];
tensor<fp32, [512]> flow_net_res_blocks_4_mlp_0_bias = const()[name = string("flow_net_res_blocks_4_mlp_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29508672)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_4_mlp_0_weight = const()[name = string("flow_net_res_blocks_4_mlp_0_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29510784)))];
tensor<fp32, [512]> flow_net_res_blocks_4_mlp_2_bias = const()[name = string("flow_net_res_blocks_4_mlp_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30559424)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_4_mlp_2_weight = const()[name = string("flow_net_res_blocks_4_mlp_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30561536)))];
tensor<fp32, [1536]> flow_net_res_blocks_5_adaLN_modulation_1_bias = const()[name = string("flow_net_res_blocks_5_adaLN_modulation_1_bias"), val = tensor<fp32, [1536]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31610176)))];
tensor<fp32, [1536, 512]> flow_net_res_blocks_5_adaLN_modulation_1_weight = const()[name = string("flow_net_res_blocks_5_adaLN_modulation_1_weight"), val = tensor<fp32, [1536, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(31616384)))];
tensor<fp32, [512]> flow_net_res_blocks_5_in_ln_bias = const()[name = string("flow_net_res_blocks_5_in_ln_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34762176)))];
tensor<fp32, [512]> flow_net_res_blocks_5_in_ln_weight = const()[name = string("flow_net_res_blocks_5_in_ln_weight"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34764288)))];
tensor<fp32, [512]> flow_net_res_blocks_5_mlp_0_bias = const()[name = string("flow_net_res_blocks_5_mlp_0_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34766400)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_5_mlp_0_weight = const()[name = string("flow_net_res_blocks_5_mlp_0_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34768512)))];
tensor<fp32, [512]> flow_net_res_blocks_5_mlp_2_bias = const()[name = string("flow_net_res_blocks_5_mlp_2_bias"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35817152)))];
tensor<fp32, [512, 512]> flow_net_res_blocks_5_mlp_2_weight = const()[name = string("flow_net_res_blocks_5_mlp_2_weight"), val = tensor<fp32, [512, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35819264)))];
tensor<fp32, [1024]> flow_net_final_layer_adaLN_modulation_1_bias = const()[name = string("flow_net_final_layer_adaLN_modulation_1_bias"), val = tensor<fp32, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36867904)))];
tensor<fp32, [1024, 512]> flow_net_final_layer_adaLN_modulation_1_weight = const()[name = string("flow_net_final_layer_adaLN_modulation_1_weight"), val = tensor<fp32, [1024, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36872064)))];
tensor<fp32, [32]> flow_net_final_layer_linear_bias = const()[name = string("flow_net_final_layer_linear_bias"), val = tensor<fp32, [32]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38969280)))];
tensor<fp32, [32, 512]> flow_net_final_layer_linear_weight = const()[name = string("flow_net_final_layer_linear_weight"), val = tensor<fp32, [32, 512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38969472)))];
int32 var_9 = const()[name = string("op_9"), val = int32(-1)];
tensor<fp32, [1, 512]> x_5 = linear(bias = flow_net_input_proj_bias, weight = flow_net_input_proj_weight, x = latent)[name = string("linear_0")];
tensor<fp32, [128]> const_0 = const()[name = string("const_0"), val = tensor<fp32, [128]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39035072)))];
tensor<fp32, [1, 128]> args_1 = mul(x = s, y = const_0)[name = string("args_1")];
tensor<fp32, [1, 128]> var_39 = cos(x = args_1)[name = string("op_39")];
tensor<fp32, [1, 128]> var_40 = sin(x = args_1)[name = string("op_40")];
bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)];
tensor<fp32, [1, 256]> input_1 = concat(axis = var_9, interleave = input_1_interleave_0, values = (var_39, var_40))[name = string("input_1")];
tensor<fp32, [1, 512]> input_3 = linear(bias = flow_net_time_embed_0_mlp_0_bias, weight = flow_net_time_embed_0_mlp_0_weight, x = input_1)[name = string("linear_1")];
tensor<fp32, [1, 512]> input_5 = silu(x = input_3)[name = string("input_5")];
tensor<fp32, [1, 512]> x_1 = linear(bias = flow_net_time_embed_0_mlp_2_bias, weight = flow_net_time_embed_0_mlp_2_weight, x = input_5)[name = string("linear_2")];
tensor<int32, [1]> reduce_mean_0_axes_0 = const()[name = string("reduce_mean_0_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_0_keep_dims_0 = const()[name = string("reduce_mean_0_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_0 = reduce_mean(axes = reduce_mean_0_axes_0, keep_dims = reduce_mean_0_keep_dims_0, x = x_1)[name = string("reduce_mean_0")];
tensor<fp32, [1, 512]> sub_0 = sub(x = x_1, y = reduce_mean_0)[name = string("sub_0")];
tensor<fp32, [1, 512]> square_0 = square(x = sub_0)[name = string("square_0")];
tensor<int32, [1]> reduce_mean_1_axes_0 = const()[name = string("reduce_mean_1_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_1_keep_dims_0 = const()[name = string("reduce_mean_1_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_1 = reduce_mean(axes = reduce_mean_1_axes_0, keep_dims = reduce_mean_1_keep_dims_0, x = square_0)[name = string("reduce_mean_1")];
fp32 real_div_0 = const()[name = string("real_div_0"), val = fp32(0x1.00804p+0)];
tensor<fp32, [1, 1]> mul_0 = mul(x = reduce_mean_1, y = real_div_0)[name = string("mul_0")];
fp32 var_56 = const()[name = string("op_56"), val = fp32(0x1.4f8b58p-17)];
tensor<fp32, [1, 1]> var_1 = add(x = mul_0, y = var_56)[name = string("var_1")];
tensor<fp32, [512]> const_1 = const()[name = string("const_1"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39035648)))];
fp32 var_59_epsilon_0 = const()[name = string("op_59_epsilon_0"), val = fp32(0x1.197998p-40)];
tensor<fp32, [1, 1]> var_59 = rsqrt(epsilon = var_59_epsilon_0, x = var_1)[name = string("op_59")];
tensor<fp32, [1, 512]> var_60 = mul(x = const_1, y = var_59)[name = string("op_60")];
tensor<fp32, [1, 512]> var_61 = mul(x = x_1, y = var_60)[name = string("op_61")];
tensor<fp32, [1, 128]> args = mul(x = t, y = const_0)[name = string("args")];
tensor<fp32, [1, 128]> var_69 = cos(x = args)[name = string("op_69")];
tensor<fp32, [1, 128]> var_70 = sin(x = args)[name = string("op_70")];
bool input_7_interleave_0 = const()[name = string("input_7_interleave_0"), val = bool(false)];
tensor<fp32, [1, 256]> input_7 = concat(axis = var_9, interleave = input_7_interleave_0, values = (var_69, var_70))[name = string("input_7")];
tensor<fp32, [1, 512]> input_9 = linear(bias = flow_net_time_embed_1_mlp_0_bias, weight = flow_net_time_embed_1_mlp_0_weight, x = input_7)[name = string("linear_3")];
tensor<fp32, [1, 512]> input_11 = silu(x = input_9)[name = string("input_11")];
tensor<fp32, [1, 512]> x_3 = linear(bias = flow_net_time_embed_1_mlp_2_bias, weight = flow_net_time_embed_1_mlp_2_weight, x = input_11)[name = string("linear_4")];
tensor<int32, [1]> reduce_mean_2_axes_0 = const()[name = string("reduce_mean_2_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_2_keep_dims_0 = const()[name = string("reduce_mean_2_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_2 = reduce_mean(axes = reduce_mean_2_axes_0, keep_dims = reduce_mean_2_keep_dims_0, x = x_3)[name = string("reduce_mean_2")];
tensor<fp32, [1, 512]> sub_2 = sub(x = x_3, y = reduce_mean_2)[name = string("sub_2")];
tensor<fp32, [1, 512]> square_1 = square(x = sub_2)[name = string("square_1")];
tensor<int32, [1]> reduce_mean_3_axes_0 = const()[name = string("reduce_mean_3_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_3_keep_dims_0 = const()[name = string("reduce_mean_3_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_3 = reduce_mean(axes = reduce_mean_3_axes_0, keep_dims = reduce_mean_3_keep_dims_0, x = square_1)[name = string("reduce_mean_3")];
fp32 real_div_1 = const()[name = string("real_div_1"), val = fp32(0x1.00804p+0)];
tensor<fp32, [1, 1]> mul_1 = mul(x = reduce_mean_3, y = real_div_1)[name = string("mul_1")];
fp32 var_86 = const()[name = string("op_86"), val = fp32(0x1.4f8b58p-17)];
tensor<fp32, [1, 1]> var_3 = add(x = mul_1, y = var_86)[name = string("var_3")];
tensor<fp32, [512]> const_3 = const()[name = string("const_3"), val = tensor<fp32, [512]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39037760)))];
fp32 var_89_epsilon_0 = const()[name = string("op_89_epsilon_0"), val = fp32(0x1.197998p-40)];
tensor<fp32, [1, 1]> var_89 = rsqrt(epsilon = var_89_epsilon_0, x = var_3)[name = string("op_89")];
tensor<fp32, [1, 512]> var_90 = mul(x = const_3, y = var_89)[name = string("op_90")];
tensor<fp32, [1, 512]> var_91 = mul(x = x_3, y = var_90)[name = string("op_91")];
tensor<fp32, [1, 512]> var_93 = add(x = var_61, y = var_91)[name = string("op_93")];
fp32 _inversed_t_combined_y_0 = const()[name = string("_inversed_t_combined_y_0"), val = fp32(0x1p-1)];
tensor<fp32, [1, 512]> _inversed_t_combined = mul(x = var_93, y = _inversed_t_combined_y_0)[name = string("_inversed_t_combined")];
tensor<fp32, [1, 512]> c = linear(bias = flow_net_cond_embed_bias, weight = flow_net_cond_embed_weight, x = transformer_out)[name = string("linear_5")];
tensor<fp32, [1, 512]> input_13 = add(x = _inversed_t_combined, y = c)[name = string("input_13")];
tensor<fp32, [1, 512]> input_15 = silu(x = input_13)[name = string("input_15")];
tensor<fp32, [1, 1536]> var_107 = linear(bias = flow_net_res_blocks_0_adaLN_modulation_1_bias, weight = flow_net_res_blocks_0_adaLN_modulation_1_weight, x = input_15)[name = string("linear_6")];
tensor<int32, [3]> var_108_split_sizes_0 = const()[name = string("op_108_split_sizes_0"), val = tensor<int32, [3]>([512, 512, 512])];
int32 var_108_axis_0 = const()[name = string("op_108_axis_0"), val = int32(-1)];
tensor<fp32, [1, 512]> var_108_0, tensor<fp32, [1, 512]> var_108_1, tensor<fp32, [1, 512]> var_108_2 = split(axis = var_108_axis_0, split_sizes = var_108_split_sizes_0, x = var_107)[name = string("op_108")];
tensor<int32, [1]> mean_1_axes_0 = const()[name = string("mean_1_axes_0"), val = tensor<int32, [1]>([-1])];
bool mean_1_keep_dims_0 = const()[name = string("mean_1_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> mean_1 = reduce_mean(axes = mean_1_axes_0, keep_dims = mean_1_keep_dims_0, x = x_5)[name = string("mean_1")];
tensor<fp32, [1, 512]> sub_4 = sub(x = x_5, y = mean_1)[name = string("sub_4")];
tensor<fp32, [1, 512]> square_2 = square(x = sub_4)[name = string("square_2")];
tensor<int32, [1]> reduce_mean_5_axes_0 = const()[name = string("reduce_mean_5_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_5_keep_dims_0 = const()[name = string("reduce_mean_5_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_5 = reduce_mean(axes = reduce_mean_5_axes_0, keep_dims = reduce_mean_5_keep_dims_0, x = square_2)[name = string("reduce_mean_5")];
fp32 var_118 = const()[name = string("op_118"), val = fp32(0x1.0c6f7ap-20)];
tensor<fp32, [1, 1]> var_119 = add(x = reduce_mean_5, y = var_118)[name = string("op_119")];
tensor<fp32, [1, 1]> var_120 = sqrt(x = var_119)[name = string("op_120")];
tensor<fp32, [1, 512]> x_7 = real_div(x = sub_4, y = var_120)[name = string("x_7")];
tensor<fp32, [1, 512]> var_122 = mul(x = x_7, y = flow_net_res_blocks_0_in_ln_weight)[name = string("op_122")];
tensor<fp32, [1, 512]> x_9 = add(x = var_122, y = flow_net_res_blocks_0_in_ln_bias)[name = string("x_9")];
fp32 var_124_promoted = const()[name = string("op_124_promoted"), val = fp32(0x1p+0)];
tensor<fp32, [1, 512]> var_125 = add(x = var_108_1, y = var_124_promoted)[name = string("op_125")];
tensor<fp32, [1, 512]> var_126 = mul(x = x_9, y = var_125)[name = string("op_126")];
tensor<fp32, [1, 512]> input_17 = add(x = var_126, y = var_108_0)[name = string("input_17")];
tensor<fp32, [1, 512]> input_19 = linear(bias = flow_net_res_blocks_0_mlp_0_bias, weight = flow_net_res_blocks_0_mlp_0_weight, x = input_17)[name = string("linear_7")];
tensor<fp32, [1, 512]> input_21 = silu(x = input_19)[name = string("input_21")];
tensor<fp32, [1, 512]> h_1 = linear(bias = flow_net_res_blocks_0_mlp_2_bias, weight = flow_net_res_blocks_0_mlp_2_weight, x = input_21)[name = string("linear_8")];
tensor<fp32, [1, 512]> var_137 = mul(x = var_108_2, y = h_1)[name = string("op_137")];
tensor<fp32, [1, 512]> x_11 = add(x = x_5, y = var_137)[name = string("x_11")];
tensor<fp32, [1, 1536]> var_146 = linear(bias = flow_net_res_blocks_1_adaLN_modulation_1_bias, weight = flow_net_res_blocks_1_adaLN_modulation_1_weight, x = input_15)[name = string("linear_9")];
tensor<int32, [3]> var_147_split_sizes_0 = const()[name = string("op_147_split_sizes_0"), val = tensor<int32, [3]>([512, 512, 512])];
int32 var_147_axis_0 = const()[name = string("op_147_axis_0"), val = int32(-1)];
tensor<fp32, [1, 512]> var_147_0, tensor<fp32, [1, 512]> var_147_1, tensor<fp32, [1, 512]> var_147_2 = split(axis = var_147_axis_0, split_sizes = var_147_split_sizes_0, x = var_146)[name = string("op_147")];
tensor<int32, [1]> mean_3_axes_0 = const()[name = string("mean_3_axes_0"), val = tensor<int32, [1]>([-1])];
bool mean_3_keep_dims_0 = const()[name = string("mean_3_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> mean_3 = reduce_mean(axes = mean_3_axes_0, keep_dims = mean_3_keep_dims_0, x = x_11)[name = string("mean_3")];
tensor<fp32, [1, 512]> sub_5 = sub(x = x_11, y = mean_3)[name = string("sub_5")];
tensor<fp32, [1, 512]> square_3 = square(x = sub_5)[name = string("square_3")];
tensor<int32, [1]> reduce_mean_7_axes_0 = const()[name = string("reduce_mean_7_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_7_keep_dims_0 = const()[name = string("reduce_mean_7_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_7 = reduce_mean(axes = reduce_mean_7_axes_0, keep_dims = reduce_mean_7_keep_dims_0, x = square_3)[name = string("reduce_mean_7")];
fp32 var_157 = const()[name = string("op_157"), val = fp32(0x1.0c6f7ap-20)];
tensor<fp32, [1, 1]> var_158 = add(x = reduce_mean_7, y = var_157)[name = string("op_158")];
tensor<fp32, [1, 1]> var_159 = sqrt(x = var_158)[name = string("op_159")];
tensor<fp32, [1, 512]> x_13 = real_div(x = sub_5, y = var_159)[name = string("x_13")];
tensor<fp32, [1, 512]> var_161 = mul(x = x_13, y = flow_net_res_blocks_1_in_ln_weight)[name = string("op_161")];
tensor<fp32, [1, 512]> x_15 = add(x = var_161, y = flow_net_res_blocks_1_in_ln_bias)[name = string("x_15")];
fp32 var_163_promoted = const()[name = string("op_163_promoted"), val = fp32(0x1p+0)];
tensor<fp32, [1, 512]> var_164 = add(x = var_147_1, y = var_163_promoted)[name = string("op_164")];
tensor<fp32, [1, 512]> var_165 = mul(x = x_15, y = var_164)[name = string("op_165")];
tensor<fp32, [1, 512]> input_25 = add(x = var_165, y = var_147_0)[name = string("input_25")];
tensor<fp32, [1, 512]> input_27 = linear(bias = flow_net_res_blocks_1_mlp_0_bias, weight = flow_net_res_blocks_1_mlp_0_weight, x = input_25)[name = string("linear_10")];
tensor<fp32, [1, 512]> input_29 = silu(x = input_27)[name = string("input_29")];
tensor<fp32, [1, 512]> h_3 = linear(bias = flow_net_res_blocks_1_mlp_2_bias, weight = flow_net_res_blocks_1_mlp_2_weight, x = input_29)[name = string("linear_11")];
tensor<fp32, [1, 512]> var_176 = mul(x = var_147_2, y = h_3)[name = string("op_176")];
tensor<fp32, [1, 512]> x_17 = add(x = x_11, y = var_176)[name = string("x_17")];
tensor<fp32, [1, 1536]> var_185 = linear(bias = flow_net_res_blocks_2_adaLN_modulation_1_bias, weight = flow_net_res_blocks_2_adaLN_modulation_1_weight, x = input_15)[name = string("linear_12")];
tensor<int32, [3]> var_186_split_sizes_0 = const()[name = string("op_186_split_sizes_0"), val = tensor<int32, [3]>([512, 512, 512])];
int32 var_186_axis_0 = const()[name = string("op_186_axis_0"), val = int32(-1)];
tensor<fp32, [1, 512]> var_186_0, tensor<fp32, [1, 512]> var_186_1, tensor<fp32, [1, 512]> var_186_2 = split(axis = var_186_axis_0, split_sizes = var_186_split_sizes_0, x = var_185)[name = string("op_186")];
tensor<int32, [1]> mean_5_axes_0 = const()[name = string("mean_5_axes_0"), val = tensor<int32, [1]>([-1])];
bool mean_5_keep_dims_0 = const()[name = string("mean_5_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> mean_5 = reduce_mean(axes = mean_5_axes_0, keep_dims = mean_5_keep_dims_0, x = x_17)[name = string("mean_5")];
tensor<fp32, [1, 512]> sub_6 = sub(x = x_17, y = mean_5)[name = string("sub_6")];
tensor<fp32, [1, 512]> square_4 = square(x = sub_6)[name = string("square_4")];
tensor<int32, [1]> reduce_mean_9_axes_0 = const()[name = string("reduce_mean_9_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_9_keep_dims_0 = const()[name = string("reduce_mean_9_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_9 = reduce_mean(axes = reduce_mean_9_axes_0, keep_dims = reduce_mean_9_keep_dims_0, x = square_4)[name = string("reduce_mean_9")];
fp32 var_196 = const()[name = string("op_196"), val = fp32(0x1.0c6f7ap-20)];
tensor<fp32, [1, 1]> var_197 = add(x = reduce_mean_9, y = var_196)[name = string("op_197")];
tensor<fp32, [1, 1]> var_198 = sqrt(x = var_197)[name = string("op_198")];
tensor<fp32, [1, 512]> x_19 = real_div(x = sub_6, y = var_198)[name = string("x_19")];
tensor<fp32, [1, 512]> var_200 = mul(x = x_19, y = flow_net_res_blocks_2_in_ln_weight)[name = string("op_200")];
tensor<fp32, [1, 512]> x_21 = add(x = var_200, y = flow_net_res_blocks_2_in_ln_bias)[name = string("x_21")];
fp32 var_202_promoted = const()[name = string("op_202_promoted"), val = fp32(0x1p+0)];
tensor<fp32, [1, 512]> var_203 = add(x = var_186_1, y = var_202_promoted)[name = string("op_203")];
tensor<fp32, [1, 512]> var_204 = mul(x = x_21, y = var_203)[name = string("op_204")];
tensor<fp32, [1, 512]> input_33 = add(x = var_204, y = var_186_0)[name = string("input_33")];
tensor<fp32, [1, 512]> input_35 = linear(bias = flow_net_res_blocks_2_mlp_0_bias, weight = flow_net_res_blocks_2_mlp_0_weight, x = input_33)[name = string("linear_13")];
tensor<fp32, [1, 512]> input_37 = silu(x = input_35)[name = string("input_37")];
tensor<fp32, [1, 512]> h_5 = linear(bias = flow_net_res_blocks_2_mlp_2_bias, weight = flow_net_res_blocks_2_mlp_2_weight, x = input_37)[name = string("linear_14")];
tensor<fp32, [1, 512]> var_215 = mul(x = var_186_2, y = h_5)[name = string("op_215")];
tensor<fp32, [1, 512]> x_23 = add(x = x_17, y = var_215)[name = string("x_23")];
tensor<fp32, [1, 1536]> var_224 = linear(bias = flow_net_res_blocks_3_adaLN_modulation_1_bias, weight = flow_net_res_blocks_3_adaLN_modulation_1_weight, x = input_15)[name = string("linear_15")];
tensor<int32, [3]> var_225_split_sizes_0 = const()[name = string("op_225_split_sizes_0"), val = tensor<int32, [3]>([512, 512, 512])];
int32 var_225_axis_0 = const()[name = string("op_225_axis_0"), val = int32(-1)];
tensor<fp32, [1, 512]> var_225_0, tensor<fp32, [1, 512]> var_225_1, tensor<fp32, [1, 512]> var_225_2 = split(axis = var_225_axis_0, split_sizes = var_225_split_sizes_0, x = var_224)[name = string("op_225")];
tensor<int32, [1]> mean_7_axes_0 = const()[name = string("mean_7_axes_0"), val = tensor<int32, [1]>([-1])];
bool mean_7_keep_dims_0 = const()[name = string("mean_7_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> mean_7 = reduce_mean(axes = mean_7_axes_0, keep_dims = mean_7_keep_dims_0, x = x_23)[name = string("mean_7")];
tensor<fp32, [1, 512]> sub_7 = sub(x = x_23, y = mean_7)[name = string("sub_7")];
tensor<fp32, [1, 512]> square_5 = square(x = sub_7)[name = string("square_5")];
tensor<int32, [1]> reduce_mean_11_axes_0 = const()[name = string("reduce_mean_11_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_11_keep_dims_0 = const()[name = string("reduce_mean_11_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_11 = reduce_mean(axes = reduce_mean_11_axes_0, keep_dims = reduce_mean_11_keep_dims_0, x = square_5)[name = string("reduce_mean_11")];
fp32 var_235 = const()[name = string("op_235"), val = fp32(0x1.0c6f7ap-20)];
tensor<fp32, [1, 1]> var_236 = add(x = reduce_mean_11, y = var_235)[name = string("op_236")];
tensor<fp32, [1, 1]> var_237 = sqrt(x = var_236)[name = string("op_237")];
tensor<fp32, [1, 512]> x_25 = real_div(x = sub_7, y = var_237)[name = string("x_25")];
tensor<fp32, [1, 512]> var_239 = mul(x = x_25, y = flow_net_res_blocks_3_in_ln_weight)[name = string("op_239")];
tensor<fp32, [1, 512]> x_27 = add(x = var_239, y = flow_net_res_blocks_3_in_ln_bias)[name = string("x_27")];
fp32 var_241_promoted = const()[name = string("op_241_promoted"), val = fp32(0x1p+0)];
tensor<fp32, [1, 512]> var_242 = add(x = var_225_1, y = var_241_promoted)[name = string("op_242")];
tensor<fp32, [1, 512]> var_243 = mul(x = x_27, y = var_242)[name = string("op_243")];
tensor<fp32, [1, 512]> input_41 = add(x = var_243, y = var_225_0)[name = string("input_41")];
tensor<fp32, [1, 512]> input_43 = linear(bias = flow_net_res_blocks_3_mlp_0_bias, weight = flow_net_res_blocks_3_mlp_0_weight, x = input_41)[name = string("linear_16")];
tensor<fp32, [1, 512]> input_45 = silu(x = input_43)[name = string("input_45")];
tensor<fp32, [1, 512]> h_7 = linear(bias = flow_net_res_blocks_3_mlp_2_bias, weight = flow_net_res_blocks_3_mlp_2_weight, x = input_45)[name = string("linear_17")];
tensor<fp32, [1, 512]> var_254 = mul(x = var_225_2, y = h_7)[name = string("op_254")];
tensor<fp32, [1, 512]> x_29 = add(x = x_23, y = var_254)[name = string("x_29")];
tensor<fp32, [1, 1536]> var_263 = linear(bias = flow_net_res_blocks_4_adaLN_modulation_1_bias, weight = flow_net_res_blocks_4_adaLN_modulation_1_weight, x = input_15)[name = string("linear_18")];
tensor<int32, [3]> var_264_split_sizes_0 = const()[name = string("op_264_split_sizes_0"), val = tensor<int32, [3]>([512, 512, 512])];
int32 var_264_axis_0 = const()[name = string("op_264_axis_0"), val = int32(-1)];
tensor<fp32, [1, 512]> var_264_0, tensor<fp32, [1, 512]> var_264_1, tensor<fp32, [1, 512]> var_264_2 = split(axis = var_264_axis_0, split_sizes = var_264_split_sizes_0, x = var_263)[name = string("op_264")];
tensor<int32, [1]> mean_9_axes_0 = const()[name = string("mean_9_axes_0"), val = tensor<int32, [1]>([-1])];
bool mean_9_keep_dims_0 = const()[name = string("mean_9_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> mean_9 = reduce_mean(axes = mean_9_axes_0, keep_dims = mean_9_keep_dims_0, x = x_29)[name = string("mean_9")];
tensor<fp32, [1, 512]> sub_8 = sub(x = x_29, y = mean_9)[name = string("sub_8")];
tensor<fp32, [1, 512]> square_6 = square(x = sub_8)[name = string("square_6")];
tensor<int32, [1]> reduce_mean_13_axes_0 = const()[name = string("reduce_mean_13_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_13_keep_dims_0 = const()[name = string("reduce_mean_13_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_13 = reduce_mean(axes = reduce_mean_13_axes_0, keep_dims = reduce_mean_13_keep_dims_0, x = square_6)[name = string("reduce_mean_13")];
fp32 var_274 = const()[name = string("op_274"), val = fp32(0x1.0c6f7ap-20)];
tensor<fp32, [1, 1]> var_275 = add(x = reduce_mean_13, y = var_274)[name = string("op_275")];
tensor<fp32, [1, 1]> var_276 = sqrt(x = var_275)[name = string("op_276")];
tensor<fp32, [1, 512]> x_31 = real_div(x = sub_8, y = var_276)[name = string("x_31")];
tensor<fp32, [1, 512]> var_278 = mul(x = x_31, y = flow_net_res_blocks_4_in_ln_weight)[name = string("op_278")];
tensor<fp32, [1, 512]> x_33 = add(x = var_278, y = flow_net_res_blocks_4_in_ln_bias)[name = string("x_33")];
fp32 var_280_promoted = const()[name = string("op_280_promoted"), val = fp32(0x1p+0)];
tensor<fp32, [1, 512]> var_281 = add(x = var_264_1, y = var_280_promoted)[name = string("op_281")];
tensor<fp32, [1, 512]> var_282 = mul(x = x_33, y = var_281)[name = string("op_282")];
tensor<fp32, [1, 512]> input_49 = add(x = var_282, y = var_264_0)[name = string("input_49")];
tensor<fp32, [1, 512]> input_51 = linear(bias = flow_net_res_blocks_4_mlp_0_bias, weight = flow_net_res_blocks_4_mlp_0_weight, x = input_49)[name = string("linear_19")];
tensor<fp32, [1, 512]> input_53 = silu(x = input_51)[name = string("input_53")];
tensor<fp32, [1, 512]> h_9 = linear(bias = flow_net_res_blocks_4_mlp_2_bias, weight = flow_net_res_blocks_4_mlp_2_weight, x = input_53)[name = string("linear_20")];
tensor<fp32, [1, 512]> var_293 = mul(x = var_264_2, y = h_9)[name = string("op_293")];
tensor<fp32, [1, 512]> x_35 = add(x = x_29, y = var_293)[name = string("x_35")];
tensor<fp32, [1, 1536]> var_302 = linear(bias = flow_net_res_blocks_5_adaLN_modulation_1_bias, weight = flow_net_res_blocks_5_adaLN_modulation_1_weight, x = input_15)[name = string("linear_21")];
tensor<int32, [3]> var_303_split_sizes_0 = const()[name = string("op_303_split_sizes_0"), val = tensor<int32, [3]>([512, 512, 512])];
int32 var_303_axis_0 = const()[name = string("op_303_axis_0"), val = int32(-1)];
tensor<fp32, [1, 512]> var_303_0, tensor<fp32, [1, 512]> var_303_1, tensor<fp32, [1, 512]> var_303_2 = split(axis = var_303_axis_0, split_sizes = var_303_split_sizes_0, x = var_302)[name = string("op_303")];
tensor<int32, [1]> mean_11_axes_0 = const()[name = string("mean_11_axes_0"), val = tensor<int32, [1]>([-1])];
bool mean_11_keep_dims_0 = const()[name = string("mean_11_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> mean_11 = reduce_mean(axes = mean_11_axes_0, keep_dims = mean_11_keep_dims_0, x = x_35)[name = string("mean_11")];
tensor<fp32, [1, 512]> sub_9 = sub(x = x_35, y = mean_11)[name = string("sub_9")];
tensor<fp32, [1, 512]> square_7 = square(x = sub_9)[name = string("square_7")];
tensor<int32, [1]> reduce_mean_15_axes_0 = const()[name = string("reduce_mean_15_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_15_keep_dims_0 = const()[name = string("reduce_mean_15_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_15 = reduce_mean(axes = reduce_mean_15_axes_0, keep_dims = reduce_mean_15_keep_dims_0, x = square_7)[name = string("reduce_mean_15")];
fp32 var_313 = const()[name = string("op_313"), val = fp32(0x1.0c6f7ap-20)];
tensor<fp32, [1, 1]> var_314 = add(x = reduce_mean_15, y = var_313)[name = string("op_314")];
tensor<fp32, [1, 1]> var_315 = sqrt(x = var_314)[name = string("op_315")];
tensor<fp32, [1, 512]> x_37 = real_div(x = sub_9, y = var_315)[name = string("x_37")];
tensor<fp32, [1, 512]> var_317 = mul(x = x_37, y = flow_net_res_blocks_5_in_ln_weight)[name = string("op_317")];
tensor<fp32, [1, 512]> x_39 = add(x = var_317, y = flow_net_res_blocks_5_in_ln_bias)[name = string("x_39")];
fp32 var_319_promoted = const()[name = string("op_319_promoted"), val = fp32(0x1p+0)];
tensor<fp32, [1, 512]> var_320 = add(x = var_303_1, y = var_319_promoted)[name = string("op_320")];
tensor<fp32, [1, 512]> var_321 = mul(x = x_39, y = var_320)[name = string("op_321")];
tensor<fp32, [1, 512]> input_57 = add(x = var_321, y = var_303_0)[name = string("input_57")];
tensor<fp32, [1, 512]> input_59 = linear(bias = flow_net_res_blocks_5_mlp_0_bias, weight = flow_net_res_blocks_5_mlp_0_weight, x = input_57)[name = string("linear_22")];
tensor<fp32, [1, 512]> input_61 = silu(x = input_59)[name = string("input_61")];
tensor<fp32, [1, 512]> h = linear(bias = flow_net_res_blocks_5_mlp_2_bias, weight = flow_net_res_blocks_5_mlp_2_weight, x = input_61)[name = string("linear_23")];
tensor<fp32, [1, 512]> var_332 = mul(x = var_303_2, y = h)[name = string("op_332")];
tensor<fp32, [1, 512]> x_41 = add(x = x_35, y = var_332)[name = string("x_41")];
tensor<fp32, [1, 1024]> var_340 = linear(bias = flow_net_final_layer_adaLN_modulation_1_bias, weight = flow_net_final_layer_adaLN_modulation_1_weight, x = input_15)[name = string("linear_24")];
tensor<int32, [2]> var_341_split_sizes_0 = const()[name = string("op_341_split_sizes_0"), val = tensor<int32, [2]>([512, 512])];
int32 var_341_axis_0 = const()[name = string("op_341_axis_0"), val = int32(-1)];
tensor<fp32, [1, 512]> var_341_0, tensor<fp32, [1, 512]> var_341_1 = split(axis = var_341_axis_0, split_sizes = var_341_split_sizes_0, x = var_340)[name = string("op_341")];
tensor<int32, [1]> mean_axes_0 = const()[name = string("mean_axes_0"), val = tensor<int32, [1]>([-1])];
bool mean_keep_dims_0 = const()[name = string("mean_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> mean = reduce_mean(axes = mean_axes_0, keep_dims = mean_keep_dims_0, x = x_41)[name = string("mean")];
tensor<fp32, [1, 512]> sub_10 = sub(x = x_41, y = mean)[name = string("sub_10")];
tensor<fp32, [1, 512]> square_8 = square(x = sub_10)[name = string("square_8")];
tensor<int32, [1]> reduce_mean_17_axes_0 = const()[name = string("reduce_mean_17_axes_0"), val = tensor<int32, [1]>([-1])];
bool reduce_mean_17_keep_dims_0 = const()[name = string("reduce_mean_17_keep_dims_0"), val = bool(true)];
tensor<fp32, [1, 1]> reduce_mean_17 = reduce_mean(axes = reduce_mean_17_axes_0, keep_dims = reduce_mean_17_keep_dims_0, x = square_8)[name = string("reduce_mean_17")];
fp32 var_348 = const()[name = string("op_348"), val = fp32(0x1.0c6f7ap-20)];
tensor<fp32, [1, 1]> var_349 = add(x = reduce_mean_17, y = var_348)[name = string("op_349")];
tensor<fp32, [1, 1]> var_350 = sqrt(x = var_349)[name = string("op_350")];
tensor<fp32, [1, 512]> x = real_div(x = sub_10, y = var_350)[name = string("x")];
fp32 var_352_promoted = const()[name = string("op_352_promoted"), val = fp32(0x1p+0)];
tensor<fp32, [1, 512]> var_353 = add(x = var_341_1, y = var_352_promoted)[name = string("op_353")];
tensor<fp32, [1, 512]> var_354 = mul(x = x, y = var_353)[name = string("op_354")];
tensor<fp32, [1, 512]> input = add(x = var_354, y = var_341_0)[name = string("input")];
tensor<fp32, [1, 32]> var_358 = linear(bias = flow_net_final_layer_linear_bias, weight = flow_net_final_layer_linear_weight, x = input)[name = string("linear_25")];
} -> (var_358);
}