| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3500.14.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.2.2"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.3.0"}})] |
| { |
| func main<ios15>(tensor<fp32, [1, 80, 3000]> logmel_data) { |
| tensor<string, []> var_44_pad_type_0 = const()[name = tensor<string, []>("op_44_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [2]> var_44_pad_0 = const()[name = tensor<string, []>("op_44_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [1]> var_44_strides_0 = const()[name = tensor<string, []>("op_44_strides_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, [1]> var_44_dilations_0 = const()[name = tensor<string, []>("op_44_dilations_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, []> var_44_groups_0 = const()[name = tensor<string, []>("op_44_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<string, []> logmel_data_to_fp16_dtype_0 = const()[name = tensor<string, []>("logmel_data_to_fp16_dtype_0"), val = tensor<string, []>("fp16")]; |
| tensor<fp16, [768, 80, 3]> const_0_to_fp16 = const()[name = tensor<string, []>("const_0_to_fp16"), val = tensor<fp16, [768, 80, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| tensor<fp16, [768]> const_1_to_fp16 = const()[name = tensor<string, []>("const_1_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368768)))]; |
| tensor<fp16, [1, 80, 3000]> logmel_data_to_fp16 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data)[name = tensor<string, []>("cast_52")]; |
| tensor<fp16, [1, 768, 3000]> var_44_cast_fp16 = conv(bias = const_1_to_fp16, dilations = var_44_dilations_0, groups = var_44_groups_0, pad = var_44_pad_0, pad_type = var_44_pad_type_0, strides = var_44_strides_0, weight = const_0_to_fp16, x = logmel_data_to_fp16)[name = tensor<string, []>("op_44_cast_fp16")]; |
| tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 768, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_44_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")]; |
| tensor<string, []> var_62_pad_type_0 = const()[name = tensor<string, []>("op_62_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [2]> var_62_pad_0 = const()[name = tensor<string, []>("op_62_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [1]> var_62_strides_0 = const()[name = tensor<string, []>("op_62_strides_0"), val = tensor<int32, [1]>([2])]; |
| tensor<int32, [1]> var_62_dilations_0 = const()[name = tensor<string, []>("op_62_dilations_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, []> var_62_groups_0 = const()[name = tensor<string, []>("op_62_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 3]> const_2_to_fp16 = const()[name = tensor<string, []>("const_2_to_fp16"), val = tensor<fp16, [768, 768, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370368)))]; |
| tensor<fp16, [768]> const_3_to_fp16 = const()[name = tensor<string, []>("const_3_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3909376)))]; |
| tensor<fp16, [1, 768, 1500]> var_62_cast_fp16 = conv(bias = const_3_to_fp16, dilations = var_62_dilations_0, groups = var_62_groups_0, pad = var_62_pad_0, pad_type = var_62_pad_type_0, strides = var_62_strides_0, weight = const_2_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("op_62_cast_fp16")]; |
| tensor<string, []> x_3_mode_0 = const()[name = tensor<string, []>("x_3_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 768, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_62_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")]; |
| tensor<fp16, [768, 1500]> var_67_to_fp16 = const()[name = tensor<string, []>("op_67_to_fp16"), val = tensor<fp16, [768, 1500]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3910976)))]; |
| tensor<fp16, [1, 768, 1500]> var_69_cast_fp16 = add(x = x_3_cast_fp16, y = var_67_to_fp16)[name = tensor<string, []>("op_69_cast_fp16")]; |
| tensor<int32, [1]> inputs_1_axes_0 = const()[name = tensor<string, []>("inputs_1_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_1_cast_fp16 = expand_dims(axes = inputs_1_axes_0, x = var_69_cast_fp16)[name = tensor<string, []>("inputs_1_cast_fp16")]; |
| tensor<int32, []> var_84 = const()[name = tensor<string, []>("op_84"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_3_axes_0 = const()[name = tensor<string, []>("input_3_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_3_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_3_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6215040)))]; |
| tensor<fp16, [768]> input_3_beta_0_to_fp16 = const()[name = tensor<string, []>("input_3_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6216640)))]; |
| tensor<fp16, []> var_100_to_fp16 = const()[name = tensor<string, []>("op_100_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_3_cast_fp16 = layer_norm(axes = input_3_axes_0, beta = input_3_beta_0_to_fp16, epsilon = var_100_to_fp16, gamma = input_3_gamma_0_to_fp16, x = inputs_1_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")]; |
| tensor<string, []> q_1_pad_type_0 = const()[name = tensor<string, []>("q_1_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_1_strides_0 = const()[name = tensor<string, []>("q_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_1_pad_0 = const()[name = tensor<string, []>("q_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_1_dilations_0 = const()[name = tensor<string, []>("q_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_1_groups_0 = const()[name = tensor<string, []>("q_1_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_135_weight_0_to_fp16 = const()[name = tensor<string, []>("op_135_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6218240)))]; |
| tensor<fp16, [768]> var_135_bias_0_to_fp16 = const()[name = tensor<string, []>("op_135_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7397952)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_135_cast_fp16 = conv(bias = var_135_bias_0_to_fp16, dilations = q_1_dilations_0, groups = q_1_groups_0, pad = q_1_pad_0, pad_type = q_1_pad_type_0, strides = q_1_strides_0, weight = var_135_weight_0_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("op_135_cast_fp16")]; |
| tensor<string, []> k_1_pad_type_0 = const()[name = tensor<string, []>("k_1_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_1_strides_0 = const()[name = tensor<string, []>("k_1_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_1_pad_0 = const()[name = tensor<string, []>("k_1_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_1_dilations_0 = const()[name = tensor<string, []>("k_1_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_1_groups_0 = const()[name = tensor<string, []>("k_1_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_0_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7399552)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_1_cast_fp16 = conv(dilations = k_1_dilations_0, groups = k_1_groups_0, pad = k_1_pad_0, pad_type = k_1_pad_type_0, strides = k_1_strides_0, weight = blocks_0_attn_key_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("k_1_cast_fp16")]; |
| tensor<string, []> var_133_pad_type_0 = const()[name = tensor<string, []>("op_133_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_133_strides_0 = const()[name = tensor<string, []>("op_133_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_133_pad_0 = const()[name = tensor<string, []>("op_133_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_133_dilations_0 = const()[name = tensor<string, []>("op_133_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_133_groups_0 = const()[name = tensor<string, []>("op_133_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_0_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8579264)))]; |
| tensor<fp16, [768]> blocks_0_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9758976)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_133_cast_fp16 = conv(bias = blocks_0_attn_value_bias_to_fp16, dilations = var_133_dilations_0, groups = var_133_groups_0, pad = var_133_pad_0, pad_type = var_133_pad_type_0, strides = var_133_strides_0, weight = blocks_0_attn_value_weight_to_fp16, x = input_3_cast_fp16)[name = tensor<string, []>("op_133_cast_fp16")]; |
| tensor<int32, [12]> tile_0 = const()[name = tensor<string, []>("tile_0"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_136_axis_0 = const()[name = tensor<string, []>("op_136_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_136_cast_fp16_11 = split(axis = var_136_axis_0, split_sizes = tile_0, x = var_135_cast_fp16)[name = tensor<string, []>("op_136_cast_fp16")]; |
| tensor<int32, [4]> var_149_perm_0 = const()[name = tensor<string, []>("op_149_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_1 = const()[name = tensor<string, []>("tile_1"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_150_axis_0 = const()[name = tensor<string, []>("op_150_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_149_cast_fp16 = transpose(perm = var_149_perm_0, x = k_1_cast_fp16)[name = tensor<string, []>("transpose_12")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_150_cast_fp16_11 = split(axis = var_150_axis_0, split_sizes = tile_1, x = var_149_cast_fp16)[name = tensor<string, []>("op_150_cast_fp16")]; |
| tensor<int32, [12]> tile_2 = const()[name = tensor<string, []>("tile_2"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_163_axis_0 = const()[name = tensor<string, []>("op_163_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_163_cast_fp16_11 = split(axis = var_163_axis_0, split_sizes = tile_2, x = var_133_cast_fp16)[name = tensor<string, []>("op_163_cast_fp16")]; |
| tensor<string, []> aw_1_equation_0 = const()[name = tensor<string, []>("aw_1_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_1_cast_fp16 = einsum(equation = aw_1_equation_0, values = (var_150_cast_fp16_0, var_136_cast_fp16_0))[name = tensor<string, []>("aw_1_cast_fp16")]; |
| tensor<string, []> aw_3_equation_0 = const()[name = tensor<string, []>("aw_3_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_3_cast_fp16 = einsum(equation = aw_3_equation_0, values = (var_150_cast_fp16_1, var_136_cast_fp16_1))[name = tensor<string, []>("aw_3_cast_fp16")]; |
| tensor<string, []> aw_5_equation_0 = const()[name = tensor<string, []>("aw_5_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_5_cast_fp16 = einsum(equation = aw_5_equation_0, values = (var_150_cast_fp16_2, var_136_cast_fp16_2))[name = tensor<string, []>("aw_5_cast_fp16")]; |
| tensor<string, []> aw_7_equation_0 = const()[name = tensor<string, []>("aw_7_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_7_cast_fp16 = einsum(equation = aw_7_equation_0, values = (var_150_cast_fp16_3, var_136_cast_fp16_3))[name = tensor<string, []>("aw_7_cast_fp16")]; |
| tensor<string, []> aw_9_equation_0 = const()[name = tensor<string, []>("aw_9_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_9_cast_fp16 = einsum(equation = aw_9_equation_0, values = (var_150_cast_fp16_4, var_136_cast_fp16_4))[name = tensor<string, []>("aw_9_cast_fp16")]; |
| tensor<string, []> aw_11_equation_0 = const()[name = tensor<string, []>("aw_11_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_11_cast_fp16 = einsum(equation = aw_11_equation_0, values = (var_150_cast_fp16_5, var_136_cast_fp16_5))[name = tensor<string, []>("aw_11_cast_fp16")]; |
| tensor<string, []> aw_13_equation_0 = const()[name = tensor<string, []>("aw_13_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_13_cast_fp16 = einsum(equation = aw_13_equation_0, values = (var_150_cast_fp16_6, var_136_cast_fp16_6))[name = tensor<string, []>("aw_13_cast_fp16")]; |
| tensor<string, []> aw_15_equation_0 = const()[name = tensor<string, []>("aw_15_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_15_cast_fp16 = einsum(equation = aw_15_equation_0, values = (var_150_cast_fp16_7, var_136_cast_fp16_7))[name = tensor<string, []>("aw_15_cast_fp16")]; |
| tensor<string, []> aw_17_equation_0 = const()[name = tensor<string, []>("aw_17_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_17_cast_fp16 = einsum(equation = aw_17_equation_0, values = (var_150_cast_fp16_8, var_136_cast_fp16_8))[name = tensor<string, []>("aw_17_cast_fp16")]; |
| tensor<string, []> aw_19_equation_0 = const()[name = tensor<string, []>("aw_19_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_19_cast_fp16 = einsum(equation = aw_19_equation_0, values = (var_150_cast_fp16_9, var_136_cast_fp16_9))[name = tensor<string, []>("aw_19_cast_fp16")]; |
| tensor<string, []> aw_21_equation_0 = const()[name = tensor<string, []>("aw_21_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_21_cast_fp16 = einsum(equation = aw_21_equation_0, values = (var_150_cast_fp16_10, var_136_cast_fp16_10))[name = tensor<string, []>("aw_21_cast_fp16")]; |
| tensor<string, []> aw_23_equation_0 = const()[name = tensor<string, []>("aw_23_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_23_cast_fp16 = einsum(equation = aw_23_equation_0, values = (var_150_cast_fp16_11, var_136_cast_fp16_11))[name = tensor<string, []>("aw_23_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_200_cast_fp16 = softmax(axis = var_84, x = aw_1_cast_fp16)[name = tensor<string, []>("op_200_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_201_cast_fp16 = softmax(axis = var_84, x = aw_3_cast_fp16)[name = tensor<string, []>("op_201_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_202_cast_fp16 = softmax(axis = var_84, x = aw_5_cast_fp16)[name = tensor<string, []>("op_202_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_203_cast_fp16 = softmax(axis = var_84, x = aw_7_cast_fp16)[name = tensor<string, []>("op_203_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_204_cast_fp16 = softmax(axis = var_84, x = aw_9_cast_fp16)[name = tensor<string, []>("op_204_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_205_cast_fp16 = softmax(axis = var_84, x = aw_11_cast_fp16)[name = tensor<string, []>("op_205_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_206_cast_fp16 = softmax(axis = var_84, x = aw_13_cast_fp16)[name = tensor<string, []>("op_206_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_207_cast_fp16 = softmax(axis = var_84, x = aw_15_cast_fp16)[name = tensor<string, []>("op_207_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_208_cast_fp16 = softmax(axis = var_84, x = aw_17_cast_fp16)[name = tensor<string, []>("op_208_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_209_cast_fp16 = softmax(axis = var_84, x = aw_19_cast_fp16)[name = tensor<string, []>("op_209_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_210_cast_fp16 = softmax(axis = var_84, x = aw_21_cast_fp16)[name = tensor<string, []>("op_210_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_211_cast_fp16 = softmax(axis = var_84, x = aw_23_cast_fp16)[name = tensor<string, []>("op_211_cast_fp16")]; |
| tensor<string, []> var_213_equation_0 = const()[name = tensor<string, []>("op_213_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_213_cast_fp16 = einsum(equation = var_213_equation_0, values = (var_163_cast_fp16_0, var_200_cast_fp16))[name = tensor<string, []>("op_213_cast_fp16")]; |
| tensor<string, []> var_215_equation_0 = const()[name = tensor<string, []>("op_215_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_215_cast_fp16 = einsum(equation = var_215_equation_0, values = (var_163_cast_fp16_1, var_201_cast_fp16))[name = tensor<string, []>("op_215_cast_fp16")]; |
| tensor<string, []> var_217_equation_0 = const()[name = tensor<string, []>("op_217_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_217_cast_fp16 = einsum(equation = var_217_equation_0, values = (var_163_cast_fp16_2, var_202_cast_fp16))[name = tensor<string, []>("op_217_cast_fp16")]; |
| tensor<string, []> var_219_equation_0 = const()[name = tensor<string, []>("op_219_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_219_cast_fp16 = einsum(equation = var_219_equation_0, values = (var_163_cast_fp16_3, var_203_cast_fp16))[name = tensor<string, []>("op_219_cast_fp16")]; |
| tensor<string, []> var_221_equation_0 = const()[name = tensor<string, []>("op_221_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_221_cast_fp16 = einsum(equation = var_221_equation_0, values = (var_163_cast_fp16_4, var_204_cast_fp16))[name = tensor<string, []>("op_221_cast_fp16")]; |
| tensor<string, []> var_223_equation_0 = const()[name = tensor<string, []>("op_223_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_223_cast_fp16 = einsum(equation = var_223_equation_0, values = (var_163_cast_fp16_5, var_205_cast_fp16))[name = tensor<string, []>("op_223_cast_fp16")]; |
| tensor<string, []> var_225_equation_0 = const()[name = tensor<string, []>("op_225_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_225_cast_fp16 = einsum(equation = var_225_equation_0, values = (var_163_cast_fp16_6, var_206_cast_fp16))[name = tensor<string, []>("op_225_cast_fp16")]; |
| tensor<string, []> var_227_equation_0 = const()[name = tensor<string, []>("op_227_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_227_cast_fp16 = einsum(equation = var_227_equation_0, values = (var_163_cast_fp16_7, var_207_cast_fp16))[name = tensor<string, []>("op_227_cast_fp16")]; |
| tensor<string, []> var_229_equation_0 = const()[name = tensor<string, []>("op_229_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_229_cast_fp16 = einsum(equation = var_229_equation_0, values = (var_163_cast_fp16_8, var_208_cast_fp16))[name = tensor<string, []>("op_229_cast_fp16")]; |
| tensor<string, []> var_231_equation_0 = const()[name = tensor<string, []>("op_231_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_231_cast_fp16 = einsum(equation = var_231_equation_0, values = (var_163_cast_fp16_9, var_209_cast_fp16))[name = tensor<string, []>("op_231_cast_fp16")]; |
| tensor<string, []> var_233_equation_0 = const()[name = tensor<string, []>("op_233_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_233_cast_fp16 = einsum(equation = var_233_equation_0, values = (var_163_cast_fp16_10, var_210_cast_fp16))[name = tensor<string, []>("op_233_cast_fp16")]; |
| tensor<string, []> var_235_equation_0 = const()[name = tensor<string, []>("op_235_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_235_cast_fp16 = einsum(equation = var_235_equation_0, values = (var_163_cast_fp16_11, var_211_cast_fp16))[name = tensor<string, []>("op_235_cast_fp16")]; |
| tensor<bool, []> input_5_interleave_0 = const()[name = tensor<string, []>("input_5_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_5_cast_fp16 = concat(axis = var_84, interleave = input_5_interleave_0, values = (var_213_cast_fp16, var_215_cast_fp16, var_217_cast_fp16, var_219_cast_fp16, var_221_cast_fp16, var_223_cast_fp16, var_225_cast_fp16, var_227_cast_fp16, var_229_cast_fp16, var_231_cast_fp16, var_233_cast_fp16, var_235_cast_fp16))[name = tensor<string, []>("input_5_cast_fp16")]; |
| tensor<string, []> var_244_pad_type_0 = const()[name = tensor<string, []>("op_244_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_244_strides_0 = const()[name = tensor<string, []>("op_244_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_244_pad_0 = const()[name = tensor<string, []>("op_244_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_244_dilations_0 = const()[name = tensor<string, []>("op_244_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_244_groups_0 = const()[name = tensor<string, []>("op_244_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_0_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9760576)))]; |
| tensor<fp16, [768]> blocks_0_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10940288)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_244_cast_fp16 = conv(bias = blocks_0_attn_out_bias_to_fp16, dilations = var_244_dilations_0, groups = var_244_groups_0, pad = var_244_pad_0, pad_type = var_244_pad_type_0, strides = var_244_strides_0, weight = blocks_0_attn_out_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("op_244_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = var_244_cast_fp16)[name = tensor<string, []>("inputs_3_cast_fp16")]; |
| tensor<int32, [1]> input_7_axes_0 = const()[name = tensor<string, []>("input_7_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_7_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_7_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10941888)))]; |
| tensor<fp16, [768]> input_7_beta_0_to_fp16 = const()[name = tensor<string, []>("input_7_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10943488)))]; |
| tensor<fp16, []> var_254_to_fp16 = const()[name = tensor<string, []>("op_254_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_7_cast_fp16 = layer_norm(axes = input_7_axes_0, beta = input_7_beta_0_to_fp16, epsilon = var_254_to_fp16, gamma = input_7_gamma_0_to_fp16, x = inputs_3_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")]; |
| tensor<string, []> input_9_pad_type_0 = const()[name = tensor<string, []>("input_9_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_9_strides_0 = const()[name = tensor<string, []>("input_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_9_pad_0 = const()[name = tensor<string, []>("input_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_9_dilations_0 = const()[name = tensor<string, []>("input_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_9_groups_0 = const()[name = tensor<string, []>("input_9_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_0_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10945088)))]; |
| tensor<fp16, [3072]> blocks_0_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15663744)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_9_cast_fp16 = conv(bias = blocks_0_mlp_0_bias_to_fp16, dilations = input_9_dilations_0, groups = input_9_groups_0, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = input_9_strides_0, weight = blocks_0_mlp_0_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")]; |
| tensor<string, []> input_11_mode_0 = const()[name = tensor<string, []>("input_11_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_11_cast_fp16 = gelu(mode = input_11_mode_0, x = input_9_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")]; |
| tensor<string, []> var_280_pad_type_0 = const()[name = tensor<string, []>("op_280_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_280_strides_0 = const()[name = tensor<string, []>("op_280_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_280_pad_0 = const()[name = tensor<string, []>("op_280_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_280_dilations_0 = const()[name = tensor<string, []>("op_280_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_280_groups_0 = const()[name = tensor<string, []>("op_280_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_0_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15669952)))]; |
| tensor<fp16, [768]> blocks_0_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20388608)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_280_cast_fp16 = conv(bias = blocks_0_mlp_2_bias_to_fp16, dilations = var_280_dilations_0, groups = var_280_groups_0, pad = var_280_pad_0, pad_type = var_280_pad_type_0, strides = var_280_strides_0, weight = blocks_0_mlp_2_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("op_280_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = var_280_cast_fp16)[name = tensor<string, []>("inputs_5_cast_fp16")]; |
| tensor<int32, []> var_289 = const()[name = tensor<string, []>("op_289"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_13_axes_0 = const()[name = tensor<string, []>("input_13_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_13_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_13_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20390208)))]; |
| tensor<fp16, [768]> input_13_beta_0_to_fp16 = const()[name = tensor<string, []>("input_13_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20391808)))]; |
| tensor<fp16, []> var_305_to_fp16 = const()[name = tensor<string, []>("op_305_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_13_cast_fp16 = layer_norm(axes = input_13_axes_0, beta = input_13_beta_0_to_fp16, epsilon = var_305_to_fp16, gamma = input_13_gamma_0_to_fp16, x = inputs_5_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")]; |
| tensor<string, []> q_3_pad_type_0 = const()[name = tensor<string, []>("q_3_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_3_strides_0 = const()[name = tensor<string, []>("q_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_3_pad_0 = const()[name = tensor<string, []>("q_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_3_dilations_0 = const()[name = tensor<string, []>("q_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_3_groups_0 = const()[name = tensor<string, []>("q_3_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_340_weight_0_to_fp16 = const()[name = tensor<string, []>("op_340_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20393408)))]; |
| tensor<fp16, [768]> var_340_bias_0_to_fp16 = const()[name = tensor<string, []>("op_340_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21573120)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_340_cast_fp16 = conv(bias = var_340_bias_0_to_fp16, dilations = q_3_dilations_0, groups = q_3_groups_0, pad = q_3_pad_0, pad_type = q_3_pad_type_0, strides = q_3_strides_0, weight = var_340_weight_0_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("op_340_cast_fp16")]; |
| tensor<string, []> k_3_pad_type_0 = const()[name = tensor<string, []>("k_3_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_3_strides_0 = const()[name = tensor<string, []>("k_3_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_3_pad_0 = const()[name = tensor<string, []>("k_3_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_3_dilations_0 = const()[name = tensor<string, []>("k_3_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_3_groups_0 = const()[name = tensor<string, []>("k_3_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_1_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21574720)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_3_cast_fp16 = conv(dilations = k_3_dilations_0, groups = k_3_groups_0, pad = k_3_pad_0, pad_type = k_3_pad_type_0, strides = k_3_strides_0, weight = blocks_1_attn_key_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("k_3_cast_fp16")]; |
| tensor<string, []> var_338_pad_type_0 = const()[name = tensor<string, []>("op_338_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_338_strides_0 = const()[name = tensor<string, []>("op_338_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_338_pad_0 = const()[name = tensor<string, []>("op_338_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_338_dilations_0 = const()[name = tensor<string, []>("op_338_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_338_groups_0 = const()[name = tensor<string, []>("op_338_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_1_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22754432)))]; |
| tensor<fp16, [768]> blocks_1_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23934144)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_338_cast_fp16 = conv(bias = blocks_1_attn_value_bias_to_fp16, dilations = var_338_dilations_0, groups = var_338_groups_0, pad = var_338_pad_0, pad_type = var_338_pad_type_0, strides = var_338_strides_0, weight = blocks_1_attn_value_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("op_338_cast_fp16")]; |
| tensor<int32, [12]> tile_3 = const()[name = tensor<string, []>("tile_3"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_341_axis_0 = const()[name = tensor<string, []>("op_341_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_341_cast_fp16_11 = split(axis = var_341_axis_0, split_sizes = tile_3, x = var_340_cast_fp16)[name = tensor<string, []>("op_341_cast_fp16")]; |
| tensor<int32, [4]> var_354_perm_0 = const()[name = tensor<string, []>("op_354_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_4 = const()[name = tensor<string, []>("tile_4"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_355_axis_0 = const()[name = tensor<string, []>("op_355_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_354_cast_fp16 = transpose(perm = var_354_perm_0, x = k_3_cast_fp16)[name = tensor<string, []>("transpose_11")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_355_cast_fp16_11 = split(axis = var_355_axis_0, split_sizes = tile_4, x = var_354_cast_fp16)[name = tensor<string, []>("op_355_cast_fp16")]; |
| tensor<int32, [12]> tile_5 = const()[name = tensor<string, []>("tile_5"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_368_axis_0 = const()[name = tensor<string, []>("op_368_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_368_cast_fp16_11 = split(axis = var_368_axis_0, split_sizes = tile_5, x = var_338_cast_fp16)[name = tensor<string, []>("op_368_cast_fp16")]; |
| tensor<string, []> aw_25_equation_0 = const()[name = tensor<string, []>("aw_25_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_25_cast_fp16 = einsum(equation = aw_25_equation_0, values = (var_355_cast_fp16_0, var_341_cast_fp16_0))[name = tensor<string, []>("aw_25_cast_fp16")]; |
| tensor<string, []> aw_27_equation_0 = const()[name = tensor<string, []>("aw_27_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_27_cast_fp16 = einsum(equation = aw_27_equation_0, values = (var_355_cast_fp16_1, var_341_cast_fp16_1))[name = tensor<string, []>("aw_27_cast_fp16")]; |
| tensor<string, []> aw_29_equation_0 = const()[name = tensor<string, []>("aw_29_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_29_cast_fp16 = einsum(equation = aw_29_equation_0, values = (var_355_cast_fp16_2, var_341_cast_fp16_2))[name = tensor<string, []>("aw_29_cast_fp16")]; |
| tensor<string, []> aw_31_equation_0 = const()[name = tensor<string, []>("aw_31_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_31_cast_fp16 = einsum(equation = aw_31_equation_0, values = (var_355_cast_fp16_3, var_341_cast_fp16_3))[name = tensor<string, []>("aw_31_cast_fp16")]; |
| tensor<string, []> aw_33_equation_0 = const()[name = tensor<string, []>("aw_33_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_33_cast_fp16 = einsum(equation = aw_33_equation_0, values = (var_355_cast_fp16_4, var_341_cast_fp16_4))[name = tensor<string, []>("aw_33_cast_fp16")]; |
| tensor<string, []> aw_35_equation_0 = const()[name = tensor<string, []>("aw_35_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_35_cast_fp16 = einsum(equation = aw_35_equation_0, values = (var_355_cast_fp16_5, var_341_cast_fp16_5))[name = tensor<string, []>("aw_35_cast_fp16")]; |
| tensor<string, []> aw_37_equation_0 = const()[name = tensor<string, []>("aw_37_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_37_cast_fp16 = einsum(equation = aw_37_equation_0, values = (var_355_cast_fp16_6, var_341_cast_fp16_6))[name = tensor<string, []>("aw_37_cast_fp16")]; |
| tensor<string, []> aw_39_equation_0 = const()[name = tensor<string, []>("aw_39_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_39_cast_fp16 = einsum(equation = aw_39_equation_0, values = (var_355_cast_fp16_7, var_341_cast_fp16_7))[name = tensor<string, []>("aw_39_cast_fp16")]; |
| tensor<string, []> aw_41_equation_0 = const()[name = tensor<string, []>("aw_41_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_41_cast_fp16 = einsum(equation = aw_41_equation_0, values = (var_355_cast_fp16_8, var_341_cast_fp16_8))[name = tensor<string, []>("aw_41_cast_fp16")]; |
| tensor<string, []> aw_43_equation_0 = const()[name = tensor<string, []>("aw_43_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_43_cast_fp16 = einsum(equation = aw_43_equation_0, values = (var_355_cast_fp16_9, var_341_cast_fp16_9))[name = tensor<string, []>("aw_43_cast_fp16")]; |
| tensor<string, []> aw_45_equation_0 = const()[name = tensor<string, []>("aw_45_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_45_cast_fp16 = einsum(equation = aw_45_equation_0, values = (var_355_cast_fp16_10, var_341_cast_fp16_10))[name = tensor<string, []>("aw_45_cast_fp16")]; |
| tensor<string, []> aw_47_equation_0 = const()[name = tensor<string, []>("aw_47_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_47_cast_fp16 = einsum(equation = aw_47_equation_0, values = (var_355_cast_fp16_11, var_341_cast_fp16_11))[name = tensor<string, []>("aw_47_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_405_cast_fp16 = softmax(axis = var_289, x = aw_25_cast_fp16)[name = tensor<string, []>("op_405_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_406_cast_fp16 = softmax(axis = var_289, x = aw_27_cast_fp16)[name = tensor<string, []>("op_406_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_407_cast_fp16 = softmax(axis = var_289, x = aw_29_cast_fp16)[name = tensor<string, []>("op_407_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_408_cast_fp16 = softmax(axis = var_289, x = aw_31_cast_fp16)[name = tensor<string, []>("op_408_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_409_cast_fp16 = softmax(axis = var_289, x = aw_33_cast_fp16)[name = tensor<string, []>("op_409_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_410_cast_fp16 = softmax(axis = var_289, x = aw_35_cast_fp16)[name = tensor<string, []>("op_410_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_411_cast_fp16 = softmax(axis = var_289, x = aw_37_cast_fp16)[name = tensor<string, []>("op_411_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_412_cast_fp16 = softmax(axis = var_289, x = aw_39_cast_fp16)[name = tensor<string, []>("op_412_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_413_cast_fp16 = softmax(axis = var_289, x = aw_41_cast_fp16)[name = tensor<string, []>("op_413_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_414_cast_fp16 = softmax(axis = var_289, x = aw_43_cast_fp16)[name = tensor<string, []>("op_414_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_415_cast_fp16 = softmax(axis = var_289, x = aw_45_cast_fp16)[name = tensor<string, []>("op_415_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_416_cast_fp16 = softmax(axis = var_289, x = aw_47_cast_fp16)[name = tensor<string, []>("op_416_cast_fp16")]; |
| tensor<string, []> var_418_equation_0 = const()[name = tensor<string, []>("op_418_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_418_cast_fp16 = einsum(equation = var_418_equation_0, values = (var_368_cast_fp16_0, var_405_cast_fp16))[name = tensor<string, []>("op_418_cast_fp16")]; |
| tensor<string, []> var_420_equation_0 = const()[name = tensor<string, []>("op_420_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_420_cast_fp16 = einsum(equation = var_420_equation_0, values = (var_368_cast_fp16_1, var_406_cast_fp16))[name = tensor<string, []>("op_420_cast_fp16")]; |
| tensor<string, []> var_422_equation_0 = const()[name = tensor<string, []>("op_422_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_422_cast_fp16 = einsum(equation = var_422_equation_0, values = (var_368_cast_fp16_2, var_407_cast_fp16))[name = tensor<string, []>("op_422_cast_fp16")]; |
| tensor<string, []> var_424_equation_0 = const()[name = tensor<string, []>("op_424_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_424_cast_fp16 = einsum(equation = var_424_equation_0, values = (var_368_cast_fp16_3, var_408_cast_fp16))[name = tensor<string, []>("op_424_cast_fp16")]; |
| tensor<string, []> var_426_equation_0 = const()[name = tensor<string, []>("op_426_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_426_cast_fp16 = einsum(equation = var_426_equation_0, values = (var_368_cast_fp16_4, var_409_cast_fp16))[name = tensor<string, []>("op_426_cast_fp16")]; |
| tensor<string, []> var_428_equation_0 = const()[name = tensor<string, []>("op_428_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_428_cast_fp16 = einsum(equation = var_428_equation_0, values = (var_368_cast_fp16_5, var_410_cast_fp16))[name = tensor<string, []>("op_428_cast_fp16")]; |
| tensor<string, []> var_430_equation_0 = const()[name = tensor<string, []>("op_430_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_430_cast_fp16 = einsum(equation = var_430_equation_0, values = (var_368_cast_fp16_6, var_411_cast_fp16))[name = tensor<string, []>("op_430_cast_fp16")]; |
| tensor<string, []> var_432_equation_0 = const()[name = tensor<string, []>("op_432_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_432_cast_fp16 = einsum(equation = var_432_equation_0, values = (var_368_cast_fp16_7, var_412_cast_fp16))[name = tensor<string, []>("op_432_cast_fp16")]; |
| tensor<string, []> var_434_equation_0 = const()[name = tensor<string, []>("op_434_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_434_cast_fp16 = einsum(equation = var_434_equation_0, values = (var_368_cast_fp16_8, var_413_cast_fp16))[name = tensor<string, []>("op_434_cast_fp16")]; |
| tensor<string, []> var_436_equation_0 = const()[name = tensor<string, []>("op_436_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_436_cast_fp16 = einsum(equation = var_436_equation_0, values = (var_368_cast_fp16_9, var_414_cast_fp16))[name = tensor<string, []>("op_436_cast_fp16")]; |
| tensor<string, []> var_438_equation_0 = const()[name = tensor<string, []>("op_438_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_438_cast_fp16 = einsum(equation = var_438_equation_0, values = (var_368_cast_fp16_10, var_415_cast_fp16))[name = tensor<string, []>("op_438_cast_fp16")]; |
| tensor<string, []> var_440_equation_0 = const()[name = tensor<string, []>("op_440_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_440_cast_fp16 = einsum(equation = var_440_equation_0, values = (var_368_cast_fp16_11, var_416_cast_fp16))[name = tensor<string, []>("op_440_cast_fp16")]; |
| tensor<bool, []> input_15_interleave_0 = const()[name = tensor<string, []>("input_15_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_15_cast_fp16 = concat(axis = var_289, interleave = input_15_interleave_0, values = (var_418_cast_fp16, var_420_cast_fp16, var_422_cast_fp16, var_424_cast_fp16, var_426_cast_fp16, var_428_cast_fp16, var_430_cast_fp16, var_432_cast_fp16, var_434_cast_fp16, var_436_cast_fp16, var_438_cast_fp16, var_440_cast_fp16))[name = tensor<string, []>("input_15_cast_fp16")]; |
| tensor<string, []> var_449_pad_type_0 = const()[name = tensor<string, []>("op_449_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_449_strides_0 = const()[name = tensor<string, []>("op_449_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_449_pad_0 = const()[name = tensor<string, []>("op_449_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_449_dilations_0 = const()[name = tensor<string, []>("op_449_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_449_groups_0 = const()[name = tensor<string, []>("op_449_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_1_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23935744)))]; |
| tensor<fp16, [768]> blocks_1_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25115456)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_449_cast_fp16 = conv(bias = blocks_1_attn_out_bias_to_fp16, dilations = var_449_dilations_0, groups = var_449_groups_0, pad = var_449_pad_0, pad_type = var_449_pad_type_0, strides = var_449_strides_0, weight = blocks_1_attn_out_weight_to_fp16, x = input_15_cast_fp16)[name = tensor<string, []>("op_449_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = var_449_cast_fp16)[name = tensor<string, []>("inputs_7_cast_fp16")]; |
| tensor<int32, [1]> input_17_axes_0 = const()[name = tensor<string, []>("input_17_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_17_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_17_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25117056)))]; |
| tensor<fp16, [768]> input_17_beta_0_to_fp16 = const()[name = tensor<string, []>("input_17_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25118656)))]; |
| tensor<fp16, []> var_459_to_fp16 = const()[name = tensor<string, []>("op_459_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_17_cast_fp16 = layer_norm(axes = input_17_axes_0, beta = input_17_beta_0_to_fp16, epsilon = var_459_to_fp16, gamma = input_17_gamma_0_to_fp16, x = inputs_7_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")]; |
| tensor<string, []> input_19_pad_type_0 = const()[name = tensor<string, []>("input_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_19_strides_0 = const()[name = tensor<string, []>("input_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_19_pad_0 = const()[name = tensor<string, []>("input_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_19_dilations_0 = const()[name = tensor<string, []>("input_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_19_groups_0 = const()[name = tensor<string, []>("input_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_1_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25120256)))]; |
| tensor<fp16, [3072]> blocks_1_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29838912)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_19_cast_fp16 = conv(bias = blocks_1_mlp_0_bias_to_fp16, dilations = input_19_dilations_0, groups = input_19_groups_0, pad = input_19_pad_0, pad_type = input_19_pad_type_0, strides = input_19_strides_0, weight = blocks_1_mlp_0_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("input_19_cast_fp16")]; |
| tensor<string, []> input_21_mode_0 = const()[name = tensor<string, []>("input_21_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_21_cast_fp16 = gelu(mode = input_21_mode_0, x = input_19_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")]; |
| tensor<string, []> var_485_pad_type_0 = const()[name = tensor<string, []>("op_485_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_485_strides_0 = const()[name = tensor<string, []>("op_485_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_485_pad_0 = const()[name = tensor<string, []>("op_485_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_485_dilations_0 = const()[name = tensor<string, []>("op_485_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_485_groups_0 = const()[name = tensor<string, []>("op_485_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_1_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29845120)))]; |
| tensor<fp16, [768]> blocks_1_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34563776)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_485_cast_fp16 = conv(bias = blocks_1_mlp_2_bias_to_fp16, dilations = var_485_dilations_0, groups = var_485_groups_0, pad = var_485_pad_0, pad_type = var_485_pad_type_0, strides = var_485_strides_0, weight = blocks_1_mlp_2_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("op_485_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = var_485_cast_fp16)[name = tensor<string, []>("inputs_9_cast_fp16")]; |
| tensor<int32, []> var_494 = const()[name = tensor<string, []>("op_494"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_23_axes_0 = const()[name = tensor<string, []>("input_23_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_23_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_23_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34565376)))]; |
| tensor<fp16, [768]> input_23_beta_0_to_fp16 = const()[name = tensor<string, []>("input_23_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34566976)))]; |
| tensor<fp16, []> var_510_to_fp16 = const()[name = tensor<string, []>("op_510_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_23_cast_fp16 = layer_norm(axes = input_23_axes_0, beta = input_23_beta_0_to_fp16, epsilon = var_510_to_fp16, gamma = input_23_gamma_0_to_fp16, x = inputs_9_cast_fp16)[name = tensor<string, []>("input_23_cast_fp16")]; |
| tensor<string, []> q_5_pad_type_0 = const()[name = tensor<string, []>("q_5_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_5_strides_0 = const()[name = tensor<string, []>("q_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_5_pad_0 = const()[name = tensor<string, []>("q_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_5_dilations_0 = const()[name = tensor<string, []>("q_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_5_groups_0 = const()[name = tensor<string, []>("q_5_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_545_weight_0_to_fp16 = const()[name = tensor<string, []>("op_545_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34568576)))]; |
| tensor<fp16, [768]> var_545_bias_0_to_fp16 = const()[name = tensor<string, []>("op_545_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35748288)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_545_cast_fp16 = conv(bias = var_545_bias_0_to_fp16, dilations = q_5_dilations_0, groups = q_5_groups_0, pad = q_5_pad_0, pad_type = q_5_pad_type_0, strides = q_5_strides_0, weight = var_545_weight_0_to_fp16, x = input_23_cast_fp16)[name = tensor<string, []>("op_545_cast_fp16")]; |
| tensor<string, []> k_5_pad_type_0 = const()[name = tensor<string, []>("k_5_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_5_strides_0 = const()[name = tensor<string, []>("k_5_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_5_pad_0 = const()[name = tensor<string, []>("k_5_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_5_dilations_0 = const()[name = tensor<string, []>("k_5_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_5_groups_0 = const()[name = tensor<string, []>("k_5_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_2_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35749888)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_5_cast_fp16 = conv(dilations = k_5_dilations_0, groups = k_5_groups_0, pad = k_5_pad_0, pad_type = k_5_pad_type_0, strides = k_5_strides_0, weight = blocks_2_attn_key_weight_to_fp16, x = input_23_cast_fp16)[name = tensor<string, []>("k_5_cast_fp16")]; |
| tensor<string, []> var_543_pad_type_0 = const()[name = tensor<string, []>("op_543_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_543_strides_0 = const()[name = tensor<string, []>("op_543_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_543_pad_0 = const()[name = tensor<string, []>("op_543_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_543_dilations_0 = const()[name = tensor<string, []>("op_543_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_543_groups_0 = const()[name = tensor<string, []>("op_543_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_2_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36929600)))]; |
| tensor<fp16, [768]> blocks_2_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38109312)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_543_cast_fp16 = conv(bias = blocks_2_attn_value_bias_to_fp16, dilations = var_543_dilations_0, groups = var_543_groups_0, pad = var_543_pad_0, pad_type = var_543_pad_type_0, strides = var_543_strides_0, weight = blocks_2_attn_value_weight_to_fp16, x = input_23_cast_fp16)[name = tensor<string, []>("op_543_cast_fp16")]; |
| tensor<int32, [12]> tile_6 = const()[name = tensor<string, []>("tile_6"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_546_axis_0 = const()[name = tensor<string, []>("op_546_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_546_cast_fp16_11 = split(axis = var_546_axis_0, split_sizes = tile_6, x = var_545_cast_fp16)[name = tensor<string, []>("op_546_cast_fp16")]; |
| tensor<int32, [4]> var_559_perm_0 = const()[name = tensor<string, []>("op_559_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_7 = const()[name = tensor<string, []>("tile_7"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_560_axis_0 = const()[name = tensor<string, []>("op_560_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_559_cast_fp16 = transpose(perm = var_559_perm_0, x = k_5_cast_fp16)[name = tensor<string, []>("transpose_10")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_560_cast_fp16_11 = split(axis = var_560_axis_0, split_sizes = tile_7, x = var_559_cast_fp16)[name = tensor<string, []>("op_560_cast_fp16")]; |
| tensor<int32, [12]> tile_8 = const()[name = tensor<string, []>("tile_8"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_573_axis_0 = const()[name = tensor<string, []>("op_573_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_573_cast_fp16_11 = split(axis = var_573_axis_0, split_sizes = tile_8, x = var_543_cast_fp16)[name = tensor<string, []>("op_573_cast_fp16")]; |
| tensor<string, []> aw_49_equation_0 = const()[name = tensor<string, []>("aw_49_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_49_cast_fp16 = einsum(equation = aw_49_equation_0, values = (var_560_cast_fp16_0, var_546_cast_fp16_0))[name = tensor<string, []>("aw_49_cast_fp16")]; |
| tensor<string, []> aw_51_equation_0 = const()[name = tensor<string, []>("aw_51_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_51_cast_fp16 = einsum(equation = aw_51_equation_0, values = (var_560_cast_fp16_1, var_546_cast_fp16_1))[name = tensor<string, []>("aw_51_cast_fp16")]; |
| tensor<string, []> aw_53_equation_0 = const()[name = tensor<string, []>("aw_53_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_53_cast_fp16 = einsum(equation = aw_53_equation_0, values = (var_560_cast_fp16_2, var_546_cast_fp16_2))[name = tensor<string, []>("aw_53_cast_fp16")]; |
| tensor<string, []> aw_55_equation_0 = const()[name = tensor<string, []>("aw_55_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_55_cast_fp16 = einsum(equation = aw_55_equation_0, values = (var_560_cast_fp16_3, var_546_cast_fp16_3))[name = tensor<string, []>("aw_55_cast_fp16")]; |
| tensor<string, []> aw_57_equation_0 = const()[name = tensor<string, []>("aw_57_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_57_cast_fp16 = einsum(equation = aw_57_equation_0, values = (var_560_cast_fp16_4, var_546_cast_fp16_4))[name = tensor<string, []>("aw_57_cast_fp16")]; |
| tensor<string, []> aw_59_equation_0 = const()[name = tensor<string, []>("aw_59_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_59_cast_fp16 = einsum(equation = aw_59_equation_0, values = (var_560_cast_fp16_5, var_546_cast_fp16_5))[name = tensor<string, []>("aw_59_cast_fp16")]; |
| tensor<string, []> aw_61_equation_0 = const()[name = tensor<string, []>("aw_61_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_61_cast_fp16 = einsum(equation = aw_61_equation_0, values = (var_560_cast_fp16_6, var_546_cast_fp16_6))[name = tensor<string, []>("aw_61_cast_fp16")]; |
| tensor<string, []> aw_63_equation_0 = const()[name = tensor<string, []>("aw_63_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_63_cast_fp16 = einsum(equation = aw_63_equation_0, values = (var_560_cast_fp16_7, var_546_cast_fp16_7))[name = tensor<string, []>("aw_63_cast_fp16")]; |
| tensor<string, []> aw_65_equation_0 = const()[name = tensor<string, []>("aw_65_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_65_cast_fp16 = einsum(equation = aw_65_equation_0, values = (var_560_cast_fp16_8, var_546_cast_fp16_8))[name = tensor<string, []>("aw_65_cast_fp16")]; |
| tensor<string, []> aw_67_equation_0 = const()[name = tensor<string, []>("aw_67_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_67_cast_fp16 = einsum(equation = aw_67_equation_0, values = (var_560_cast_fp16_9, var_546_cast_fp16_9))[name = tensor<string, []>("aw_67_cast_fp16")]; |
| tensor<string, []> aw_69_equation_0 = const()[name = tensor<string, []>("aw_69_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_69_cast_fp16 = einsum(equation = aw_69_equation_0, values = (var_560_cast_fp16_10, var_546_cast_fp16_10))[name = tensor<string, []>("aw_69_cast_fp16")]; |
| tensor<string, []> aw_71_equation_0 = const()[name = tensor<string, []>("aw_71_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_71_cast_fp16 = einsum(equation = aw_71_equation_0, values = (var_560_cast_fp16_11, var_546_cast_fp16_11))[name = tensor<string, []>("aw_71_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_610_cast_fp16 = softmax(axis = var_494, x = aw_49_cast_fp16)[name = tensor<string, []>("op_610_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_611_cast_fp16 = softmax(axis = var_494, x = aw_51_cast_fp16)[name = tensor<string, []>("op_611_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_612_cast_fp16 = softmax(axis = var_494, x = aw_53_cast_fp16)[name = tensor<string, []>("op_612_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_613_cast_fp16 = softmax(axis = var_494, x = aw_55_cast_fp16)[name = tensor<string, []>("op_613_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_614_cast_fp16 = softmax(axis = var_494, x = aw_57_cast_fp16)[name = tensor<string, []>("op_614_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_615_cast_fp16 = softmax(axis = var_494, x = aw_59_cast_fp16)[name = tensor<string, []>("op_615_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_616_cast_fp16 = softmax(axis = var_494, x = aw_61_cast_fp16)[name = tensor<string, []>("op_616_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_617_cast_fp16 = softmax(axis = var_494, x = aw_63_cast_fp16)[name = tensor<string, []>("op_617_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_618_cast_fp16 = softmax(axis = var_494, x = aw_65_cast_fp16)[name = tensor<string, []>("op_618_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_619_cast_fp16 = softmax(axis = var_494, x = aw_67_cast_fp16)[name = tensor<string, []>("op_619_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_620_cast_fp16 = softmax(axis = var_494, x = aw_69_cast_fp16)[name = tensor<string, []>("op_620_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_621_cast_fp16 = softmax(axis = var_494, x = aw_71_cast_fp16)[name = tensor<string, []>("op_621_cast_fp16")]; |
| tensor<string, []> var_623_equation_0 = const()[name = tensor<string, []>("op_623_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_623_cast_fp16 = einsum(equation = var_623_equation_0, values = (var_573_cast_fp16_0, var_610_cast_fp16))[name = tensor<string, []>("op_623_cast_fp16")]; |
| tensor<string, []> var_625_equation_0 = const()[name = tensor<string, []>("op_625_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_625_cast_fp16 = einsum(equation = var_625_equation_0, values = (var_573_cast_fp16_1, var_611_cast_fp16))[name = tensor<string, []>("op_625_cast_fp16")]; |
| tensor<string, []> var_627_equation_0 = const()[name = tensor<string, []>("op_627_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_627_cast_fp16 = einsum(equation = var_627_equation_0, values = (var_573_cast_fp16_2, var_612_cast_fp16))[name = tensor<string, []>("op_627_cast_fp16")]; |
| tensor<string, []> var_629_equation_0 = const()[name = tensor<string, []>("op_629_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_629_cast_fp16 = einsum(equation = var_629_equation_0, values = (var_573_cast_fp16_3, var_613_cast_fp16))[name = tensor<string, []>("op_629_cast_fp16")]; |
| tensor<string, []> var_631_equation_0 = const()[name = tensor<string, []>("op_631_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_631_cast_fp16 = einsum(equation = var_631_equation_0, values = (var_573_cast_fp16_4, var_614_cast_fp16))[name = tensor<string, []>("op_631_cast_fp16")]; |
| tensor<string, []> var_633_equation_0 = const()[name = tensor<string, []>("op_633_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_633_cast_fp16 = einsum(equation = var_633_equation_0, values = (var_573_cast_fp16_5, var_615_cast_fp16))[name = tensor<string, []>("op_633_cast_fp16")]; |
| tensor<string, []> var_635_equation_0 = const()[name = tensor<string, []>("op_635_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_635_cast_fp16 = einsum(equation = var_635_equation_0, values = (var_573_cast_fp16_6, var_616_cast_fp16))[name = tensor<string, []>("op_635_cast_fp16")]; |
| tensor<string, []> var_637_equation_0 = const()[name = tensor<string, []>("op_637_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_637_cast_fp16 = einsum(equation = var_637_equation_0, values = (var_573_cast_fp16_7, var_617_cast_fp16))[name = tensor<string, []>("op_637_cast_fp16")]; |
| tensor<string, []> var_639_equation_0 = const()[name = tensor<string, []>("op_639_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_639_cast_fp16 = einsum(equation = var_639_equation_0, values = (var_573_cast_fp16_8, var_618_cast_fp16))[name = tensor<string, []>("op_639_cast_fp16")]; |
| tensor<string, []> var_641_equation_0 = const()[name = tensor<string, []>("op_641_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_641_cast_fp16 = einsum(equation = var_641_equation_0, values = (var_573_cast_fp16_9, var_619_cast_fp16))[name = tensor<string, []>("op_641_cast_fp16")]; |
| tensor<string, []> var_643_equation_0 = const()[name = tensor<string, []>("op_643_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_643_cast_fp16 = einsum(equation = var_643_equation_0, values = (var_573_cast_fp16_10, var_620_cast_fp16))[name = tensor<string, []>("op_643_cast_fp16")]; |
| tensor<string, []> var_645_equation_0 = const()[name = tensor<string, []>("op_645_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_645_cast_fp16 = einsum(equation = var_645_equation_0, values = (var_573_cast_fp16_11, var_621_cast_fp16))[name = tensor<string, []>("op_645_cast_fp16")]; |
| tensor<bool, []> input_25_interleave_0 = const()[name = tensor<string, []>("input_25_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_25_cast_fp16 = concat(axis = var_494, interleave = input_25_interleave_0, values = (var_623_cast_fp16, var_625_cast_fp16, var_627_cast_fp16, var_629_cast_fp16, var_631_cast_fp16, var_633_cast_fp16, var_635_cast_fp16, var_637_cast_fp16, var_639_cast_fp16, var_641_cast_fp16, var_643_cast_fp16, var_645_cast_fp16))[name = tensor<string, []>("input_25_cast_fp16")]; |
| tensor<string, []> var_654_pad_type_0 = const()[name = tensor<string, []>("op_654_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_654_strides_0 = const()[name = tensor<string, []>("op_654_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_654_pad_0 = const()[name = tensor<string, []>("op_654_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_654_dilations_0 = const()[name = tensor<string, []>("op_654_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_654_groups_0 = const()[name = tensor<string, []>("op_654_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_2_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38110912)))]; |
| tensor<fp16, [768]> blocks_2_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39290624)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_654_cast_fp16 = conv(bias = blocks_2_attn_out_bias_to_fp16, dilations = var_654_dilations_0, groups = var_654_groups_0, pad = var_654_pad_0, pad_type = var_654_pad_type_0, strides = var_654_strides_0, weight = blocks_2_attn_out_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("op_654_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = var_654_cast_fp16)[name = tensor<string, []>("inputs_11_cast_fp16")]; |
| tensor<int32, [1]> input_27_axes_0 = const()[name = tensor<string, []>("input_27_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_27_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_27_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39292224)))]; |
| tensor<fp16, [768]> input_27_beta_0_to_fp16 = const()[name = tensor<string, []>("input_27_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39293824)))]; |
| tensor<fp16, []> var_664_to_fp16 = const()[name = tensor<string, []>("op_664_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_27_cast_fp16 = layer_norm(axes = input_27_axes_0, beta = input_27_beta_0_to_fp16, epsilon = var_664_to_fp16, gamma = input_27_gamma_0_to_fp16, x = inputs_11_cast_fp16)[name = tensor<string, []>("input_27_cast_fp16")]; |
| tensor<string, []> input_29_pad_type_0 = const()[name = tensor<string, []>("input_29_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_29_strides_0 = const()[name = tensor<string, []>("input_29_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_29_pad_0 = const()[name = tensor<string, []>("input_29_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_29_dilations_0 = const()[name = tensor<string, []>("input_29_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_29_groups_0 = const()[name = tensor<string, []>("input_29_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_2_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39295424)))]; |
| tensor<fp16, [3072]> blocks_2_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44014080)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_29_cast_fp16 = conv(bias = blocks_2_mlp_0_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = blocks_2_mlp_0_weight_to_fp16, x = input_27_cast_fp16)[name = tensor<string, []>("input_29_cast_fp16")]; |
| tensor<string, []> input_31_mode_0 = const()[name = tensor<string, []>("input_31_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")]; |
| tensor<string, []> var_690_pad_type_0 = const()[name = tensor<string, []>("op_690_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_690_strides_0 = const()[name = tensor<string, []>("op_690_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_690_pad_0 = const()[name = tensor<string, []>("op_690_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_690_dilations_0 = const()[name = tensor<string, []>("op_690_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_690_groups_0 = const()[name = tensor<string, []>("op_690_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_2_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44020288)))]; |
| tensor<fp16, [768]> blocks_2_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48738944)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_690_cast_fp16 = conv(bias = blocks_2_mlp_2_bias_to_fp16, dilations = var_690_dilations_0, groups = var_690_groups_0, pad = var_690_pad_0, pad_type = var_690_pad_type_0, strides = var_690_strides_0, weight = blocks_2_mlp_2_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("op_690_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = var_690_cast_fp16)[name = tensor<string, []>("inputs_13_cast_fp16")]; |
| tensor<int32, []> var_699 = const()[name = tensor<string, []>("op_699"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_33_axes_0 = const()[name = tensor<string, []>("input_33_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_33_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_33_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48740544)))]; |
| tensor<fp16, [768]> input_33_beta_0_to_fp16 = const()[name = tensor<string, []>("input_33_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48742144)))]; |
| tensor<fp16, []> var_715_to_fp16 = const()[name = tensor<string, []>("op_715_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_33_cast_fp16 = layer_norm(axes = input_33_axes_0, beta = input_33_beta_0_to_fp16, epsilon = var_715_to_fp16, gamma = input_33_gamma_0_to_fp16, x = inputs_13_cast_fp16)[name = tensor<string, []>("input_33_cast_fp16")]; |
| tensor<string, []> q_7_pad_type_0 = const()[name = tensor<string, []>("q_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_7_strides_0 = const()[name = tensor<string, []>("q_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_7_pad_0 = const()[name = tensor<string, []>("q_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_7_dilations_0 = const()[name = tensor<string, []>("q_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_7_groups_0 = const()[name = tensor<string, []>("q_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_750_weight_0_to_fp16 = const()[name = tensor<string, []>("op_750_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48743744)))]; |
| tensor<fp16, [768]> var_750_bias_0_to_fp16 = const()[name = tensor<string, []>("op_750_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49923456)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_750_cast_fp16 = conv(bias = var_750_bias_0_to_fp16, dilations = q_7_dilations_0, groups = q_7_groups_0, pad = q_7_pad_0, pad_type = q_7_pad_type_0, strides = q_7_strides_0, weight = var_750_weight_0_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("op_750_cast_fp16")]; |
| tensor<string, []> k_7_pad_type_0 = const()[name = tensor<string, []>("k_7_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_7_strides_0 = const()[name = tensor<string, []>("k_7_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_7_pad_0 = const()[name = tensor<string, []>("k_7_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_7_dilations_0 = const()[name = tensor<string, []>("k_7_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_7_groups_0 = const()[name = tensor<string, []>("k_7_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_3_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49925056)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_7_cast_fp16 = conv(dilations = k_7_dilations_0, groups = k_7_groups_0, pad = k_7_pad_0, pad_type = k_7_pad_type_0, strides = k_7_strides_0, weight = blocks_3_attn_key_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("k_7_cast_fp16")]; |
| tensor<string, []> var_748_pad_type_0 = const()[name = tensor<string, []>("op_748_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_748_strides_0 = const()[name = tensor<string, []>("op_748_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_748_pad_0 = const()[name = tensor<string, []>("op_748_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_748_dilations_0 = const()[name = tensor<string, []>("op_748_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_748_groups_0 = const()[name = tensor<string, []>("op_748_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_3_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51104768)))]; |
| tensor<fp16, [768]> blocks_3_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52284480)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_748_cast_fp16 = conv(bias = blocks_3_attn_value_bias_to_fp16, dilations = var_748_dilations_0, groups = var_748_groups_0, pad = var_748_pad_0, pad_type = var_748_pad_type_0, strides = var_748_strides_0, weight = blocks_3_attn_value_weight_to_fp16, x = input_33_cast_fp16)[name = tensor<string, []>("op_748_cast_fp16")]; |
| tensor<int32, [12]> tile_9 = const()[name = tensor<string, []>("tile_9"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_751_axis_0 = const()[name = tensor<string, []>("op_751_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_751_cast_fp16_11 = split(axis = var_751_axis_0, split_sizes = tile_9, x = var_750_cast_fp16)[name = tensor<string, []>("op_751_cast_fp16")]; |
| tensor<int32, [4]> var_764_perm_0 = const()[name = tensor<string, []>("op_764_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_10 = const()[name = tensor<string, []>("tile_10"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_765_axis_0 = const()[name = tensor<string, []>("op_765_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_764_cast_fp16 = transpose(perm = var_764_perm_0, x = k_7_cast_fp16)[name = tensor<string, []>("transpose_9")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_765_cast_fp16_11 = split(axis = var_765_axis_0, split_sizes = tile_10, x = var_764_cast_fp16)[name = tensor<string, []>("op_765_cast_fp16")]; |
| tensor<int32, [12]> tile_11 = const()[name = tensor<string, []>("tile_11"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_778_axis_0 = const()[name = tensor<string, []>("op_778_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_778_cast_fp16_11 = split(axis = var_778_axis_0, split_sizes = tile_11, x = var_748_cast_fp16)[name = tensor<string, []>("op_778_cast_fp16")]; |
| tensor<string, []> aw_73_equation_0 = const()[name = tensor<string, []>("aw_73_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_73_cast_fp16 = einsum(equation = aw_73_equation_0, values = (var_765_cast_fp16_0, var_751_cast_fp16_0))[name = tensor<string, []>("aw_73_cast_fp16")]; |
| tensor<string, []> aw_75_equation_0 = const()[name = tensor<string, []>("aw_75_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_75_cast_fp16 = einsum(equation = aw_75_equation_0, values = (var_765_cast_fp16_1, var_751_cast_fp16_1))[name = tensor<string, []>("aw_75_cast_fp16")]; |
| tensor<string, []> aw_77_equation_0 = const()[name = tensor<string, []>("aw_77_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_77_cast_fp16 = einsum(equation = aw_77_equation_0, values = (var_765_cast_fp16_2, var_751_cast_fp16_2))[name = tensor<string, []>("aw_77_cast_fp16")]; |
| tensor<string, []> aw_79_equation_0 = const()[name = tensor<string, []>("aw_79_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_79_cast_fp16 = einsum(equation = aw_79_equation_0, values = (var_765_cast_fp16_3, var_751_cast_fp16_3))[name = tensor<string, []>("aw_79_cast_fp16")]; |
| tensor<string, []> aw_81_equation_0 = const()[name = tensor<string, []>("aw_81_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_81_cast_fp16 = einsum(equation = aw_81_equation_0, values = (var_765_cast_fp16_4, var_751_cast_fp16_4))[name = tensor<string, []>("aw_81_cast_fp16")]; |
| tensor<string, []> aw_83_equation_0 = const()[name = tensor<string, []>("aw_83_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_83_cast_fp16 = einsum(equation = aw_83_equation_0, values = (var_765_cast_fp16_5, var_751_cast_fp16_5))[name = tensor<string, []>("aw_83_cast_fp16")]; |
| tensor<string, []> aw_85_equation_0 = const()[name = tensor<string, []>("aw_85_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_85_cast_fp16 = einsum(equation = aw_85_equation_0, values = (var_765_cast_fp16_6, var_751_cast_fp16_6))[name = tensor<string, []>("aw_85_cast_fp16")]; |
| tensor<string, []> aw_87_equation_0 = const()[name = tensor<string, []>("aw_87_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_87_cast_fp16 = einsum(equation = aw_87_equation_0, values = (var_765_cast_fp16_7, var_751_cast_fp16_7))[name = tensor<string, []>("aw_87_cast_fp16")]; |
| tensor<string, []> aw_89_equation_0 = const()[name = tensor<string, []>("aw_89_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_89_cast_fp16 = einsum(equation = aw_89_equation_0, values = (var_765_cast_fp16_8, var_751_cast_fp16_8))[name = tensor<string, []>("aw_89_cast_fp16")]; |
| tensor<string, []> aw_91_equation_0 = const()[name = tensor<string, []>("aw_91_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_91_cast_fp16 = einsum(equation = aw_91_equation_0, values = (var_765_cast_fp16_9, var_751_cast_fp16_9))[name = tensor<string, []>("aw_91_cast_fp16")]; |
| tensor<string, []> aw_93_equation_0 = const()[name = tensor<string, []>("aw_93_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_93_cast_fp16 = einsum(equation = aw_93_equation_0, values = (var_765_cast_fp16_10, var_751_cast_fp16_10))[name = tensor<string, []>("aw_93_cast_fp16")]; |
| tensor<string, []> aw_95_equation_0 = const()[name = tensor<string, []>("aw_95_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_95_cast_fp16 = einsum(equation = aw_95_equation_0, values = (var_765_cast_fp16_11, var_751_cast_fp16_11))[name = tensor<string, []>("aw_95_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_815_cast_fp16 = softmax(axis = var_699, x = aw_73_cast_fp16)[name = tensor<string, []>("op_815_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_816_cast_fp16 = softmax(axis = var_699, x = aw_75_cast_fp16)[name = tensor<string, []>("op_816_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_817_cast_fp16 = softmax(axis = var_699, x = aw_77_cast_fp16)[name = tensor<string, []>("op_817_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_818_cast_fp16 = softmax(axis = var_699, x = aw_79_cast_fp16)[name = tensor<string, []>("op_818_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_819_cast_fp16 = softmax(axis = var_699, x = aw_81_cast_fp16)[name = tensor<string, []>("op_819_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_820_cast_fp16 = softmax(axis = var_699, x = aw_83_cast_fp16)[name = tensor<string, []>("op_820_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_821_cast_fp16 = softmax(axis = var_699, x = aw_85_cast_fp16)[name = tensor<string, []>("op_821_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_822_cast_fp16 = softmax(axis = var_699, x = aw_87_cast_fp16)[name = tensor<string, []>("op_822_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_823_cast_fp16 = softmax(axis = var_699, x = aw_89_cast_fp16)[name = tensor<string, []>("op_823_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_824_cast_fp16 = softmax(axis = var_699, x = aw_91_cast_fp16)[name = tensor<string, []>("op_824_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_825_cast_fp16 = softmax(axis = var_699, x = aw_93_cast_fp16)[name = tensor<string, []>("op_825_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_826_cast_fp16 = softmax(axis = var_699, x = aw_95_cast_fp16)[name = tensor<string, []>("op_826_cast_fp16")]; |
| tensor<string, []> var_828_equation_0 = const()[name = tensor<string, []>("op_828_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_828_cast_fp16 = einsum(equation = var_828_equation_0, values = (var_778_cast_fp16_0, var_815_cast_fp16))[name = tensor<string, []>("op_828_cast_fp16")]; |
| tensor<string, []> var_830_equation_0 = const()[name = tensor<string, []>("op_830_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_830_cast_fp16 = einsum(equation = var_830_equation_0, values = (var_778_cast_fp16_1, var_816_cast_fp16))[name = tensor<string, []>("op_830_cast_fp16")]; |
| tensor<string, []> var_832_equation_0 = const()[name = tensor<string, []>("op_832_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_832_cast_fp16 = einsum(equation = var_832_equation_0, values = (var_778_cast_fp16_2, var_817_cast_fp16))[name = tensor<string, []>("op_832_cast_fp16")]; |
| tensor<string, []> var_834_equation_0 = const()[name = tensor<string, []>("op_834_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_834_cast_fp16 = einsum(equation = var_834_equation_0, values = (var_778_cast_fp16_3, var_818_cast_fp16))[name = tensor<string, []>("op_834_cast_fp16")]; |
| tensor<string, []> var_836_equation_0 = const()[name = tensor<string, []>("op_836_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_836_cast_fp16 = einsum(equation = var_836_equation_0, values = (var_778_cast_fp16_4, var_819_cast_fp16))[name = tensor<string, []>("op_836_cast_fp16")]; |
| tensor<string, []> var_838_equation_0 = const()[name = tensor<string, []>("op_838_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_838_cast_fp16 = einsum(equation = var_838_equation_0, values = (var_778_cast_fp16_5, var_820_cast_fp16))[name = tensor<string, []>("op_838_cast_fp16")]; |
| tensor<string, []> var_840_equation_0 = const()[name = tensor<string, []>("op_840_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_840_cast_fp16 = einsum(equation = var_840_equation_0, values = (var_778_cast_fp16_6, var_821_cast_fp16))[name = tensor<string, []>("op_840_cast_fp16")]; |
| tensor<string, []> var_842_equation_0 = const()[name = tensor<string, []>("op_842_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_842_cast_fp16 = einsum(equation = var_842_equation_0, values = (var_778_cast_fp16_7, var_822_cast_fp16))[name = tensor<string, []>("op_842_cast_fp16")]; |
| tensor<string, []> var_844_equation_0 = const()[name = tensor<string, []>("op_844_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_844_cast_fp16 = einsum(equation = var_844_equation_0, values = (var_778_cast_fp16_8, var_823_cast_fp16))[name = tensor<string, []>("op_844_cast_fp16")]; |
| tensor<string, []> var_846_equation_0 = const()[name = tensor<string, []>("op_846_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_846_cast_fp16 = einsum(equation = var_846_equation_0, values = (var_778_cast_fp16_9, var_824_cast_fp16))[name = tensor<string, []>("op_846_cast_fp16")]; |
| tensor<string, []> var_848_equation_0 = const()[name = tensor<string, []>("op_848_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_848_cast_fp16 = einsum(equation = var_848_equation_0, values = (var_778_cast_fp16_10, var_825_cast_fp16))[name = tensor<string, []>("op_848_cast_fp16")]; |
| tensor<string, []> var_850_equation_0 = const()[name = tensor<string, []>("op_850_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_850_cast_fp16 = einsum(equation = var_850_equation_0, values = (var_778_cast_fp16_11, var_826_cast_fp16))[name = tensor<string, []>("op_850_cast_fp16")]; |
| tensor<bool, []> input_35_interleave_0 = const()[name = tensor<string, []>("input_35_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_35_cast_fp16 = concat(axis = var_699, interleave = input_35_interleave_0, values = (var_828_cast_fp16, var_830_cast_fp16, var_832_cast_fp16, var_834_cast_fp16, var_836_cast_fp16, var_838_cast_fp16, var_840_cast_fp16, var_842_cast_fp16, var_844_cast_fp16, var_846_cast_fp16, var_848_cast_fp16, var_850_cast_fp16))[name = tensor<string, []>("input_35_cast_fp16")]; |
| tensor<string, []> var_859_pad_type_0 = const()[name = tensor<string, []>("op_859_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_859_strides_0 = const()[name = tensor<string, []>("op_859_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_859_pad_0 = const()[name = tensor<string, []>("op_859_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_859_dilations_0 = const()[name = tensor<string, []>("op_859_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_859_groups_0 = const()[name = tensor<string, []>("op_859_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_3_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52286080)))]; |
| tensor<fp16, [768]> blocks_3_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53465792)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_859_cast_fp16 = conv(bias = blocks_3_attn_out_bias_to_fp16, dilations = var_859_dilations_0, groups = var_859_groups_0, pad = var_859_pad_0, pad_type = var_859_pad_type_0, strides = var_859_strides_0, weight = blocks_3_attn_out_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("op_859_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = var_859_cast_fp16)[name = tensor<string, []>("inputs_15_cast_fp16")]; |
| tensor<int32, [1]> input_37_axes_0 = const()[name = tensor<string, []>("input_37_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_37_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_37_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53467392)))]; |
| tensor<fp16, [768]> input_37_beta_0_to_fp16 = const()[name = tensor<string, []>("input_37_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53468992)))]; |
| tensor<fp16, []> var_869_to_fp16 = const()[name = tensor<string, []>("op_869_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_37_cast_fp16 = layer_norm(axes = input_37_axes_0, beta = input_37_beta_0_to_fp16, epsilon = var_869_to_fp16, gamma = input_37_gamma_0_to_fp16, x = inputs_15_cast_fp16)[name = tensor<string, []>("input_37_cast_fp16")]; |
| tensor<string, []> input_39_pad_type_0 = const()[name = tensor<string, []>("input_39_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_39_strides_0 = const()[name = tensor<string, []>("input_39_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_39_pad_0 = const()[name = tensor<string, []>("input_39_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_39_dilations_0 = const()[name = tensor<string, []>("input_39_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_39_groups_0 = const()[name = tensor<string, []>("input_39_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_3_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53470592)))]; |
| tensor<fp16, [3072]> blocks_3_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58189248)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_39_cast_fp16 = conv(bias = blocks_3_mlp_0_bias_to_fp16, dilations = input_39_dilations_0, groups = input_39_groups_0, pad = input_39_pad_0, pad_type = input_39_pad_type_0, strides = input_39_strides_0, weight = blocks_3_mlp_0_weight_to_fp16, x = input_37_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")]; |
| tensor<string, []> input_41_mode_0 = const()[name = tensor<string, []>("input_41_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_41_cast_fp16 = gelu(mode = input_41_mode_0, x = input_39_cast_fp16)[name = tensor<string, []>("input_41_cast_fp16")]; |
| tensor<string, []> var_895_pad_type_0 = const()[name = tensor<string, []>("op_895_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_895_strides_0 = const()[name = tensor<string, []>("op_895_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_895_pad_0 = const()[name = tensor<string, []>("op_895_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_895_dilations_0 = const()[name = tensor<string, []>("op_895_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_895_groups_0 = const()[name = tensor<string, []>("op_895_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_3_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58195456)))]; |
| tensor<fp16, [768]> blocks_3_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62914112)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_895_cast_fp16 = conv(bias = blocks_3_mlp_2_bias_to_fp16, dilations = var_895_dilations_0, groups = var_895_groups_0, pad = var_895_pad_0, pad_type = var_895_pad_type_0, strides = var_895_strides_0, weight = blocks_3_mlp_2_weight_to_fp16, x = input_41_cast_fp16)[name = tensor<string, []>("op_895_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = var_895_cast_fp16)[name = tensor<string, []>("inputs_17_cast_fp16")]; |
| tensor<int32, []> var_904 = const()[name = tensor<string, []>("op_904"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_43_axes_0 = const()[name = tensor<string, []>("input_43_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_43_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_43_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62915712)))]; |
| tensor<fp16, [768]> input_43_beta_0_to_fp16 = const()[name = tensor<string, []>("input_43_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62917312)))]; |
| tensor<fp16, []> var_920_to_fp16 = const()[name = tensor<string, []>("op_920_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_43_cast_fp16 = layer_norm(axes = input_43_axes_0, beta = input_43_beta_0_to_fp16, epsilon = var_920_to_fp16, gamma = input_43_gamma_0_to_fp16, x = inputs_17_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")]; |
| tensor<string, []> q_9_pad_type_0 = const()[name = tensor<string, []>("q_9_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_9_strides_0 = const()[name = tensor<string, []>("q_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_9_pad_0 = const()[name = tensor<string, []>("q_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_9_dilations_0 = const()[name = tensor<string, []>("q_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_9_groups_0 = const()[name = tensor<string, []>("q_9_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_955_weight_0_to_fp16 = const()[name = tensor<string, []>("op_955_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62918912)))]; |
| tensor<fp16, [768]> var_955_bias_0_to_fp16 = const()[name = tensor<string, []>("op_955_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64098624)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_955_cast_fp16 = conv(bias = var_955_bias_0_to_fp16, dilations = q_9_dilations_0, groups = q_9_groups_0, pad = q_9_pad_0, pad_type = q_9_pad_type_0, strides = q_9_strides_0, weight = var_955_weight_0_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("op_955_cast_fp16")]; |
| tensor<string, []> k_9_pad_type_0 = const()[name = tensor<string, []>("k_9_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_9_strides_0 = const()[name = tensor<string, []>("k_9_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_9_pad_0 = const()[name = tensor<string, []>("k_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_9_dilations_0 = const()[name = tensor<string, []>("k_9_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_9_groups_0 = const()[name = tensor<string, []>("k_9_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_4_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64100224)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_9_cast_fp16 = conv(dilations = k_9_dilations_0, groups = k_9_groups_0, pad = k_9_pad_0, pad_type = k_9_pad_type_0, strides = k_9_strides_0, weight = blocks_4_attn_key_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("k_9_cast_fp16")]; |
| tensor<string, []> var_953_pad_type_0 = const()[name = tensor<string, []>("op_953_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_953_strides_0 = const()[name = tensor<string, []>("op_953_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_953_pad_0 = const()[name = tensor<string, []>("op_953_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_953_dilations_0 = const()[name = tensor<string, []>("op_953_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_953_groups_0 = const()[name = tensor<string, []>("op_953_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_4_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65279936)))]; |
| tensor<fp16, [768]> blocks_4_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66459648)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_953_cast_fp16 = conv(bias = blocks_4_attn_value_bias_to_fp16, dilations = var_953_dilations_0, groups = var_953_groups_0, pad = var_953_pad_0, pad_type = var_953_pad_type_0, strides = var_953_strides_0, weight = blocks_4_attn_value_weight_to_fp16, x = input_43_cast_fp16)[name = tensor<string, []>("op_953_cast_fp16")]; |
| tensor<int32, [12]> tile_12 = const()[name = tensor<string, []>("tile_12"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_956_axis_0 = const()[name = tensor<string, []>("op_956_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_956_cast_fp16_11 = split(axis = var_956_axis_0, split_sizes = tile_12, x = var_955_cast_fp16)[name = tensor<string, []>("op_956_cast_fp16")]; |
| tensor<int32, [4]> var_969_perm_0 = const()[name = tensor<string, []>("op_969_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_13 = const()[name = tensor<string, []>("tile_13"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_970_axis_0 = const()[name = tensor<string, []>("op_970_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_969_cast_fp16 = transpose(perm = var_969_perm_0, x = k_9_cast_fp16)[name = tensor<string, []>("transpose_8")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_970_cast_fp16_11 = split(axis = var_970_axis_0, split_sizes = tile_13, x = var_969_cast_fp16)[name = tensor<string, []>("op_970_cast_fp16")]; |
| tensor<int32, [12]> tile_14 = const()[name = tensor<string, []>("tile_14"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_983_axis_0 = const()[name = tensor<string, []>("op_983_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_983_cast_fp16_11 = split(axis = var_983_axis_0, split_sizes = tile_14, x = var_953_cast_fp16)[name = tensor<string, []>("op_983_cast_fp16")]; |
| tensor<string, []> aw_97_equation_0 = const()[name = tensor<string, []>("aw_97_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_97_cast_fp16 = einsum(equation = aw_97_equation_0, values = (var_970_cast_fp16_0, var_956_cast_fp16_0))[name = tensor<string, []>("aw_97_cast_fp16")]; |
| tensor<string, []> aw_99_equation_0 = const()[name = tensor<string, []>("aw_99_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_99_cast_fp16 = einsum(equation = aw_99_equation_0, values = (var_970_cast_fp16_1, var_956_cast_fp16_1))[name = tensor<string, []>("aw_99_cast_fp16")]; |
| tensor<string, []> aw_101_equation_0 = const()[name = tensor<string, []>("aw_101_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_101_cast_fp16 = einsum(equation = aw_101_equation_0, values = (var_970_cast_fp16_2, var_956_cast_fp16_2))[name = tensor<string, []>("aw_101_cast_fp16")]; |
| tensor<string, []> aw_103_equation_0 = const()[name = tensor<string, []>("aw_103_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_103_cast_fp16 = einsum(equation = aw_103_equation_0, values = (var_970_cast_fp16_3, var_956_cast_fp16_3))[name = tensor<string, []>("aw_103_cast_fp16")]; |
| tensor<string, []> aw_105_equation_0 = const()[name = tensor<string, []>("aw_105_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_105_cast_fp16 = einsum(equation = aw_105_equation_0, values = (var_970_cast_fp16_4, var_956_cast_fp16_4))[name = tensor<string, []>("aw_105_cast_fp16")]; |
| tensor<string, []> aw_107_equation_0 = const()[name = tensor<string, []>("aw_107_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_107_cast_fp16 = einsum(equation = aw_107_equation_0, values = (var_970_cast_fp16_5, var_956_cast_fp16_5))[name = tensor<string, []>("aw_107_cast_fp16")]; |
| tensor<string, []> aw_109_equation_0 = const()[name = tensor<string, []>("aw_109_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_109_cast_fp16 = einsum(equation = aw_109_equation_0, values = (var_970_cast_fp16_6, var_956_cast_fp16_6))[name = tensor<string, []>("aw_109_cast_fp16")]; |
| tensor<string, []> aw_111_equation_0 = const()[name = tensor<string, []>("aw_111_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_111_cast_fp16 = einsum(equation = aw_111_equation_0, values = (var_970_cast_fp16_7, var_956_cast_fp16_7))[name = tensor<string, []>("aw_111_cast_fp16")]; |
| tensor<string, []> aw_113_equation_0 = const()[name = tensor<string, []>("aw_113_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_113_cast_fp16 = einsum(equation = aw_113_equation_0, values = (var_970_cast_fp16_8, var_956_cast_fp16_8))[name = tensor<string, []>("aw_113_cast_fp16")]; |
| tensor<string, []> aw_115_equation_0 = const()[name = tensor<string, []>("aw_115_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_115_cast_fp16 = einsum(equation = aw_115_equation_0, values = (var_970_cast_fp16_9, var_956_cast_fp16_9))[name = tensor<string, []>("aw_115_cast_fp16")]; |
| tensor<string, []> aw_117_equation_0 = const()[name = tensor<string, []>("aw_117_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_117_cast_fp16 = einsum(equation = aw_117_equation_0, values = (var_970_cast_fp16_10, var_956_cast_fp16_10))[name = tensor<string, []>("aw_117_cast_fp16")]; |
| tensor<string, []> aw_119_equation_0 = const()[name = tensor<string, []>("aw_119_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_119_cast_fp16 = einsum(equation = aw_119_equation_0, values = (var_970_cast_fp16_11, var_956_cast_fp16_11))[name = tensor<string, []>("aw_119_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1020_cast_fp16 = softmax(axis = var_904, x = aw_97_cast_fp16)[name = tensor<string, []>("op_1020_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1021_cast_fp16 = softmax(axis = var_904, x = aw_99_cast_fp16)[name = tensor<string, []>("op_1021_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1022_cast_fp16 = softmax(axis = var_904, x = aw_101_cast_fp16)[name = tensor<string, []>("op_1022_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1023_cast_fp16 = softmax(axis = var_904, x = aw_103_cast_fp16)[name = tensor<string, []>("op_1023_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1024_cast_fp16 = softmax(axis = var_904, x = aw_105_cast_fp16)[name = tensor<string, []>("op_1024_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1025_cast_fp16 = softmax(axis = var_904, x = aw_107_cast_fp16)[name = tensor<string, []>("op_1025_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1026_cast_fp16 = softmax(axis = var_904, x = aw_109_cast_fp16)[name = tensor<string, []>("op_1026_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1027_cast_fp16 = softmax(axis = var_904, x = aw_111_cast_fp16)[name = tensor<string, []>("op_1027_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1028_cast_fp16 = softmax(axis = var_904, x = aw_113_cast_fp16)[name = tensor<string, []>("op_1028_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1029_cast_fp16 = softmax(axis = var_904, x = aw_115_cast_fp16)[name = tensor<string, []>("op_1029_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1030_cast_fp16 = softmax(axis = var_904, x = aw_117_cast_fp16)[name = tensor<string, []>("op_1030_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1031_cast_fp16 = softmax(axis = var_904, x = aw_119_cast_fp16)[name = tensor<string, []>("op_1031_cast_fp16")]; |
| tensor<string, []> var_1033_equation_0 = const()[name = tensor<string, []>("op_1033_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1033_cast_fp16 = einsum(equation = var_1033_equation_0, values = (var_983_cast_fp16_0, var_1020_cast_fp16))[name = tensor<string, []>("op_1033_cast_fp16")]; |
| tensor<string, []> var_1035_equation_0 = const()[name = tensor<string, []>("op_1035_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1035_cast_fp16 = einsum(equation = var_1035_equation_0, values = (var_983_cast_fp16_1, var_1021_cast_fp16))[name = tensor<string, []>("op_1035_cast_fp16")]; |
| tensor<string, []> var_1037_equation_0 = const()[name = tensor<string, []>("op_1037_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1037_cast_fp16 = einsum(equation = var_1037_equation_0, values = (var_983_cast_fp16_2, var_1022_cast_fp16))[name = tensor<string, []>("op_1037_cast_fp16")]; |
| tensor<string, []> var_1039_equation_0 = const()[name = tensor<string, []>("op_1039_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1039_cast_fp16 = einsum(equation = var_1039_equation_0, values = (var_983_cast_fp16_3, var_1023_cast_fp16))[name = tensor<string, []>("op_1039_cast_fp16")]; |
| tensor<string, []> var_1041_equation_0 = const()[name = tensor<string, []>("op_1041_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1041_cast_fp16 = einsum(equation = var_1041_equation_0, values = (var_983_cast_fp16_4, var_1024_cast_fp16))[name = tensor<string, []>("op_1041_cast_fp16")]; |
| tensor<string, []> var_1043_equation_0 = const()[name = tensor<string, []>("op_1043_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1043_cast_fp16 = einsum(equation = var_1043_equation_0, values = (var_983_cast_fp16_5, var_1025_cast_fp16))[name = tensor<string, []>("op_1043_cast_fp16")]; |
| tensor<string, []> var_1045_equation_0 = const()[name = tensor<string, []>("op_1045_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1045_cast_fp16 = einsum(equation = var_1045_equation_0, values = (var_983_cast_fp16_6, var_1026_cast_fp16))[name = tensor<string, []>("op_1045_cast_fp16")]; |
| tensor<string, []> var_1047_equation_0 = const()[name = tensor<string, []>("op_1047_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1047_cast_fp16 = einsum(equation = var_1047_equation_0, values = (var_983_cast_fp16_7, var_1027_cast_fp16))[name = tensor<string, []>("op_1047_cast_fp16")]; |
| tensor<string, []> var_1049_equation_0 = const()[name = tensor<string, []>("op_1049_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1049_cast_fp16 = einsum(equation = var_1049_equation_0, values = (var_983_cast_fp16_8, var_1028_cast_fp16))[name = tensor<string, []>("op_1049_cast_fp16")]; |
| tensor<string, []> var_1051_equation_0 = const()[name = tensor<string, []>("op_1051_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1051_cast_fp16 = einsum(equation = var_1051_equation_0, values = (var_983_cast_fp16_9, var_1029_cast_fp16))[name = tensor<string, []>("op_1051_cast_fp16")]; |
| tensor<string, []> var_1053_equation_0 = const()[name = tensor<string, []>("op_1053_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1053_cast_fp16 = einsum(equation = var_1053_equation_0, values = (var_983_cast_fp16_10, var_1030_cast_fp16))[name = tensor<string, []>("op_1053_cast_fp16")]; |
| tensor<string, []> var_1055_equation_0 = const()[name = tensor<string, []>("op_1055_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1055_cast_fp16 = einsum(equation = var_1055_equation_0, values = (var_983_cast_fp16_11, var_1031_cast_fp16))[name = tensor<string, []>("op_1055_cast_fp16")]; |
| tensor<bool, []> input_45_interleave_0 = const()[name = tensor<string, []>("input_45_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_45_cast_fp16 = concat(axis = var_904, interleave = input_45_interleave_0, values = (var_1033_cast_fp16, var_1035_cast_fp16, var_1037_cast_fp16, var_1039_cast_fp16, var_1041_cast_fp16, var_1043_cast_fp16, var_1045_cast_fp16, var_1047_cast_fp16, var_1049_cast_fp16, var_1051_cast_fp16, var_1053_cast_fp16, var_1055_cast_fp16))[name = tensor<string, []>("input_45_cast_fp16")]; |
| tensor<string, []> var_1064_pad_type_0 = const()[name = tensor<string, []>("op_1064_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1064_strides_0 = const()[name = tensor<string, []>("op_1064_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1064_pad_0 = const()[name = tensor<string, []>("op_1064_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1064_dilations_0 = const()[name = tensor<string, []>("op_1064_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1064_groups_0 = const()[name = tensor<string, []>("op_1064_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_4_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66461248)))]; |
| tensor<fp16, [768]> blocks_4_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67640960)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1064_cast_fp16 = conv(bias = blocks_4_attn_out_bias_to_fp16, dilations = var_1064_dilations_0, groups = var_1064_groups_0, pad = var_1064_pad_0, pad_type = var_1064_pad_type_0, strides = var_1064_strides_0, weight = blocks_4_attn_out_weight_to_fp16, x = input_45_cast_fp16)[name = tensor<string, []>("op_1064_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = var_1064_cast_fp16)[name = tensor<string, []>("inputs_19_cast_fp16")]; |
| tensor<int32, [1]> input_47_axes_0 = const()[name = tensor<string, []>("input_47_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_47_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_47_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67642560)))]; |
| tensor<fp16, [768]> input_47_beta_0_to_fp16 = const()[name = tensor<string, []>("input_47_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67644160)))]; |
| tensor<fp16, []> var_1074_to_fp16 = const()[name = tensor<string, []>("op_1074_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_47_cast_fp16 = layer_norm(axes = input_47_axes_0, beta = input_47_beta_0_to_fp16, epsilon = var_1074_to_fp16, gamma = input_47_gamma_0_to_fp16, x = inputs_19_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")]; |
| tensor<string, []> input_49_pad_type_0 = const()[name = tensor<string, []>("input_49_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_49_strides_0 = const()[name = tensor<string, []>("input_49_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_49_pad_0 = const()[name = tensor<string, []>("input_49_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_49_dilations_0 = const()[name = tensor<string, []>("input_49_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_49_groups_0 = const()[name = tensor<string, []>("input_49_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_4_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67645760)))]; |
| tensor<fp16, [3072]> blocks_4_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72364416)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_49_cast_fp16 = conv(bias = blocks_4_mlp_0_bias_to_fp16, dilations = input_49_dilations_0, groups = input_49_groups_0, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = input_49_strides_0, weight = blocks_4_mlp_0_weight_to_fp16, x = input_47_cast_fp16)[name = tensor<string, []>("input_49_cast_fp16")]; |
| tensor<string, []> input_51_mode_0 = const()[name = tensor<string, []>("input_51_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_51_cast_fp16 = gelu(mode = input_51_mode_0, x = input_49_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")]; |
| tensor<string, []> var_1100_pad_type_0 = const()[name = tensor<string, []>("op_1100_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1100_strides_0 = const()[name = tensor<string, []>("op_1100_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1100_pad_0 = const()[name = tensor<string, []>("op_1100_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1100_dilations_0 = const()[name = tensor<string, []>("op_1100_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1100_groups_0 = const()[name = tensor<string, []>("op_1100_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_4_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72370624)))]; |
| tensor<fp16, [768]> blocks_4_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77089280)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1100_cast_fp16 = conv(bias = blocks_4_mlp_2_bias_to_fp16, dilations = var_1100_dilations_0, groups = var_1100_groups_0, pad = var_1100_pad_0, pad_type = var_1100_pad_type_0, strides = var_1100_strides_0, weight = blocks_4_mlp_2_weight_to_fp16, x = input_51_cast_fp16)[name = tensor<string, []>("op_1100_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = var_1100_cast_fp16)[name = tensor<string, []>("inputs_21_cast_fp16")]; |
| tensor<int32, []> var_1109 = const()[name = tensor<string, []>("op_1109"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_53_axes_0 = const()[name = tensor<string, []>("input_53_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_53_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_53_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77090880)))]; |
| tensor<fp16, [768]> input_53_beta_0_to_fp16 = const()[name = tensor<string, []>("input_53_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77092480)))]; |
| tensor<fp16, []> var_1125_to_fp16 = const()[name = tensor<string, []>("op_1125_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_53_cast_fp16 = layer_norm(axes = input_53_axes_0, beta = input_53_beta_0_to_fp16, epsilon = var_1125_to_fp16, gamma = input_53_gamma_0_to_fp16, x = inputs_21_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")]; |
| tensor<string, []> q_11_pad_type_0 = const()[name = tensor<string, []>("q_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_11_strides_0 = const()[name = tensor<string, []>("q_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_11_pad_0 = const()[name = tensor<string, []>("q_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_11_dilations_0 = const()[name = tensor<string, []>("q_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_11_groups_0 = const()[name = tensor<string, []>("q_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_1160_weight_0_to_fp16 = const()[name = tensor<string, []>("op_1160_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77094080)))]; |
| tensor<fp16, [768]> var_1160_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1160_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78273792)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1160_cast_fp16 = conv(bias = var_1160_bias_0_to_fp16, dilations = q_11_dilations_0, groups = q_11_groups_0, pad = q_11_pad_0, pad_type = q_11_pad_type_0, strides = q_11_strides_0, weight = var_1160_weight_0_to_fp16, x = input_53_cast_fp16)[name = tensor<string, []>("op_1160_cast_fp16")]; |
| tensor<string, []> k_11_pad_type_0 = const()[name = tensor<string, []>("k_11_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_11_strides_0 = const()[name = tensor<string, []>("k_11_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_11_pad_0 = const()[name = tensor<string, []>("k_11_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_11_dilations_0 = const()[name = tensor<string, []>("k_11_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_11_groups_0 = const()[name = tensor<string, []>("k_11_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_5_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78275392)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_11_cast_fp16 = conv(dilations = k_11_dilations_0, groups = k_11_groups_0, pad = k_11_pad_0, pad_type = k_11_pad_type_0, strides = k_11_strides_0, weight = blocks_5_attn_key_weight_to_fp16, x = input_53_cast_fp16)[name = tensor<string, []>("k_11_cast_fp16")]; |
| tensor<string, []> var_1158_pad_type_0 = const()[name = tensor<string, []>("op_1158_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1158_strides_0 = const()[name = tensor<string, []>("op_1158_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1158_pad_0 = const()[name = tensor<string, []>("op_1158_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1158_dilations_0 = const()[name = tensor<string, []>("op_1158_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1158_groups_0 = const()[name = tensor<string, []>("op_1158_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_5_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79455104)))]; |
| tensor<fp16, [768]> blocks_5_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80634816)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1158_cast_fp16 = conv(bias = blocks_5_attn_value_bias_to_fp16, dilations = var_1158_dilations_0, groups = var_1158_groups_0, pad = var_1158_pad_0, pad_type = var_1158_pad_type_0, strides = var_1158_strides_0, weight = blocks_5_attn_value_weight_to_fp16, x = input_53_cast_fp16)[name = tensor<string, []>("op_1158_cast_fp16")]; |
| tensor<int32, [12]> tile_15 = const()[name = tensor<string, []>("tile_15"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1161_axis_0 = const()[name = tensor<string, []>("op_1161_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1161_cast_fp16_11 = split(axis = var_1161_axis_0, split_sizes = tile_15, x = var_1160_cast_fp16)[name = tensor<string, []>("op_1161_cast_fp16")]; |
| tensor<int32, [4]> var_1174_perm_0 = const()[name = tensor<string, []>("op_1174_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_16 = const()[name = tensor<string, []>("tile_16"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1175_axis_0 = const()[name = tensor<string, []>("op_1175_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_1174_cast_fp16 = transpose(perm = var_1174_perm_0, x = k_11_cast_fp16)[name = tensor<string, []>("transpose_7")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_1175_cast_fp16_11 = split(axis = var_1175_axis_0, split_sizes = tile_16, x = var_1174_cast_fp16)[name = tensor<string, []>("op_1175_cast_fp16")]; |
| tensor<int32, [12]> tile_17 = const()[name = tensor<string, []>("tile_17"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1188_axis_0 = const()[name = tensor<string, []>("op_1188_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1188_cast_fp16_11 = split(axis = var_1188_axis_0, split_sizes = tile_17, x = var_1158_cast_fp16)[name = tensor<string, []>("op_1188_cast_fp16")]; |
| tensor<string, []> aw_121_equation_0 = const()[name = tensor<string, []>("aw_121_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_121_cast_fp16 = einsum(equation = aw_121_equation_0, values = (var_1175_cast_fp16_0, var_1161_cast_fp16_0))[name = tensor<string, []>("aw_121_cast_fp16")]; |
| tensor<string, []> aw_123_equation_0 = const()[name = tensor<string, []>("aw_123_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_123_cast_fp16 = einsum(equation = aw_123_equation_0, values = (var_1175_cast_fp16_1, var_1161_cast_fp16_1))[name = tensor<string, []>("aw_123_cast_fp16")]; |
| tensor<string, []> aw_125_equation_0 = const()[name = tensor<string, []>("aw_125_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_125_cast_fp16 = einsum(equation = aw_125_equation_0, values = (var_1175_cast_fp16_2, var_1161_cast_fp16_2))[name = tensor<string, []>("aw_125_cast_fp16")]; |
| tensor<string, []> aw_127_equation_0 = const()[name = tensor<string, []>("aw_127_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_127_cast_fp16 = einsum(equation = aw_127_equation_0, values = (var_1175_cast_fp16_3, var_1161_cast_fp16_3))[name = tensor<string, []>("aw_127_cast_fp16")]; |
| tensor<string, []> aw_129_equation_0 = const()[name = tensor<string, []>("aw_129_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_129_cast_fp16 = einsum(equation = aw_129_equation_0, values = (var_1175_cast_fp16_4, var_1161_cast_fp16_4))[name = tensor<string, []>("aw_129_cast_fp16")]; |
| tensor<string, []> aw_131_equation_0 = const()[name = tensor<string, []>("aw_131_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_131_cast_fp16 = einsum(equation = aw_131_equation_0, values = (var_1175_cast_fp16_5, var_1161_cast_fp16_5))[name = tensor<string, []>("aw_131_cast_fp16")]; |
| tensor<string, []> aw_133_equation_0 = const()[name = tensor<string, []>("aw_133_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_133_cast_fp16 = einsum(equation = aw_133_equation_0, values = (var_1175_cast_fp16_6, var_1161_cast_fp16_6))[name = tensor<string, []>("aw_133_cast_fp16")]; |
| tensor<string, []> aw_135_equation_0 = const()[name = tensor<string, []>("aw_135_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_135_cast_fp16 = einsum(equation = aw_135_equation_0, values = (var_1175_cast_fp16_7, var_1161_cast_fp16_7))[name = tensor<string, []>("aw_135_cast_fp16")]; |
| tensor<string, []> aw_137_equation_0 = const()[name = tensor<string, []>("aw_137_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_137_cast_fp16 = einsum(equation = aw_137_equation_0, values = (var_1175_cast_fp16_8, var_1161_cast_fp16_8))[name = tensor<string, []>("aw_137_cast_fp16")]; |
| tensor<string, []> aw_139_equation_0 = const()[name = tensor<string, []>("aw_139_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_139_cast_fp16 = einsum(equation = aw_139_equation_0, values = (var_1175_cast_fp16_9, var_1161_cast_fp16_9))[name = tensor<string, []>("aw_139_cast_fp16")]; |
| tensor<string, []> aw_141_equation_0 = const()[name = tensor<string, []>("aw_141_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_141_cast_fp16 = einsum(equation = aw_141_equation_0, values = (var_1175_cast_fp16_10, var_1161_cast_fp16_10))[name = tensor<string, []>("aw_141_cast_fp16")]; |
| tensor<string, []> aw_143_equation_0 = const()[name = tensor<string, []>("aw_143_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_143_cast_fp16 = einsum(equation = aw_143_equation_0, values = (var_1175_cast_fp16_11, var_1161_cast_fp16_11))[name = tensor<string, []>("aw_143_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1225_cast_fp16 = softmax(axis = var_1109, x = aw_121_cast_fp16)[name = tensor<string, []>("op_1225_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1226_cast_fp16 = softmax(axis = var_1109, x = aw_123_cast_fp16)[name = tensor<string, []>("op_1226_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1227_cast_fp16 = softmax(axis = var_1109, x = aw_125_cast_fp16)[name = tensor<string, []>("op_1227_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1228_cast_fp16 = softmax(axis = var_1109, x = aw_127_cast_fp16)[name = tensor<string, []>("op_1228_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1229_cast_fp16 = softmax(axis = var_1109, x = aw_129_cast_fp16)[name = tensor<string, []>("op_1229_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1230_cast_fp16 = softmax(axis = var_1109, x = aw_131_cast_fp16)[name = tensor<string, []>("op_1230_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1231_cast_fp16 = softmax(axis = var_1109, x = aw_133_cast_fp16)[name = tensor<string, []>("op_1231_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1232_cast_fp16 = softmax(axis = var_1109, x = aw_135_cast_fp16)[name = tensor<string, []>("op_1232_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1233_cast_fp16 = softmax(axis = var_1109, x = aw_137_cast_fp16)[name = tensor<string, []>("op_1233_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1234_cast_fp16 = softmax(axis = var_1109, x = aw_139_cast_fp16)[name = tensor<string, []>("op_1234_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1235_cast_fp16 = softmax(axis = var_1109, x = aw_141_cast_fp16)[name = tensor<string, []>("op_1235_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1236_cast_fp16 = softmax(axis = var_1109, x = aw_143_cast_fp16)[name = tensor<string, []>("op_1236_cast_fp16")]; |
| tensor<string, []> var_1238_equation_0 = const()[name = tensor<string, []>("op_1238_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1238_cast_fp16 = einsum(equation = var_1238_equation_0, values = (var_1188_cast_fp16_0, var_1225_cast_fp16))[name = tensor<string, []>("op_1238_cast_fp16")]; |
| tensor<string, []> var_1240_equation_0 = const()[name = tensor<string, []>("op_1240_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1240_cast_fp16 = einsum(equation = var_1240_equation_0, values = (var_1188_cast_fp16_1, var_1226_cast_fp16))[name = tensor<string, []>("op_1240_cast_fp16")]; |
| tensor<string, []> var_1242_equation_0 = const()[name = tensor<string, []>("op_1242_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1242_cast_fp16 = einsum(equation = var_1242_equation_0, values = (var_1188_cast_fp16_2, var_1227_cast_fp16))[name = tensor<string, []>("op_1242_cast_fp16")]; |
| tensor<string, []> var_1244_equation_0 = const()[name = tensor<string, []>("op_1244_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1244_cast_fp16 = einsum(equation = var_1244_equation_0, values = (var_1188_cast_fp16_3, var_1228_cast_fp16))[name = tensor<string, []>("op_1244_cast_fp16")]; |
| tensor<string, []> var_1246_equation_0 = const()[name = tensor<string, []>("op_1246_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1246_cast_fp16 = einsum(equation = var_1246_equation_0, values = (var_1188_cast_fp16_4, var_1229_cast_fp16))[name = tensor<string, []>("op_1246_cast_fp16")]; |
| tensor<string, []> var_1248_equation_0 = const()[name = tensor<string, []>("op_1248_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1248_cast_fp16 = einsum(equation = var_1248_equation_0, values = (var_1188_cast_fp16_5, var_1230_cast_fp16))[name = tensor<string, []>("op_1248_cast_fp16")]; |
| tensor<string, []> var_1250_equation_0 = const()[name = tensor<string, []>("op_1250_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1250_cast_fp16 = einsum(equation = var_1250_equation_0, values = (var_1188_cast_fp16_6, var_1231_cast_fp16))[name = tensor<string, []>("op_1250_cast_fp16")]; |
| tensor<string, []> var_1252_equation_0 = const()[name = tensor<string, []>("op_1252_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1252_cast_fp16 = einsum(equation = var_1252_equation_0, values = (var_1188_cast_fp16_7, var_1232_cast_fp16))[name = tensor<string, []>("op_1252_cast_fp16")]; |
| tensor<string, []> var_1254_equation_0 = const()[name = tensor<string, []>("op_1254_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1254_cast_fp16 = einsum(equation = var_1254_equation_0, values = (var_1188_cast_fp16_8, var_1233_cast_fp16))[name = tensor<string, []>("op_1254_cast_fp16")]; |
| tensor<string, []> var_1256_equation_0 = const()[name = tensor<string, []>("op_1256_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1256_cast_fp16 = einsum(equation = var_1256_equation_0, values = (var_1188_cast_fp16_9, var_1234_cast_fp16))[name = tensor<string, []>("op_1256_cast_fp16")]; |
| tensor<string, []> var_1258_equation_0 = const()[name = tensor<string, []>("op_1258_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1258_cast_fp16 = einsum(equation = var_1258_equation_0, values = (var_1188_cast_fp16_10, var_1235_cast_fp16))[name = tensor<string, []>("op_1258_cast_fp16")]; |
| tensor<string, []> var_1260_equation_0 = const()[name = tensor<string, []>("op_1260_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1260_cast_fp16 = einsum(equation = var_1260_equation_0, values = (var_1188_cast_fp16_11, var_1236_cast_fp16))[name = tensor<string, []>("op_1260_cast_fp16")]; |
| tensor<bool, []> input_55_interleave_0 = const()[name = tensor<string, []>("input_55_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_55_cast_fp16 = concat(axis = var_1109, interleave = input_55_interleave_0, values = (var_1238_cast_fp16, var_1240_cast_fp16, var_1242_cast_fp16, var_1244_cast_fp16, var_1246_cast_fp16, var_1248_cast_fp16, var_1250_cast_fp16, var_1252_cast_fp16, var_1254_cast_fp16, var_1256_cast_fp16, var_1258_cast_fp16, var_1260_cast_fp16))[name = tensor<string, []>("input_55_cast_fp16")]; |
| tensor<string, []> var_1269_pad_type_0 = const()[name = tensor<string, []>("op_1269_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1269_strides_0 = const()[name = tensor<string, []>("op_1269_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1269_pad_0 = const()[name = tensor<string, []>("op_1269_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1269_dilations_0 = const()[name = tensor<string, []>("op_1269_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1269_groups_0 = const()[name = tensor<string, []>("op_1269_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_5_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80636416)))]; |
| tensor<fp16, [768]> blocks_5_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81816128)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1269_cast_fp16 = conv(bias = blocks_5_attn_out_bias_to_fp16, dilations = var_1269_dilations_0, groups = var_1269_groups_0, pad = var_1269_pad_0, pad_type = var_1269_pad_type_0, strides = var_1269_strides_0, weight = blocks_5_attn_out_weight_to_fp16, x = input_55_cast_fp16)[name = tensor<string, []>("op_1269_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = var_1269_cast_fp16)[name = tensor<string, []>("inputs_23_cast_fp16")]; |
| tensor<int32, [1]> input_57_axes_0 = const()[name = tensor<string, []>("input_57_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_57_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_57_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81817728)))]; |
| tensor<fp16, [768]> input_57_beta_0_to_fp16 = const()[name = tensor<string, []>("input_57_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81819328)))]; |
| tensor<fp16, []> var_1279_to_fp16 = const()[name = tensor<string, []>("op_1279_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_57_cast_fp16 = layer_norm(axes = input_57_axes_0, beta = input_57_beta_0_to_fp16, epsilon = var_1279_to_fp16, gamma = input_57_gamma_0_to_fp16, x = inputs_23_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")]; |
| tensor<string, []> input_59_pad_type_0 = const()[name = tensor<string, []>("input_59_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_59_strides_0 = const()[name = tensor<string, []>("input_59_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_59_pad_0 = const()[name = tensor<string, []>("input_59_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_59_dilations_0 = const()[name = tensor<string, []>("input_59_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_59_groups_0 = const()[name = tensor<string, []>("input_59_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_5_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81820928)))]; |
| tensor<fp16, [3072]> blocks_5_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86539584)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_59_cast_fp16 = conv(bias = blocks_5_mlp_0_bias_to_fp16, dilations = input_59_dilations_0, groups = input_59_groups_0, pad = input_59_pad_0, pad_type = input_59_pad_type_0, strides = input_59_strides_0, weight = blocks_5_mlp_0_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")]; |
| tensor<string, []> input_61_mode_0 = const()[name = tensor<string, []>("input_61_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_61_cast_fp16 = gelu(mode = input_61_mode_0, x = input_59_cast_fp16)[name = tensor<string, []>("input_61_cast_fp16")]; |
| tensor<string, []> var_1305_pad_type_0 = const()[name = tensor<string, []>("op_1305_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1305_strides_0 = const()[name = tensor<string, []>("op_1305_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1305_pad_0 = const()[name = tensor<string, []>("op_1305_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1305_dilations_0 = const()[name = tensor<string, []>("op_1305_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1305_groups_0 = const()[name = tensor<string, []>("op_1305_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_5_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86545792)))]; |
| tensor<fp16, [768]> blocks_5_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91264448)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1305_cast_fp16 = conv(bias = blocks_5_mlp_2_bias_to_fp16, dilations = var_1305_dilations_0, groups = var_1305_groups_0, pad = var_1305_pad_0, pad_type = var_1305_pad_type_0, strides = var_1305_strides_0, weight = blocks_5_mlp_2_weight_to_fp16, x = input_61_cast_fp16)[name = tensor<string, []>("op_1305_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = var_1305_cast_fp16)[name = tensor<string, []>("inputs_25_cast_fp16")]; |
| tensor<int32, []> var_1314 = const()[name = tensor<string, []>("op_1314"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_63_axes_0 = const()[name = tensor<string, []>("input_63_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_63_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_63_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91266048)))]; |
| tensor<fp16, [768]> input_63_beta_0_to_fp16 = const()[name = tensor<string, []>("input_63_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91267648)))]; |
| tensor<fp16, []> var_1330_to_fp16 = const()[name = tensor<string, []>("op_1330_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_63_cast_fp16 = layer_norm(axes = input_63_axes_0, beta = input_63_beta_0_to_fp16, epsilon = var_1330_to_fp16, gamma = input_63_gamma_0_to_fp16, x = inputs_25_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")]; |
| tensor<string, []> q_13_pad_type_0 = const()[name = tensor<string, []>("q_13_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_13_strides_0 = const()[name = tensor<string, []>("q_13_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_13_pad_0 = const()[name = tensor<string, []>("q_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_13_dilations_0 = const()[name = tensor<string, []>("q_13_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_13_groups_0 = const()[name = tensor<string, []>("q_13_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_1365_weight_0_to_fp16 = const()[name = tensor<string, []>("op_1365_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91269248)))]; |
| tensor<fp16, [768]> var_1365_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1365_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92448960)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1365_cast_fp16 = conv(bias = var_1365_bias_0_to_fp16, dilations = q_13_dilations_0, groups = q_13_groups_0, pad = q_13_pad_0, pad_type = q_13_pad_type_0, strides = q_13_strides_0, weight = var_1365_weight_0_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("op_1365_cast_fp16")]; |
| tensor<string, []> k_13_pad_type_0 = const()[name = tensor<string, []>("k_13_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_13_strides_0 = const()[name = tensor<string, []>("k_13_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_13_pad_0 = const()[name = tensor<string, []>("k_13_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_13_dilations_0 = const()[name = tensor<string, []>("k_13_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_13_groups_0 = const()[name = tensor<string, []>("k_13_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_6_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92450560)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_13_cast_fp16 = conv(dilations = k_13_dilations_0, groups = k_13_groups_0, pad = k_13_pad_0, pad_type = k_13_pad_type_0, strides = k_13_strides_0, weight = blocks_6_attn_key_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("k_13_cast_fp16")]; |
| tensor<string, []> var_1363_pad_type_0 = const()[name = tensor<string, []>("op_1363_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1363_strides_0 = const()[name = tensor<string, []>("op_1363_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1363_pad_0 = const()[name = tensor<string, []>("op_1363_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1363_dilations_0 = const()[name = tensor<string, []>("op_1363_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1363_groups_0 = const()[name = tensor<string, []>("op_1363_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_6_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93630272)))]; |
| tensor<fp16, [768]> blocks_6_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94809984)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1363_cast_fp16 = conv(bias = blocks_6_attn_value_bias_to_fp16, dilations = var_1363_dilations_0, groups = var_1363_groups_0, pad = var_1363_pad_0, pad_type = var_1363_pad_type_0, strides = var_1363_strides_0, weight = blocks_6_attn_value_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("op_1363_cast_fp16")]; |
| tensor<int32, [12]> tile_18 = const()[name = tensor<string, []>("tile_18"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1366_axis_0 = const()[name = tensor<string, []>("op_1366_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1366_cast_fp16_11 = split(axis = var_1366_axis_0, split_sizes = tile_18, x = var_1365_cast_fp16)[name = tensor<string, []>("op_1366_cast_fp16")]; |
| tensor<int32, [4]> var_1379_perm_0 = const()[name = tensor<string, []>("op_1379_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_19 = const()[name = tensor<string, []>("tile_19"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1380_axis_0 = const()[name = tensor<string, []>("op_1380_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_1379_cast_fp16 = transpose(perm = var_1379_perm_0, x = k_13_cast_fp16)[name = tensor<string, []>("transpose_6")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_1380_cast_fp16_11 = split(axis = var_1380_axis_0, split_sizes = tile_19, x = var_1379_cast_fp16)[name = tensor<string, []>("op_1380_cast_fp16")]; |
| tensor<int32, [12]> tile_20 = const()[name = tensor<string, []>("tile_20"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1393_axis_0 = const()[name = tensor<string, []>("op_1393_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1393_cast_fp16_11 = split(axis = var_1393_axis_0, split_sizes = tile_20, x = var_1363_cast_fp16)[name = tensor<string, []>("op_1393_cast_fp16")]; |
| tensor<string, []> aw_145_equation_0 = const()[name = tensor<string, []>("aw_145_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_145_cast_fp16 = einsum(equation = aw_145_equation_0, values = (var_1380_cast_fp16_0, var_1366_cast_fp16_0))[name = tensor<string, []>("aw_145_cast_fp16")]; |
| tensor<string, []> aw_147_equation_0 = const()[name = tensor<string, []>("aw_147_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_147_cast_fp16 = einsum(equation = aw_147_equation_0, values = (var_1380_cast_fp16_1, var_1366_cast_fp16_1))[name = tensor<string, []>("aw_147_cast_fp16")]; |
| tensor<string, []> aw_149_equation_0 = const()[name = tensor<string, []>("aw_149_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_149_cast_fp16 = einsum(equation = aw_149_equation_0, values = (var_1380_cast_fp16_2, var_1366_cast_fp16_2))[name = tensor<string, []>("aw_149_cast_fp16")]; |
| tensor<string, []> aw_151_equation_0 = const()[name = tensor<string, []>("aw_151_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_151_cast_fp16 = einsum(equation = aw_151_equation_0, values = (var_1380_cast_fp16_3, var_1366_cast_fp16_3))[name = tensor<string, []>("aw_151_cast_fp16")]; |
| tensor<string, []> aw_153_equation_0 = const()[name = tensor<string, []>("aw_153_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_153_cast_fp16 = einsum(equation = aw_153_equation_0, values = (var_1380_cast_fp16_4, var_1366_cast_fp16_4))[name = tensor<string, []>("aw_153_cast_fp16")]; |
| tensor<string, []> aw_155_equation_0 = const()[name = tensor<string, []>("aw_155_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_155_cast_fp16 = einsum(equation = aw_155_equation_0, values = (var_1380_cast_fp16_5, var_1366_cast_fp16_5))[name = tensor<string, []>("aw_155_cast_fp16")]; |
| tensor<string, []> aw_157_equation_0 = const()[name = tensor<string, []>("aw_157_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_157_cast_fp16 = einsum(equation = aw_157_equation_0, values = (var_1380_cast_fp16_6, var_1366_cast_fp16_6))[name = tensor<string, []>("aw_157_cast_fp16")]; |
| tensor<string, []> aw_159_equation_0 = const()[name = tensor<string, []>("aw_159_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_159_cast_fp16 = einsum(equation = aw_159_equation_0, values = (var_1380_cast_fp16_7, var_1366_cast_fp16_7))[name = tensor<string, []>("aw_159_cast_fp16")]; |
| tensor<string, []> aw_161_equation_0 = const()[name = tensor<string, []>("aw_161_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_161_cast_fp16 = einsum(equation = aw_161_equation_0, values = (var_1380_cast_fp16_8, var_1366_cast_fp16_8))[name = tensor<string, []>("aw_161_cast_fp16")]; |
| tensor<string, []> aw_163_equation_0 = const()[name = tensor<string, []>("aw_163_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_163_cast_fp16 = einsum(equation = aw_163_equation_0, values = (var_1380_cast_fp16_9, var_1366_cast_fp16_9))[name = tensor<string, []>("aw_163_cast_fp16")]; |
| tensor<string, []> aw_165_equation_0 = const()[name = tensor<string, []>("aw_165_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_165_cast_fp16 = einsum(equation = aw_165_equation_0, values = (var_1380_cast_fp16_10, var_1366_cast_fp16_10))[name = tensor<string, []>("aw_165_cast_fp16")]; |
| tensor<string, []> aw_167_equation_0 = const()[name = tensor<string, []>("aw_167_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_167_cast_fp16 = einsum(equation = aw_167_equation_0, values = (var_1380_cast_fp16_11, var_1366_cast_fp16_11))[name = tensor<string, []>("aw_167_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1430_cast_fp16 = softmax(axis = var_1314, x = aw_145_cast_fp16)[name = tensor<string, []>("op_1430_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1431_cast_fp16 = softmax(axis = var_1314, x = aw_147_cast_fp16)[name = tensor<string, []>("op_1431_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1432_cast_fp16 = softmax(axis = var_1314, x = aw_149_cast_fp16)[name = tensor<string, []>("op_1432_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1433_cast_fp16 = softmax(axis = var_1314, x = aw_151_cast_fp16)[name = tensor<string, []>("op_1433_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1434_cast_fp16 = softmax(axis = var_1314, x = aw_153_cast_fp16)[name = tensor<string, []>("op_1434_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1435_cast_fp16 = softmax(axis = var_1314, x = aw_155_cast_fp16)[name = tensor<string, []>("op_1435_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1436_cast_fp16 = softmax(axis = var_1314, x = aw_157_cast_fp16)[name = tensor<string, []>("op_1436_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1437_cast_fp16 = softmax(axis = var_1314, x = aw_159_cast_fp16)[name = tensor<string, []>("op_1437_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1438_cast_fp16 = softmax(axis = var_1314, x = aw_161_cast_fp16)[name = tensor<string, []>("op_1438_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1439_cast_fp16 = softmax(axis = var_1314, x = aw_163_cast_fp16)[name = tensor<string, []>("op_1439_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1440_cast_fp16 = softmax(axis = var_1314, x = aw_165_cast_fp16)[name = tensor<string, []>("op_1440_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1441_cast_fp16 = softmax(axis = var_1314, x = aw_167_cast_fp16)[name = tensor<string, []>("op_1441_cast_fp16")]; |
| tensor<string, []> var_1443_equation_0 = const()[name = tensor<string, []>("op_1443_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1443_cast_fp16 = einsum(equation = var_1443_equation_0, values = (var_1393_cast_fp16_0, var_1430_cast_fp16))[name = tensor<string, []>("op_1443_cast_fp16")]; |
| tensor<string, []> var_1445_equation_0 = const()[name = tensor<string, []>("op_1445_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1445_cast_fp16 = einsum(equation = var_1445_equation_0, values = (var_1393_cast_fp16_1, var_1431_cast_fp16))[name = tensor<string, []>("op_1445_cast_fp16")]; |
| tensor<string, []> var_1447_equation_0 = const()[name = tensor<string, []>("op_1447_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1447_cast_fp16 = einsum(equation = var_1447_equation_0, values = (var_1393_cast_fp16_2, var_1432_cast_fp16))[name = tensor<string, []>("op_1447_cast_fp16")]; |
| tensor<string, []> var_1449_equation_0 = const()[name = tensor<string, []>("op_1449_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1449_cast_fp16 = einsum(equation = var_1449_equation_0, values = (var_1393_cast_fp16_3, var_1433_cast_fp16))[name = tensor<string, []>("op_1449_cast_fp16")]; |
| tensor<string, []> var_1451_equation_0 = const()[name = tensor<string, []>("op_1451_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1451_cast_fp16 = einsum(equation = var_1451_equation_0, values = (var_1393_cast_fp16_4, var_1434_cast_fp16))[name = tensor<string, []>("op_1451_cast_fp16")]; |
| tensor<string, []> var_1453_equation_0 = const()[name = tensor<string, []>("op_1453_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1453_cast_fp16 = einsum(equation = var_1453_equation_0, values = (var_1393_cast_fp16_5, var_1435_cast_fp16))[name = tensor<string, []>("op_1453_cast_fp16")]; |
| tensor<string, []> var_1455_equation_0 = const()[name = tensor<string, []>("op_1455_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1455_cast_fp16 = einsum(equation = var_1455_equation_0, values = (var_1393_cast_fp16_6, var_1436_cast_fp16))[name = tensor<string, []>("op_1455_cast_fp16")]; |
| tensor<string, []> var_1457_equation_0 = const()[name = tensor<string, []>("op_1457_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1457_cast_fp16 = einsum(equation = var_1457_equation_0, values = (var_1393_cast_fp16_7, var_1437_cast_fp16))[name = tensor<string, []>("op_1457_cast_fp16")]; |
| tensor<string, []> var_1459_equation_0 = const()[name = tensor<string, []>("op_1459_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1459_cast_fp16 = einsum(equation = var_1459_equation_0, values = (var_1393_cast_fp16_8, var_1438_cast_fp16))[name = tensor<string, []>("op_1459_cast_fp16")]; |
| tensor<string, []> var_1461_equation_0 = const()[name = tensor<string, []>("op_1461_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1461_cast_fp16 = einsum(equation = var_1461_equation_0, values = (var_1393_cast_fp16_9, var_1439_cast_fp16))[name = tensor<string, []>("op_1461_cast_fp16")]; |
| tensor<string, []> var_1463_equation_0 = const()[name = tensor<string, []>("op_1463_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1463_cast_fp16 = einsum(equation = var_1463_equation_0, values = (var_1393_cast_fp16_10, var_1440_cast_fp16))[name = tensor<string, []>("op_1463_cast_fp16")]; |
| tensor<string, []> var_1465_equation_0 = const()[name = tensor<string, []>("op_1465_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1465_cast_fp16 = einsum(equation = var_1465_equation_0, values = (var_1393_cast_fp16_11, var_1441_cast_fp16))[name = tensor<string, []>("op_1465_cast_fp16")]; |
| tensor<bool, []> input_65_interleave_0 = const()[name = tensor<string, []>("input_65_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_65_cast_fp16 = concat(axis = var_1314, interleave = input_65_interleave_0, values = (var_1443_cast_fp16, var_1445_cast_fp16, var_1447_cast_fp16, var_1449_cast_fp16, var_1451_cast_fp16, var_1453_cast_fp16, var_1455_cast_fp16, var_1457_cast_fp16, var_1459_cast_fp16, var_1461_cast_fp16, var_1463_cast_fp16, var_1465_cast_fp16))[name = tensor<string, []>("input_65_cast_fp16")]; |
| tensor<string, []> var_1474_pad_type_0 = const()[name = tensor<string, []>("op_1474_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1474_strides_0 = const()[name = tensor<string, []>("op_1474_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1474_pad_0 = const()[name = tensor<string, []>("op_1474_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1474_dilations_0 = const()[name = tensor<string, []>("op_1474_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1474_groups_0 = const()[name = tensor<string, []>("op_1474_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_6_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94811584)))]; |
| tensor<fp16, [768]> blocks_6_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95991296)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1474_cast_fp16 = conv(bias = blocks_6_attn_out_bias_to_fp16, dilations = var_1474_dilations_0, groups = var_1474_groups_0, pad = var_1474_pad_0, pad_type = var_1474_pad_type_0, strides = var_1474_strides_0, weight = blocks_6_attn_out_weight_to_fp16, x = input_65_cast_fp16)[name = tensor<string, []>("op_1474_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = var_1474_cast_fp16)[name = tensor<string, []>("inputs_27_cast_fp16")]; |
| tensor<int32, [1]> input_67_axes_0 = const()[name = tensor<string, []>("input_67_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_67_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_67_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95992896)))]; |
| tensor<fp16, [768]> input_67_beta_0_to_fp16 = const()[name = tensor<string, []>("input_67_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95994496)))]; |
| tensor<fp16, []> var_1484_to_fp16 = const()[name = tensor<string, []>("op_1484_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_67_cast_fp16 = layer_norm(axes = input_67_axes_0, beta = input_67_beta_0_to_fp16, epsilon = var_1484_to_fp16, gamma = input_67_gamma_0_to_fp16, x = inputs_27_cast_fp16)[name = tensor<string, []>("input_67_cast_fp16")]; |
| tensor<string, []> input_69_pad_type_0 = const()[name = tensor<string, []>("input_69_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_69_strides_0 = const()[name = tensor<string, []>("input_69_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_69_pad_0 = const()[name = tensor<string, []>("input_69_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_69_dilations_0 = const()[name = tensor<string, []>("input_69_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_69_groups_0 = const()[name = tensor<string, []>("input_69_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_6_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95996096)))]; |
| tensor<fp16, [3072]> blocks_6_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100714752)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_69_cast_fp16 = conv(bias = blocks_6_mlp_0_bias_to_fp16, dilations = input_69_dilations_0, groups = input_69_groups_0, pad = input_69_pad_0, pad_type = input_69_pad_type_0, strides = input_69_strides_0, weight = blocks_6_mlp_0_weight_to_fp16, x = input_67_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")]; |
| tensor<string, []> input_71_mode_0 = const()[name = tensor<string, []>("input_71_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_71_cast_fp16 = gelu(mode = input_71_mode_0, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")]; |
| tensor<string, []> var_1510_pad_type_0 = const()[name = tensor<string, []>("op_1510_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1510_strides_0 = const()[name = tensor<string, []>("op_1510_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1510_pad_0 = const()[name = tensor<string, []>("op_1510_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1510_dilations_0 = const()[name = tensor<string, []>("op_1510_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1510_groups_0 = const()[name = tensor<string, []>("op_1510_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_6_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100720960)))]; |
| tensor<fp16, [768]> blocks_6_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105439616)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1510_cast_fp16 = conv(bias = blocks_6_mlp_2_bias_to_fp16, dilations = var_1510_dilations_0, groups = var_1510_groups_0, pad = var_1510_pad_0, pad_type = var_1510_pad_type_0, strides = var_1510_strides_0, weight = blocks_6_mlp_2_weight_to_fp16, x = input_71_cast_fp16)[name = tensor<string, []>("op_1510_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = var_1510_cast_fp16)[name = tensor<string, []>("inputs_29_cast_fp16")]; |
| tensor<int32, []> var_1519 = const()[name = tensor<string, []>("op_1519"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_73_axes_0 = const()[name = tensor<string, []>("input_73_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_73_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_73_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105441216)))]; |
| tensor<fp16, [768]> input_73_beta_0_to_fp16 = const()[name = tensor<string, []>("input_73_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105442816)))]; |
| tensor<fp16, []> var_1535_to_fp16 = const()[name = tensor<string, []>("op_1535_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = input_73_beta_0_to_fp16, epsilon = var_1535_to_fp16, gamma = input_73_gamma_0_to_fp16, x = inputs_29_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")]; |
| tensor<string, []> q_15_pad_type_0 = const()[name = tensor<string, []>("q_15_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_15_strides_0 = const()[name = tensor<string, []>("q_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_15_pad_0 = const()[name = tensor<string, []>("q_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_15_dilations_0 = const()[name = tensor<string, []>("q_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_15_groups_0 = const()[name = tensor<string, []>("q_15_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_1570_weight_0_to_fp16 = const()[name = tensor<string, []>("op_1570_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105444416)))]; |
| tensor<fp16, [768]> var_1570_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1570_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106624128)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1570_cast_fp16 = conv(bias = var_1570_bias_0_to_fp16, dilations = q_15_dilations_0, groups = q_15_groups_0, pad = q_15_pad_0, pad_type = q_15_pad_type_0, strides = q_15_strides_0, weight = var_1570_weight_0_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("op_1570_cast_fp16")]; |
| tensor<string, []> k_15_pad_type_0 = const()[name = tensor<string, []>("k_15_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_15_strides_0 = const()[name = tensor<string, []>("k_15_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_15_pad_0 = const()[name = tensor<string, []>("k_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_15_dilations_0 = const()[name = tensor<string, []>("k_15_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_15_groups_0 = const()[name = tensor<string, []>("k_15_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_7_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106625728)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_15_cast_fp16 = conv(dilations = k_15_dilations_0, groups = k_15_groups_0, pad = k_15_pad_0, pad_type = k_15_pad_type_0, strides = k_15_strides_0, weight = blocks_7_attn_key_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("k_15_cast_fp16")]; |
| tensor<string, []> var_1568_pad_type_0 = const()[name = tensor<string, []>("op_1568_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1568_strides_0 = const()[name = tensor<string, []>("op_1568_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1568_pad_0 = const()[name = tensor<string, []>("op_1568_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1568_dilations_0 = const()[name = tensor<string, []>("op_1568_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1568_groups_0 = const()[name = tensor<string, []>("op_1568_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_7_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107805440)))]; |
| tensor<fp16, [768]> blocks_7_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108985152)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1568_cast_fp16 = conv(bias = blocks_7_attn_value_bias_to_fp16, dilations = var_1568_dilations_0, groups = var_1568_groups_0, pad = var_1568_pad_0, pad_type = var_1568_pad_type_0, strides = var_1568_strides_0, weight = blocks_7_attn_value_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("op_1568_cast_fp16")]; |
| tensor<int32, [12]> tile_21 = const()[name = tensor<string, []>("tile_21"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1571_axis_0 = const()[name = tensor<string, []>("op_1571_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1571_cast_fp16_11 = split(axis = var_1571_axis_0, split_sizes = tile_21, x = var_1570_cast_fp16)[name = tensor<string, []>("op_1571_cast_fp16")]; |
| tensor<int32, [4]> var_1584_perm_0 = const()[name = tensor<string, []>("op_1584_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_22 = const()[name = tensor<string, []>("tile_22"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1585_axis_0 = const()[name = tensor<string, []>("op_1585_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_1584_cast_fp16 = transpose(perm = var_1584_perm_0, x = k_15_cast_fp16)[name = tensor<string, []>("transpose_5")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_1585_cast_fp16_11 = split(axis = var_1585_axis_0, split_sizes = tile_22, x = var_1584_cast_fp16)[name = tensor<string, []>("op_1585_cast_fp16")]; |
| tensor<int32, [12]> tile_23 = const()[name = tensor<string, []>("tile_23"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1598_axis_0 = const()[name = tensor<string, []>("op_1598_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1598_cast_fp16_11 = split(axis = var_1598_axis_0, split_sizes = tile_23, x = var_1568_cast_fp16)[name = tensor<string, []>("op_1598_cast_fp16")]; |
| tensor<string, []> aw_169_equation_0 = const()[name = tensor<string, []>("aw_169_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_169_cast_fp16 = einsum(equation = aw_169_equation_0, values = (var_1585_cast_fp16_0, var_1571_cast_fp16_0))[name = tensor<string, []>("aw_169_cast_fp16")]; |
| tensor<string, []> aw_171_equation_0 = const()[name = tensor<string, []>("aw_171_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_171_cast_fp16 = einsum(equation = aw_171_equation_0, values = (var_1585_cast_fp16_1, var_1571_cast_fp16_1))[name = tensor<string, []>("aw_171_cast_fp16")]; |
| tensor<string, []> aw_173_equation_0 = const()[name = tensor<string, []>("aw_173_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_173_cast_fp16 = einsum(equation = aw_173_equation_0, values = (var_1585_cast_fp16_2, var_1571_cast_fp16_2))[name = tensor<string, []>("aw_173_cast_fp16")]; |
| tensor<string, []> aw_175_equation_0 = const()[name = tensor<string, []>("aw_175_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_175_cast_fp16 = einsum(equation = aw_175_equation_0, values = (var_1585_cast_fp16_3, var_1571_cast_fp16_3))[name = tensor<string, []>("aw_175_cast_fp16")]; |
| tensor<string, []> aw_177_equation_0 = const()[name = tensor<string, []>("aw_177_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_177_cast_fp16 = einsum(equation = aw_177_equation_0, values = (var_1585_cast_fp16_4, var_1571_cast_fp16_4))[name = tensor<string, []>("aw_177_cast_fp16")]; |
| tensor<string, []> aw_179_equation_0 = const()[name = tensor<string, []>("aw_179_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_179_cast_fp16 = einsum(equation = aw_179_equation_0, values = (var_1585_cast_fp16_5, var_1571_cast_fp16_5))[name = tensor<string, []>("aw_179_cast_fp16")]; |
| tensor<string, []> aw_181_equation_0 = const()[name = tensor<string, []>("aw_181_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_181_cast_fp16 = einsum(equation = aw_181_equation_0, values = (var_1585_cast_fp16_6, var_1571_cast_fp16_6))[name = tensor<string, []>("aw_181_cast_fp16")]; |
| tensor<string, []> aw_183_equation_0 = const()[name = tensor<string, []>("aw_183_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_183_cast_fp16 = einsum(equation = aw_183_equation_0, values = (var_1585_cast_fp16_7, var_1571_cast_fp16_7))[name = tensor<string, []>("aw_183_cast_fp16")]; |
| tensor<string, []> aw_185_equation_0 = const()[name = tensor<string, []>("aw_185_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_185_cast_fp16 = einsum(equation = aw_185_equation_0, values = (var_1585_cast_fp16_8, var_1571_cast_fp16_8))[name = tensor<string, []>("aw_185_cast_fp16")]; |
| tensor<string, []> aw_187_equation_0 = const()[name = tensor<string, []>("aw_187_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_187_cast_fp16 = einsum(equation = aw_187_equation_0, values = (var_1585_cast_fp16_9, var_1571_cast_fp16_9))[name = tensor<string, []>("aw_187_cast_fp16")]; |
| tensor<string, []> aw_189_equation_0 = const()[name = tensor<string, []>("aw_189_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_189_cast_fp16 = einsum(equation = aw_189_equation_0, values = (var_1585_cast_fp16_10, var_1571_cast_fp16_10))[name = tensor<string, []>("aw_189_cast_fp16")]; |
| tensor<string, []> aw_191_equation_0 = const()[name = tensor<string, []>("aw_191_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_191_cast_fp16 = einsum(equation = aw_191_equation_0, values = (var_1585_cast_fp16_11, var_1571_cast_fp16_11))[name = tensor<string, []>("aw_191_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1635_cast_fp16 = softmax(axis = var_1519, x = aw_169_cast_fp16)[name = tensor<string, []>("op_1635_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1636_cast_fp16 = softmax(axis = var_1519, x = aw_171_cast_fp16)[name = tensor<string, []>("op_1636_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1637_cast_fp16 = softmax(axis = var_1519, x = aw_173_cast_fp16)[name = tensor<string, []>("op_1637_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1638_cast_fp16 = softmax(axis = var_1519, x = aw_175_cast_fp16)[name = tensor<string, []>("op_1638_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1639_cast_fp16 = softmax(axis = var_1519, x = aw_177_cast_fp16)[name = tensor<string, []>("op_1639_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1640_cast_fp16 = softmax(axis = var_1519, x = aw_179_cast_fp16)[name = tensor<string, []>("op_1640_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1641_cast_fp16 = softmax(axis = var_1519, x = aw_181_cast_fp16)[name = tensor<string, []>("op_1641_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1642_cast_fp16 = softmax(axis = var_1519, x = aw_183_cast_fp16)[name = tensor<string, []>("op_1642_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1643_cast_fp16 = softmax(axis = var_1519, x = aw_185_cast_fp16)[name = tensor<string, []>("op_1643_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1644_cast_fp16 = softmax(axis = var_1519, x = aw_187_cast_fp16)[name = tensor<string, []>("op_1644_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1645_cast_fp16 = softmax(axis = var_1519, x = aw_189_cast_fp16)[name = tensor<string, []>("op_1645_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1646_cast_fp16 = softmax(axis = var_1519, x = aw_191_cast_fp16)[name = tensor<string, []>("op_1646_cast_fp16")]; |
| tensor<string, []> var_1648_equation_0 = const()[name = tensor<string, []>("op_1648_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1648_cast_fp16 = einsum(equation = var_1648_equation_0, values = (var_1598_cast_fp16_0, var_1635_cast_fp16))[name = tensor<string, []>("op_1648_cast_fp16")]; |
| tensor<string, []> var_1650_equation_0 = const()[name = tensor<string, []>("op_1650_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1650_cast_fp16 = einsum(equation = var_1650_equation_0, values = (var_1598_cast_fp16_1, var_1636_cast_fp16))[name = tensor<string, []>("op_1650_cast_fp16")]; |
| tensor<string, []> var_1652_equation_0 = const()[name = tensor<string, []>("op_1652_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1652_cast_fp16 = einsum(equation = var_1652_equation_0, values = (var_1598_cast_fp16_2, var_1637_cast_fp16))[name = tensor<string, []>("op_1652_cast_fp16")]; |
| tensor<string, []> var_1654_equation_0 = const()[name = tensor<string, []>("op_1654_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1654_cast_fp16 = einsum(equation = var_1654_equation_0, values = (var_1598_cast_fp16_3, var_1638_cast_fp16))[name = tensor<string, []>("op_1654_cast_fp16")]; |
| tensor<string, []> var_1656_equation_0 = const()[name = tensor<string, []>("op_1656_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1656_cast_fp16 = einsum(equation = var_1656_equation_0, values = (var_1598_cast_fp16_4, var_1639_cast_fp16))[name = tensor<string, []>("op_1656_cast_fp16")]; |
| tensor<string, []> var_1658_equation_0 = const()[name = tensor<string, []>("op_1658_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1658_cast_fp16 = einsum(equation = var_1658_equation_0, values = (var_1598_cast_fp16_5, var_1640_cast_fp16))[name = tensor<string, []>("op_1658_cast_fp16")]; |
| tensor<string, []> var_1660_equation_0 = const()[name = tensor<string, []>("op_1660_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1660_cast_fp16 = einsum(equation = var_1660_equation_0, values = (var_1598_cast_fp16_6, var_1641_cast_fp16))[name = tensor<string, []>("op_1660_cast_fp16")]; |
| tensor<string, []> var_1662_equation_0 = const()[name = tensor<string, []>("op_1662_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1662_cast_fp16 = einsum(equation = var_1662_equation_0, values = (var_1598_cast_fp16_7, var_1642_cast_fp16))[name = tensor<string, []>("op_1662_cast_fp16")]; |
| tensor<string, []> var_1664_equation_0 = const()[name = tensor<string, []>("op_1664_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1664_cast_fp16 = einsum(equation = var_1664_equation_0, values = (var_1598_cast_fp16_8, var_1643_cast_fp16))[name = tensor<string, []>("op_1664_cast_fp16")]; |
| tensor<string, []> var_1666_equation_0 = const()[name = tensor<string, []>("op_1666_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1666_cast_fp16 = einsum(equation = var_1666_equation_0, values = (var_1598_cast_fp16_9, var_1644_cast_fp16))[name = tensor<string, []>("op_1666_cast_fp16")]; |
| tensor<string, []> var_1668_equation_0 = const()[name = tensor<string, []>("op_1668_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1668_cast_fp16 = einsum(equation = var_1668_equation_0, values = (var_1598_cast_fp16_10, var_1645_cast_fp16))[name = tensor<string, []>("op_1668_cast_fp16")]; |
| tensor<string, []> var_1670_equation_0 = const()[name = tensor<string, []>("op_1670_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1670_cast_fp16 = einsum(equation = var_1670_equation_0, values = (var_1598_cast_fp16_11, var_1646_cast_fp16))[name = tensor<string, []>("op_1670_cast_fp16")]; |
| tensor<bool, []> input_75_interleave_0 = const()[name = tensor<string, []>("input_75_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_75_cast_fp16 = concat(axis = var_1519, interleave = input_75_interleave_0, values = (var_1648_cast_fp16, var_1650_cast_fp16, var_1652_cast_fp16, var_1654_cast_fp16, var_1656_cast_fp16, var_1658_cast_fp16, var_1660_cast_fp16, var_1662_cast_fp16, var_1664_cast_fp16, var_1666_cast_fp16, var_1668_cast_fp16, var_1670_cast_fp16))[name = tensor<string, []>("input_75_cast_fp16")]; |
| tensor<string, []> var_1679_pad_type_0 = const()[name = tensor<string, []>("op_1679_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1679_strides_0 = const()[name = tensor<string, []>("op_1679_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1679_pad_0 = const()[name = tensor<string, []>("op_1679_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1679_dilations_0 = const()[name = tensor<string, []>("op_1679_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1679_groups_0 = const()[name = tensor<string, []>("op_1679_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_7_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108986752)))]; |
| tensor<fp16, [768]> blocks_7_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110166464)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1679_cast_fp16 = conv(bias = blocks_7_attn_out_bias_to_fp16, dilations = var_1679_dilations_0, groups = var_1679_groups_0, pad = var_1679_pad_0, pad_type = var_1679_pad_type_0, strides = var_1679_strides_0, weight = blocks_7_attn_out_weight_to_fp16, x = input_75_cast_fp16)[name = tensor<string, []>("op_1679_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_31_cast_fp16 = add(x = inputs_29_cast_fp16, y = var_1679_cast_fp16)[name = tensor<string, []>("inputs_31_cast_fp16")]; |
| tensor<int32, [1]> input_77_axes_0 = const()[name = tensor<string, []>("input_77_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_77_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_77_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110168064)))]; |
| tensor<fp16, [768]> input_77_beta_0_to_fp16 = const()[name = tensor<string, []>("input_77_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110169664)))]; |
| tensor<fp16, []> var_1689_to_fp16 = const()[name = tensor<string, []>("op_1689_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_77_cast_fp16 = layer_norm(axes = input_77_axes_0, beta = input_77_beta_0_to_fp16, epsilon = var_1689_to_fp16, gamma = input_77_gamma_0_to_fp16, x = inputs_31_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")]; |
| tensor<string, []> input_79_pad_type_0 = const()[name = tensor<string, []>("input_79_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_79_strides_0 = const()[name = tensor<string, []>("input_79_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_79_pad_0 = const()[name = tensor<string, []>("input_79_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_79_dilations_0 = const()[name = tensor<string, []>("input_79_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_79_groups_0 = const()[name = tensor<string, []>("input_79_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_7_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110171264)))]; |
| tensor<fp16, [3072]> blocks_7_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114889920)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_79_cast_fp16 = conv(bias = blocks_7_mlp_0_bias_to_fp16, dilations = input_79_dilations_0, groups = input_79_groups_0, pad = input_79_pad_0, pad_type = input_79_pad_type_0, strides = input_79_strides_0, weight = blocks_7_mlp_0_weight_to_fp16, x = input_77_cast_fp16)[name = tensor<string, []>("input_79_cast_fp16")]; |
| tensor<string, []> input_81_mode_0 = const()[name = tensor<string, []>("input_81_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_81_cast_fp16 = gelu(mode = input_81_mode_0, x = input_79_cast_fp16)[name = tensor<string, []>("input_81_cast_fp16")]; |
| tensor<string, []> var_1715_pad_type_0 = const()[name = tensor<string, []>("op_1715_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1715_strides_0 = const()[name = tensor<string, []>("op_1715_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1715_pad_0 = const()[name = tensor<string, []>("op_1715_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1715_dilations_0 = const()[name = tensor<string, []>("op_1715_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1715_groups_0 = const()[name = tensor<string, []>("op_1715_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_7_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114896128)))]; |
| tensor<fp16, [768]> blocks_7_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119614784)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1715_cast_fp16 = conv(bias = blocks_7_mlp_2_bias_to_fp16, dilations = var_1715_dilations_0, groups = var_1715_groups_0, pad = var_1715_pad_0, pad_type = var_1715_pad_type_0, strides = var_1715_strides_0, weight = blocks_7_mlp_2_weight_to_fp16, x = input_81_cast_fp16)[name = tensor<string, []>("op_1715_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = var_1715_cast_fp16)[name = tensor<string, []>("inputs_33_cast_fp16")]; |
| tensor<int32, []> var_1724 = const()[name = tensor<string, []>("op_1724"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_83_axes_0 = const()[name = tensor<string, []>("input_83_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_83_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_83_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119616384)))]; |
| tensor<fp16, [768]> input_83_beta_0_to_fp16 = const()[name = tensor<string, []>("input_83_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119617984)))]; |
| tensor<fp16, []> var_1740_to_fp16 = const()[name = tensor<string, []>("op_1740_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_83_cast_fp16 = layer_norm(axes = input_83_axes_0, beta = input_83_beta_0_to_fp16, epsilon = var_1740_to_fp16, gamma = input_83_gamma_0_to_fp16, x = inputs_33_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")]; |
| tensor<string, []> q_17_pad_type_0 = const()[name = tensor<string, []>("q_17_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_17_strides_0 = const()[name = tensor<string, []>("q_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_17_pad_0 = const()[name = tensor<string, []>("q_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_17_dilations_0 = const()[name = tensor<string, []>("q_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_17_groups_0 = const()[name = tensor<string, []>("q_17_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_1775_weight_0_to_fp16 = const()[name = tensor<string, []>("op_1775_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119619584)))]; |
| tensor<fp16, [768]> var_1775_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1775_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120799296)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1775_cast_fp16 = conv(bias = var_1775_bias_0_to_fp16, dilations = q_17_dilations_0, groups = q_17_groups_0, pad = q_17_pad_0, pad_type = q_17_pad_type_0, strides = q_17_strides_0, weight = var_1775_weight_0_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("op_1775_cast_fp16")]; |
| tensor<string, []> k_17_pad_type_0 = const()[name = tensor<string, []>("k_17_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_17_strides_0 = const()[name = tensor<string, []>("k_17_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_17_pad_0 = const()[name = tensor<string, []>("k_17_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_17_dilations_0 = const()[name = tensor<string, []>("k_17_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_17_groups_0 = const()[name = tensor<string, []>("k_17_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_8_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120800896)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_17_cast_fp16 = conv(dilations = k_17_dilations_0, groups = k_17_groups_0, pad = k_17_pad_0, pad_type = k_17_pad_type_0, strides = k_17_strides_0, weight = blocks_8_attn_key_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("k_17_cast_fp16")]; |
| tensor<string, []> var_1773_pad_type_0 = const()[name = tensor<string, []>("op_1773_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1773_strides_0 = const()[name = tensor<string, []>("op_1773_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1773_pad_0 = const()[name = tensor<string, []>("op_1773_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1773_dilations_0 = const()[name = tensor<string, []>("op_1773_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1773_groups_0 = const()[name = tensor<string, []>("op_1773_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_8_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121980608)))]; |
| tensor<fp16, [768]> blocks_8_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123160320)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1773_cast_fp16 = conv(bias = blocks_8_attn_value_bias_to_fp16, dilations = var_1773_dilations_0, groups = var_1773_groups_0, pad = var_1773_pad_0, pad_type = var_1773_pad_type_0, strides = var_1773_strides_0, weight = blocks_8_attn_value_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("op_1773_cast_fp16")]; |
| tensor<int32, [12]> tile_24 = const()[name = tensor<string, []>("tile_24"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1776_axis_0 = const()[name = tensor<string, []>("op_1776_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1776_cast_fp16_11 = split(axis = var_1776_axis_0, split_sizes = tile_24, x = var_1775_cast_fp16)[name = tensor<string, []>("op_1776_cast_fp16")]; |
| tensor<int32, [4]> var_1789_perm_0 = const()[name = tensor<string, []>("op_1789_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_25 = const()[name = tensor<string, []>("tile_25"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1790_axis_0 = const()[name = tensor<string, []>("op_1790_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_1789_cast_fp16 = transpose(perm = var_1789_perm_0, x = k_17_cast_fp16)[name = tensor<string, []>("transpose_4")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_1790_cast_fp16_11 = split(axis = var_1790_axis_0, split_sizes = tile_25, x = var_1789_cast_fp16)[name = tensor<string, []>("op_1790_cast_fp16")]; |
| tensor<int32, [12]> tile_26 = const()[name = tensor<string, []>("tile_26"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1803_axis_0 = const()[name = tensor<string, []>("op_1803_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1803_cast_fp16_11 = split(axis = var_1803_axis_0, split_sizes = tile_26, x = var_1773_cast_fp16)[name = tensor<string, []>("op_1803_cast_fp16")]; |
| tensor<string, []> aw_193_equation_0 = const()[name = tensor<string, []>("aw_193_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_193_cast_fp16 = einsum(equation = aw_193_equation_0, values = (var_1790_cast_fp16_0, var_1776_cast_fp16_0))[name = tensor<string, []>("aw_193_cast_fp16")]; |
| tensor<string, []> aw_195_equation_0 = const()[name = tensor<string, []>("aw_195_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_195_cast_fp16 = einsum(equation = aw_195_equation_0, values = (var_1790_cast_fp16_1, var_1776_cast_fp16_1))[name = tensor<string, []>("aw_195_cast_fp16")]; |
| tensor<string, []> aw_197_equation_0 = const()[name = tensor<string, []>("aw_197_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_197_cast_fp16 = einsum(equation = aw_197_equation_0, values = (var_1790_cast_fp16_2, var_1776_cast_fp16_2))[name = tensor<string, []>("aw_197_cast_fp16")]; |
| tensor<string, []> aw_199_equation_0 = const()[name = tensor<string, []>("aw_199_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_199_cast_fp16 = einsum(equation = aw_199_equation_0, values = (var_1790_cast_fp16_3, var_1776_cast_fp16_3))[name = tensor<string, []>("aw_199_cast_fp16")]; |
| tensor<string, []> aw_201_equation_0 = const()[name = tensor<string, []>("aw_201_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_201_cast_fp16 = einsum(equation = aw_201_equation_0, values = (var_1790_cast_fp16_4, var_1776_cast_fp16_4))[name = tensor<string, []>("aw_201_cast_fp16")]; |
| tensor<string, []> aw_203_equation_0 = const()[name = tensor<string, []>("aw_203_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_203_cast_fp16 = einsum(equation = aw_203_equation_0, values = (var_1790_cast_fp16_5, var_1776_cast_fp16_5))[name = tensor<string, []>("aw_203_cast_fp16")]; |
| tensor<string, []> aw_205_equation_0 = const()[name = tensor<string, []>("aw_205_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_205_cast_fp16 = einsum(equation = aw_205_equation_0, values = (var_1790_cast_fp16_6, var_1776_cast_fp16_6))[name = tensor<string, []>("aw_205_cast_fp16")]; |
| tensor<string, []> aw_207_equation_0 = const()[name = tensor<string, []>("aw_207_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_207_cast_fp16 = einsum(equation = aw_207_equation_0, values = (var_1790_cast_fp16_7, var_1776_cast_fp16_7))[name = tensor<string, []>("aw_207_cast_fp16")]; |
| tensor<string, []> aw_209_equation_0 = const()[name = tensor<string, []>("aw_209_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_209_cast_fp16 = einsum(equation = aw_209_equation_0, values = (var_1790_cast_fp16_8, var_1776_cast_fp16_8))[name = tensor<string, []>("aw_209_cast_fp16")]; |
| tensor<string, []> aw_211_equation_0 = const()[name = tensor<string, []>("aw_211_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_211_cast_fp16 = einsum(equation = aw_211_equation_0, values = (var_1790_cast_fp16_9, var_1776_cast_fp16_9))[name = tensor<string, []>("aw_211_cast_fp16")]; |
| tensor<string, []> aw_213_equation_0 = const()[name = tensor<string, []>("aw_213_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_213_cast_fp16 = einsum(equation = aw_213_equation_0, values = (var_1790_cast_fp16_10, var_1776_cast_fp16_10))[name = tensor<string, []>("aw_213_cast_fp16")]; |
| tensor<string, []> aw_215_equation_0 = const()[name = tensor<string, []>("aw_215_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_215_cast_fp16 = einsum(equation = aw_215_equation_0, values = (var_1790_cast_fp16_11, var_1776_cast_fp16_11))[name = tensor<string, []>("aw_215_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1840_cast_fp16 = softmax(axis = var_1724, x = aw_193_cast_fp16)[name = tensor<string, []>("op_1840_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1841_cast_fp16 = softmax(axis = var_1724, x = aw_195_cast_fp16)[name = tensor<string, []>("op_1841_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1842_cast_fp16 = softmax(axis = var_1724, x = aw_197_cast_fp16)[name = tensor<string, []>("op_1842_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1843_cast_fp16 = softmax(axis = var_1724, x = aw_199_cast_fp16)[name = tensor<string, []>("op_1843_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1844_cast_fp16 = softmax(axis = var_1724, x = aw_201_cast_fp16)[name = tensor<string, []>("op_1844_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1845_cast_fp16 = softmax(axis = var_1724, x = aw_203_cast_fp16)[name = tensor<string, []>("op_1845_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1846_cast_fp16 = softmax(axis = var_1724, x = aw_205_cast_fp16)[name = tensor<string, []>("op_1846_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1847_cast_fp16 = softmax(axis = var_1724, x = aw_207_cast_fp16)[name = tensor<string, []>("op_1847_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1848_cast_fp16 = softmax(axis = var_1724, x = aw_209_cast_fp16)[name = tensor<string, []>("op_1848_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1849_cast_fp16 = softmax(axis = var_1724, x = aw_211_cast_fp16)[name = tensor<string, []>("op_1849_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1850_cast_fp16 = softmax(axis = var_1724, x = aw_213_cast_fp16)[name = tensor<string, []>("op_1850_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_1851_cast_fp16 = softmax(axis = var_1724, x = aw_215_cast_fp16)[name = tensor<string, []>("op_1851_cast_fp16")]; |
| tensor<string, []> var_1853_equation_0 = const()[name = tensor<string, []>("op_1853_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1853_cast_fp16 = einsum(equation = var_1853_equation_0, values = (var_1803_cast_fp16_0, var_1840_cast_fp16))[name = tensor<string, []>("op_1853_cast_fp16")]; |
| tensor<string, []> var_1855_equation_0 = const()[name = tensor<string, []>("op_1855_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1855_cast_fp16 = einsum(equation = var_1855_equation_0, values = (var_1803_cast_fp16_1, var_1841_cast_fp16))[name = tensor<string, []>("op_1855_cast_fp16")]; |
| tensor<string, []> var_1857_equation_0 = const()[name = tensor<string, []>("op_1857_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1857_cast_fp16 = einsum(equation = var_1857_equation_0, values = (var_1803_cast_fp16_2, var_1842_cast_fp16))[name = tensor<string, []>("op_1857_cast_fp16")]; |
| tensor<string, []> var_1859_equation_0 = const()[name = tensor<string, []>("op_1859_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1859_cast_fp16 = einsum(equation = var_1859_equation_0, values = (var_1803_cast_fp16_3, var_1843_cast_fp16))[name = tensor<string, []>("op_1859_cast_fp16")]; |
| tensor<string, []> var_1861_equation_0 = const()[name = tensor<string, []>("op_1861_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1861_cast_fp16 = einsum(equation = var_1861_equation_0, values = (var_1803_cast_fp16_4, var_1844_cast_fp16))[name = tensor<string, []>("op_1861_cast_fp16")]; |
| tensor<string, []> var_1863_equation_0 = const()[name = tensor<string, []>("op_1863_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1863_cast_fp16 = einsum(equation = var_1863_equation_0, values = (var_1803_cast_fp16_5, var_1845_cast_fp16))[name = tensor<string, []>("op_1863_cast_fp16")]; |
| tensor<string, []> var_1865_equation_0 = const()[name = tensor<string, []>("op_1865_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1865_cast_fp16 = einsum(equation = var_1865_equation_0, values = (var_1803_cast_fp16_6, var_1846_cast_fp16))[name = tensor<string, []>("op_1865_cast_fp16")]; |
| tensor<string, []> var_1867_equation_0 = const()[name = tensor<string, []>("op_1867_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1867_cast_fp16 = einsum(equation = var_1867_equation_0, values = (var_1803_cast_fp16_7, var_1847_cast_fp16))[name = tensor<string, []>("op_1867_cast_fp16")]; |
| tensor<string, []> var_1869_equation_0 = const()[name = tensor<string, []>("op_1869_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1869_cast_fp16 = einsum(equation = var_1869_equation_0, values = (var_1803_cast_fp16_8, var_1848_cast_fp16))[name = tensor<string, []>("op_1869_cast_fp16")]; |
| tensor<string, []> var_1871_equation_0 = const()[name = tensor<string, []>("op_1871_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1871_cast_fp16 = einsum(equation = var_1871_equation_0, values = (var_1803_cast_fp16_9, var_1849_cast_fp16))[name = tensor<string, []>("op_1871_cast_fp16")]; |
| tensor<string, []> var_1873_equation_0 = const()[name = tensor<string, []>("op_1873_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1873_cast_fp16 = einsum(equation = var_1873_equation_0, values = (var_1803_cast_fp16_10, var_1850_cast_fp16))[name = tensor<string, []>("op_1873_cast_fp16")]; |
| tensor<string, []> var_1875_equation_0 = const()[name = tensor<string, []>("op_1875_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1875_cast_fp16 = einsum(equation = var_1875_equation_0, values = (var_1803_cast_fp16_11, var_1851_cast_fp16))[name = tensor<string, []>("op_1875_cast_fp16")]; |
| tensor<bool, []> input_85_interleave_0 = const()[name = tensor<string, []>("input_85_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_85_cast_fp16 = concat(axis = var_1724, interleave = input_85_interleave_0, values = (var_1853_cast_fp16, var_1855_cast_fp16, var_1857_cast_fp16, var_1859_cast_fp16, var_1861_cast_fp16, var_1863_cast_fp16, var_1865_cast_fp16, var_1867_cast_fp16, var_1869_cast_fp16, var_1871_cast_fp16, var_1873_cast_fp16, var_1875_cast_fp16))[name = tensor<string, []>("input_85_cast_fp16")]; |
| tensor<string, []> var_1884_pad_type_0 = const()[name = tensor<string, []>("op_1884_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1884_strides_0 = const()[name = tensor<string, []>("op_1884_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1884_pad_0 = const()[name = tensor<string, []>("op_1884_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1884_dilations_0 = const()[name = tensor<string, []>("op_1884_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1884_groups_0 = const()[name = tensor<string, []>("op_1884_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_8_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123161920)))]; |
| tensor<fp16, [768]> blocks_8_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124341632)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1884_cast_fp16 = conv(bias = blocks_8_attn_out_bias_to_fp16, dilations = var_1884_dilations_0, groups = var_1884_groups_0, pad = var_1884_pad_0, pad_type = var_1884_pad_type_0, strides = var_1884_strides_0, weight = blocks_8_attn_out_weight_to_fp16, x = input_85_cast_fp16)[name = tensor<string, []>("op_1884_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = var_1884_cast_fp16)[name = tensor<string, []>("inputs_35_cast_fp16")]; |
| tensor<int32, [1]> input_87_axes_0 = const()[name = tensor<string, []>("input_87_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_87_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_87_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124343232)))]; |
| tensor<fp16, [768]> input_87_beta_0_to_fp16 = const()[name = tensor<string, []>("input_87_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124344832)))]; |
| tensor<fp16, []> var_1894_to_fp16 = const()[name = tensor<string, []>("op_1894_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_87_cast_fp16 = layer_norm(axes = input_87_axes_0, beta = input_87_beta_0_to_fp16, epsilon = var_1894_to_fp16, gamma = input_87_gamma_0_to_fp16, x = inputs_35_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")]; |
| tensor<string, []> input_89_pad_type_0 = const()[name = tensor<string, []>("input_89_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_89_strides_0 = const()[name = tensor<string, []>("input_89_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_89_pad_0 = const()[name = tensor<string, []>("input_89_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_89_dilations_0 = const()[name = tensor<string, []>("input_89_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_89_groups_0 = const()[name = tensor<string, []>("input_89_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_8_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124346432)))]; |
| tensor<fp16, [3072]> blocks_8_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129065088)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_89_cast_fp16 = conv(bias = blocks_8_mlp_0_bias_to_fp16, dilations = input_89_dilations_0, groups = input_89_groups_0, pad = input_89_pad_0, pad_type = input_89_pad_type_0, strides = input_89_strides_0, weight = blocks_8_mlp_0_weight_to_fp16, x = input_87_cast_fp16)[name = tensor<string, []>("input_89_cast_fp16")]; |
| tensor<string, []> input_91_mode_0 = const()[name = tensor<string, []>("input_91_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_91_cast_fp16 = gelu(mode = input_91_mode_0, x = input_89_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")]; |
| tensor<string, []> var_1920_pad_type_0 = const()[name = tensor<string, []>("op_1920_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1920_strides_0 = const()[name = tensor<string, []>("op_1920_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1920_pad_0 = const()[name = tensor<string, []>("op_1920_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1920_dilations_0 = const()[name = tensor<string, []>("op_1920_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1920_groups_0 = const()[name = tensor<string, []>("op_1920_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_8_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129071296)))]; |
| tensor<fp16, [768]> blocks_8_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133789952)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1920_cast_fp16 = conv(bias = blocks_8_mlp_2_bias_to_fp16, dilations = var_1920_dilations_0, groups = var_1920_groups_0, pad = var_1920_pad_0, pad_type = var_1920_pad_type_0, strides = var_1920_strides_0, weight = blocks_8_mlp_2_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("op_1920_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = var_1920_cast_fp16)[name = tensor<string, []>("inputs_37_cast_fp16")]; |
| tensor<int32, []> var_1929 = const()[name = tensor<string, []>("op_1929"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_93_axes_0 = const()[name = tensor<string, []>("input_93_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_93_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_93_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133791552)))]; |
| tensor<fp16, [768]> input_93_beta_0_to_fp16 = const()[name = tensor<string, []>("input_93_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133793152)))]; |
| tensor<fp16, []> var_1945_to_fp16 = const()[name = tensor<string, []>("op_1945_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_93_cast_fp16 = layer_norm(axes = input_93_axes_0, beta = input_93_beta_0_to_fp16, epsilon = var_1945_to_fp16, gamma = input_93_gamma_0_to_fp16, x = inputs_37_cast_fp16)[name = tensor<string, []>("input_93_cast_fp16")]; |
| tensor<string, []> q_19_pad_type_0 = const()[name = tensor<string, []>("q_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_19_strides_0 = const()[name = tensor<string, []>("q_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_19_pad_0 = const()[name = tensor<string, []>("q_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_19_dilations_0 = const()[name = tensor<string, []>("q_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_19_groups_0 = const()[name = tensor<string, []>("q_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_1980_weight_0_to_fp16 = const()[name = tensor<string, []>("op_1980_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133794752)))]; |
| tensor<fp16, [768]> var_1980_bias_0_to_fp16 = const()[name = tensor<string, []>("op_1980_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134974464)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1980_cast_fp16 = conv(bias = var_1980_bias_0_to_fp16, dilations = q_19_dilations_0, groups = q_19_groups_0, pad = q_19_pad_0, pad_type = q_19_pad_type_0, strides = q_19_strides_0, weight = var_1980_weight_0_to_fp16, x = input_93_cast_fp16)[name = tensor<string, []>("op_1980_cast_fp16")]; |
| tensor<string, []> k_19_pad_type_0 = const()[name = tensor<string, []>("k_19_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_19_strides_0 = const()[name = tensor<string, []>("k_19_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_19_pad_0 = const()[name = tensor<string, []>("k_19_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_19_dilations_0 = const()[name = tensor<string, []>("k_19_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_19_groups_0 = const()[name = tensor<string, []>("k_19_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_9_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134976064)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_19_cast_fp16 = conv(dilations = k_19_dilations_0, groups = k_19_groups_0, pad = k_19_pad_0, pad_type = k_19_pad_type_0, strides = k_19_strides_0, weight = blocks_9_attn_key_weight_to_fp16, x = input_93_cast_fp16)[name = tensor<string, []>("k_19_cast_fp16")]; |
| tensor<string, []> var_1978_pad_type_0 = const()[name = tensor<string, []>("op_1978_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_1978_strides_0 = const()[name = tensor<string, []>("op_1978_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_1978_pad_0 = const()[name = tensor<string, []>("op_1978_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_1978_dilations_0 = const()[name = tensor<string, []>("op_1978_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_1978_groups_0 = const()[name = tensor<string, []>("op_1978_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_9_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136155776)))]; |
| tensor<fp16, [768]> blocks_9_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137335488)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_1978_cast_fp16 = conv(bias = blocks_9_attn_value_bias_to_fp16, dilations = var_1978_dilations_0, groups = var_1978_groups_0, pad = var_1978_pad_0, pad_type = var_1978_pad_type_0, strides = var_1978_strides_0, weight = blocks_9_attn_value_weight_to_fp16, x = input_93_cast_fp16)[name = tensor<string, []>("op_1978_cast_fp16")]; |
| tensor<int32, [12]> tile_27 = const()[name = tensor<string, []>("tile_27"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1981_axis_0 = const()[name = tensor<string, []>("op_1981_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_1981_cast_fp16_11 = split(axis = var_1981_axis_0, split_sizes = tile_27, x = var_1980_cast_fp16)[name = tensor<string, []>("op_1981_cast_fp16")]; |
| tensor<int32, [4]> var_1994_perm_0 = const()[name = tensor<string, []>("op_1994_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_28 = const()[name = tensor<string, []>("tile_28"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_1995_axis_0 = const()[name = tensor<string, []>("op_1995_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_1994_cast_fp16 = transpose(perm = var_1994_perm_0, x = k_19_cast_fp16)[name = tensor<string, []>("transpose_3")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_1995_cast_fp16_11 = split(axis = var_1995_axis_0, split_sizes = tile_28, x = var_1994_cast_fp16)[name = tensor<string, []>("op_1995_cast_fp16")]; |
| tensor<int32, [12]> tile_29 = const()[name = tensor<string, []>("tile_29"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_2008_axis_0 = const()[name = tensor<string, []>("op_2008_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_2008_cast_fp16_11 = split(axis = var_2008_axis_0, split_sizes = tile_29, x = var_1978_cast_fp16)[name = tensor<string, []>("op_2008_cast_fp16")]; |
| tensor<string, []> aw_217_equation_0 = const()[name = tensor<string, []>("aw_217_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_217_cast_fp16 = einsum(equation = aw_217_equation_0, values = (var_1995_cast_fp16_0, var_1981_cast_fp16_0))[name = tensor<string, []>("aw_217_cast_fp16")]; |
| tensor<string, []> aw_219_equation_0 = const()[name = tensor<string, []>("aw_219_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_219_cast_fp16 = einsum(equation = aw_219_equation_0, values = (var_1995_cast_fp16_1, var_1981_cast_fp16_1))[name = tensor<string, []>("aw_219_cast_fp16")]; |
| tensor<string, []> aw_221_equation_0 = const()[name = tensor<string, []>("aw_221_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_221_cast_fp16 = einsum(equation = aw_221_equation_0, values = (var_1995_cast_fp16_2, var_1981_cast_fp16_2))[name = tensor<string, []>("aw_221_cast_fp16")]; |
| tensor<string, []> aw_223_equation_0 = const()[name = tensor<string, []>("aw_223_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_223_cast_fp16 = einsum(equation = aw_223_equation_0, values = (var_1995_cast_fp16_3, var_1981_cast_fp16_3))[name = tensor<string, []>("aw_223_cast_fp16")]; |
| tensor<string, []> aw_225_equation_0 = const()[name = tensor<string, []>("aw_225_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_225_cast_fp16 = einsum(equation = aw_225_equation_0, values = (var_1995_cast_fp16_4, var_1981_cast_fp16_4))[name = tensor<string, []>("aw_225_cast_fp16")]; |
| tensor<string, []> aw_227_equation_0 = const()[name = tensor<string, []>("aw_227_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_227_cast_fp16 = einsum(equation = aw_227_equation_0, values = (var_1995_cast_fp16_5, var_1981_cast_fp16_5))[name = tensor<string, []>("aw_227_cast_fp16")]; |
| tensor<string, []> aw_229_equation_0 = const()[name = tensor<string, []>("aw_229_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_229_cast_fp16 = einsum(equation = aw_229_equation_0, values = (var_1995_cast_fp16_6, var_1981_cast_fp16_6))[name = tensor<string, []>("aw_229_cast_fp16")]; |
| tensor<string, []> aw_231_equation_0 = const()[name = tensor<string, []>("aw_231_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_231_cast_fp16 = einsum(equation = aw_231_equation_0, values = (var_1995_cast_fp16_7, var_1981_cast_fp16_7))[name = tensor<string, []>("aw_231_cast_fp16")]; |
| tensor<string, []> aw_233_equation_0 = const()[name = tensor<string, []>("aw_233_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_233_cast_fp16 = einsum(equation = aw_233_equation_0, values = (var_1995_cast_fp16_8, var_1981_cast_fp16_8))[name = tensor<string, []>("aw_233_cast_fp16")]; |
| tensor<string, []> aw_235_equation_0 = const()[name = tensor<string, []>("aw_235_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_235_cast_fp16 = einsum(equation = aw_235_equation_0, values = (var_1995_cast_fp16_9, var_1981_cast_fp16_9))[name = tensor<string, []>("aw_235_cast_fp16")]; |
| tensor<string, []> aw_237_equation_0 = const()[name = tensor<string, []>("aw_237_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_237_cast_fp16 = einsum(equation = aw_237_equation_0, values = (var_1995_cast_fp16_10, var_1981_cast_fp16_10))[name = tensor<string, []>("aw_237_cast_fp16")]; |
| tensor<string, []> aw_239_equation_0 = const()[name = tensor<string, []>("aw_239_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_239_cast_fp16 = einsum(equation = aw_239_equation_0, values = (var_1995_cast_fp16_11, var_1981_cast_fp16_11))[name = tensor<string, []>("aw_239_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2045_cast_fp16 = softmax(axis = var_1929, x = aw_217_cast_fp16)[name = tensor<string, []>("op_2045_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2046_cast_fp16 = softmax(axis = var_1929, x = aw_219_cast_fp16)[name = tensor<string, []>("op_2046_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2047_cast_fp16 = softmax(axis = var_1929, x = aw_221_cast_fp16)[name = tensor<string, []>("op_2047_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2048_cast_fp16 = softmax(axis = var_1929, x = aw_223_cast_fp16)[name = tensor<string, []>("op_2048_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2049_cast_fp16 = softmax(axis = var_1929, x = aw_225_cast_fp16)[name = tensor<string, []>("op_2049_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2050_cast_fp16 = softmax(axis = var_1929, x = aw_227_cast_fp16)[name = tensor<string, []>("op_2050_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2051_cast_fp16 = softmax(axis = var_1929, x = aw_229_cast_fp16)[name = tensor<string, []>("op_2051_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2052_cast_fp16 = softmax(axis = var_1929, x = aw_231_cast_fp16)[name = tensor<string, []>("op_2052_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2053_cast_fp16 = softmax(axis = var_1929, x = aw_233_cast_fp16)[name = tensor<string, []>("op_2053_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2054_cast_fp16 = softmax(axis = var_1929, x = aw_235_cast_fp16)[name = tensor<string, []>("op_2054_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2055_cast_fp16 = softmax(axis = var_1929, x = aw_237_cast_fp16)[name = tensor<string, []>("op_2055_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2056_cast_fp16 = softmax(axis = var_1929, x = aw_239_cast_fp16)[name = tensor<string, []>("op_2056_cast_fp16")]; |
| tensor<string, []> var_2058_equation_0 = const()[name = tensor<string, []>("op_2058_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2058_cast_fp16 = einsum(equation = var_2058_equation_0, values = (var_2008_cast_fp16_0, var_2045_cast_fp16))[name = tensor<string, []>("op_2058_cast_fp16")]; |
| tensor<string, []> var_2060_equation_0 = const()[name = tensor<string, []>("op_2060_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2060_cast_fp16 = einsum(equation = var_2060_equation_0, values = (var_2008_cast_fp16_1, var_2046_cast_fp16))[name = tensor<string, []>("op_2060_cast_fp16")]; |
| tensor<string, []> var_2062_equation_0 = const()[name = tensor<string, []>("op_2062_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2062_cast_fp16 = einsum(equation = var_2062_equation_0, values = (var_2008_cast_fp16_2, var_2047_cast_fp16))[name = tensor<string, []>("op_2062_cast_fp16")]; |
| tensor<string, []> var_2064_equation_0 = const()[name = tensor<string, []>("op_2064_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2064_cast_fp16 = einsum(equation = var_2064_equation_0, values = (var_2008_cast_fp16_3, var_2048_cast_fp16))[name = tensor<string, []>("op_2064_cast_fp16")]; |
| tensor<string, []> var_2066_equation_0 = const()[name = tensor<string, []>("op_2066_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2066_cast_fp16 = einsum(equation = var_2066_equation_0, values = (var_2008_cast_fp16_4, var_2049_cast_fp16))[name = tensor<string, []>("op_2066_cast_fp16")]; |
| tensor<string, []> var_2068_equation_0 = const()[name = tensor<string, []>("op_2068_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2068_cast_fp16 = einsum(equation = var_2068_equation_0, values = (var_2008_cast_fp16_5, var_2050_cast_fp16))[name = tensor<string, []>("op_2068_cast_fp16")]; |
| tensor<string, []> var_2070_equation_0 = const()[name = tensor<string, []>("op_2070_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2070_cast_fp16 = einsum(equation = var_2070_equation_0, values = (var_2008_cast_fp16_6, var_2051_cast_fp16))[name = tensor<string, []>("op_2070_cast_fp16")]; |
| tensor<string, []> var_2072_equation_0 = const()[name = tensor<string, []>("op_2072_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2072_cast_fp16 = einsum(equation = var_2072_equation_0, values = (var_2008_cast_fp16_7, var_2052_cast_fp16))[name = tensor<string, []>("op_2072_cast_fp16")]; |
| tensor<string, []> var_2074_equation_0 = const()[name = tensor<string, []>("op_2074_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2074_cast_fp16 = einsum(equation = var_2074_equation_0, values = (var_2008_cast_fp16_8, var_2053_cast_fp16))[name = tensor<string, []>("op_2074_cast_fp16")]; |
| tensor<string, []> var_2076_equation_0 = const()[name = tensor<string, []>("op_2076_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2076_cast_fp16 = einsum(equation = var_2076_equation_0, values = (var_2008_cast_fp16_9, var_2054_cast_fp16))[name = tensor<string, []>("op_2076_cast_fp16")]; |
| tensor<string, []> var_2078_equation_0 = const()[name = tensor<string, []>("op_2078_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2078_cast_fp16 = einsum(equation = var_2078_equation_0, values = (var_2008_cast_fp16_10, var_2055_cast_fp16))[name = tensor<string, []>("op_2078_cast_fp16")]; |
| tensor<string, []> var_2080_equation_0 = const()[name = tensor<string, []>("op_2080_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2080_cast_fp16 = einsum(equation = var_2080_equation_0, values = (var_2008_cast_fp16_11, var_2056_cast_fp16))[name = tensor<string, []>("op_2080_cast_fp16")]; |
| tensor<bool, []> input_95_interleave_0 = const()[name = tensor<string, []>("input_95_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_95_cast_fp16 = concat(axis = var_1929, interleave = input_95_interleave_0, values = (var_2058_cast_fp16, var_2060_cast_fp16, var_2062_cast_fp16, var_2064_cast_fp16, var_2066_cast_fp16, var_2068_cast_fp16, var_2070_cast_fp16, var_2072_cast_fp16, var_2074_cast_fp16, var_2076_cast_fp16, var_2078_cast_fp16, var_2080_cast_fp16))[name = tensor<string, []>("input_95_cast_fp16")]; |
| tensor<string, []> var_2089_pad_type_0 = const()[name = tensor<string, []>("op_2089_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_2089_strides_0 = const()[name = tensor<string, []>("op_2089_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_2089_pad_0 = const()[name = tensor<string, []>("op_2089_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_2089_dilations_0 = const()[name = tensor<string, []>("op_2089_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_2089_groups_0 = const()[name = tensor<string, []>("op_2089_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_9_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137337088)))]; |
| tensor<fp16, [768]> blocks_9_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138516800)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2089_cast_fp16 = conv(bias = blocks_9_attn_out_bias_to_fp16, dilations = var_2089_dilations_0, groups = var_2089_groups_0, pad = var_2089_pad_0, pad_type = var_2089_pad_type_0, strides = var_2089_strides_0, weight = blocks_9_attn_out_weight_to_fp16, x = input_95_cast_fp16)[name = tensor<string, []>("op_2089_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = var_2089_cast_fp16)[name = tensor<string, []>("inputs_39_cast_fp16")]; |
| tensor<int32, [1]> input_97_axes_0 = const()[name = tensor<string, []>("input_97_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_97_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_97_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138518400)))]; |
| tensor<fp16, [768]> input_97_beta_0_to_fp16 = const()[name = tensor<string, []>("input_97_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138520000)))]; |
| tensor<fp16, []> var_2099_to_fp16 = const()[name = tensor<string, []>("op_2099_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_97_cast_fp16 = layer_norm(axes = input_97_axes_0, beta = input_97_beta_0_to_fp16, epsilon = var_2099_to_fp16, gamma = input_97_gamma_0_to_fp16, x = inputs_39_cast_fp16)[name = tensor<string, []>("input_97_cast_fp16")]; |
| tensor<string, []> input_99_pad_type_0 = const()[name = tensor<string, []>("input_99_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_99_strides_0 = const()[name = tensor<string, []>("input_99_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_99_pad_0 = const()[name = tensor<string, []>("input_99_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_99_dilations_0 = const()[name = tensor<string, []>("input_99_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_99_groups_0 = const()[name = tensor<string, []>("input_99_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_9_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138521600)))]; |
| tensor<fp16, [3072]> blocks_9_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143240256)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_99_cast_fp16 = conv(bias = blocks_9_mlp_0_bias_to_fp16, dilations = input_99_dilations_0, groups = input_99_groups_0, pad = input_99_pad_0, pad_type = input_99_pad_type_0, strides = input_99_strides_0, weight = blocks_9_mlp_0_weight_to_fp16, x = input_97_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")]; |
| tensor<string, []> input_101_mode_0 = const()[name = tensor<string, []>("input_101_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_101_cast_fp16 = gelu(mode = input_101_mode_0, x = input_99_cast_fp16)[name = tensor<string, []>("input_101_cast_fp16")]; |
| tensor<string, []> var_2125_pad_type_0 = const()[name = tensor<string, []>("op_2125_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_2125_strides_0 = const()[name = tensor<string, []>("op_2125_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_2125_pad_0 = const()[name = tensor<string, []>("op_2125_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_2125_dilations_0 = const()[name = tensor<string, []>("op_2125_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_2125_groups_0 = const()[name = tensor<string, []>("op_2125_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_9_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143246464)))]; |
| tensor<fp16, [768]> blocks_9_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147965120)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2125_cast_fp16 = conv(bias = blocks_9_mlp_2_bias_to_fp16, dilations = var_2125_dilations_0, groups = var_2125_groups_0, pad = var_2125_pad_0, pad_type = var_2125_pad_type_0, strides = var_2125_strides_0, weight = blocks_9_mlp_2_weight_to_fp16, x = input_101_cast_fp16)[name = tensor<string, []>("op_2125_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_41_cast_fp16 = add(x = inputs_39_cast_fp16, y = var_2125_cast_fp16)[name = tensor<string, []>("inputs_41_cast_fp16")]; |
| tensor<int32, []> var_2134 = const()[name = tensor<string, []>("op_2134"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_103_axes_0 = const()[name = tensor<string, []>("input_103_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_103_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_103_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147966720)))]; |
| tensor<fp16, [768]> input_103_beta_0_to_fp16 = const()[name = tensor<string, []>("input_103_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147968320)))]; |
| tensor<fp16, []> var_2150_to_fp16 = const()[name = tensor<string, []>("op_2150_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_103_cast_fp16 = layer_norm(axes = input_103_axes_0, beta = input_103_beta_0_to_fp16, epsilon = var_2150_to_fp16, gamma = input_103_gamma_0_to_fp16, x = inputs_41_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")]; |
| tensor<string, []> q_21_pad_type_0 = const()[name = tensor<string, []>("q_21_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_21_strides_0 = const()[name = tensor<string, []>("q_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_21_pad_0 = const()[name = tensor<string, []>("q_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_21_dilations_0 = const()[name = tensor<string, []>("q_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_21_groups_0 = const()[name = tensor<string, []>("q_21_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_2185_weight_0_to_fp16 = const()[name = tensor<string, []>("op_2185_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147969920)))]; |
| tensor<fp16, [768]> var_2185_bias_0_to_fp16 = const()[name = tensor<string, []>("op_2185_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149149632)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2185_cast_fp16 = conv(bias = var_2185_bias_0_to_fp16, dilations = q_21_dilations_0, groups = q_21_groups_0, pad = q_21_pad_0, pad_type = q_21_pad_type_0, strides = q_21_strides_0, weight = var_2185_weight_0_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("op_2185_cast_fp16")]; |
| tensor<string, []> k_21_pad_type_0 = const()[name = tensor<string, []>("k_21_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_21_strides_0 = const()[name = tensor<string, []>("k_21_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_21_pad_0 = const()[name = tensor<string, []>("k_21_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_21_dilations_0 = const()[name = tensor<string, []>("k_21_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_21_groups_0 = const()[name = tensor<string, []>("k_21_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_10_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149151232)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_21_cast_fp16 = conv(dilations = k_21_dilations_0, groups = k_21_groups_0, pad = k_21_pad_0, pad_type = k_21_pad_type_0, strides = k_21_strides_0, weight = blocks_10_attn_key_weight_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("k_21_cast_fp16")]; |
| tensor<string, []> var_2183_pad_type_0 = const()[name = tensor<string, []>("op_2183_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_2183_strides_0 = const()[name = tensor<string, []>("op_2183_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_2183_pad_0 = const()[name = tensor<string, []>("op_2183_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_2183_dilations_0 = const()[name = tensor<string, []>("op_2183_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_2183_groups_0 = const()[name = tensor<string, []>("op_2183_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_10_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150330944)))]; |
| tensor<fp16, [768]> blocks_10_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151510656)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2183_cast_fp16 = conv(bias = blocks_10_attn_value_bias_to_fp16, dilations = var_2183_dilations_0, groups = var_2183_groups_0, pad = var_2183_pad_0, pad_type = var_2183_pad_type_0, strides = var_2183_strides_0, weight = blocks_10_attn_value_weight_to_fp16, x = input_103_cast_fp16)[name = tensor<string, []>("op_2183_cast_fp16")]; |
| tensor<int32, [12]> tile_30 = const()[name = tensor<string, []>("tile_30"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_2186_axis_0 = const()[name = tensor<string, []>("op_2186_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_2186_cast_fp16_11 = split(axis = var_2186_axis_0, split_sizes = tile_30, x = var_2185_cast_fp16)[name = tensor<string, []>("op_2186_cast_fp16")]; |
| tensor<int32, [4]> var_2199_perm_0 = const()[name = tensor<string, []>("op_2199_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_31 = const()[name = tensor<string, []>("tile_31"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_2200_axis_0 = const()[name = tensor<string, []>("op_2200_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_2199_cast_fp16 = transpose(perm = var_2199_perm_0, x = k_21_cast_fp16)[name = tensor<string, []>("transpose_2")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_2200_cast_fp16_11 = split(axis = var_2200_axis_0, split_sizes = tile_31, x = var_2199_cast_fp16)[name = tensor<string, []>("op_2200_cast_fp16")]; |
| tensor<int32, [12]> tile_32 = const()[name = tensor<string, []>("tile_32"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_2213_axis_0 = const()[name = tensor<string, []>("op_2213_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_2213_cast_fp16_11 = split(axis = var_2213_axis_0, split_sizes = tile_32, x = var_2183_cast_fp16)[name = tensor<string, []>("op_2213_cast_fp16")]; |
| tensor<string, []> aw_241_equation_0 = const()[name = tensor<string, []>("aw_241_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_241_cast_fp16 = einsum(equation = aw_241_equation_0, values = (var_2200_cast_fp16_0, var_2186_cast_fp16_0))[name = tensor<string, []>("aw_241_cast_fp16")]; |
| tensor<string, []> aw_243_equation_0 = const()[name = tensor<string, []>("aw_243_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_243_cast_fp16 = einsum(equation = aw_243_equation_0, values = (var_2200_cast_fp16_1, var_2186_cast_fp16_1))[name = tensor<string, []>("aw_243_cast_fp16")]; |
| tensor<string, []> aw_245_equation_0 = const()[name = tensor<string, []>("aw_245_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_245_cast_fp16 = einsum(equation = aw_245_equation_0, values = (var_2200_cast_fp16_2, var_2186_cast_fp16_2))[name = tensor<string, []>("aw_245_cast_fp16")]; |
| tensor<string, []> aw_247_equation_0 = const()[name = tensor<string, []>("aw_247_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_247_cast_fp16 = einsum(equation = aw_247_equation_0, values = (var_2200_cast_fp16_3, var_2186_cast_fp16_3))[name = tensor<string, []>("aw_247_cast_fp16")]; |
| tensor<string, []> aw_249_equation_0 = const()[name = tensor<string, []>("aw_249_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_249_cast_fp16 = einsum(equation = aw_249_equation_0, values = (var_2200_cast_fp16_4, var_2186_cast_fp16_4))[name = tensor<string, []>("aw_249_cast_fp16")]; |
| tensor<string, []> aw_251_equation_0 = const()[name = tensor<string, []>("aw_251_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_251_cast_fp16 = einsum(equation = aw_251_equation_0, values = (var_2200_cast_fp16_5, var_2186_cast_fp16_5))[name = tensor<string, []>("aw_251_cast_fp16")]; |
| tensor<string, []> aw_253_equation_0 = const()[name = tensor<string, []>("aw_253_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_253_cast_fp16 = einsum(equation = aw_253_equation_0, values = (var_2200_cast_fp16_6, var_2186_cast_fp16_6))[name = tensor<string, []>("aw_253_cast_fp16")]; |
| tensor<string, []> aw_255_equation_0 = const()[name = tensor<string, []>("aw_255_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_255_cast_fp16 = einsum(equation = aw_255_equation_0, values = (var_2200_cast_fp16_7, var_2186_cast_fp16_7))[name = tensor<string, []>("aw_255_cast_fp16")]; |
| tensor<string, []> aw_257_equation_0 = const()[name = tensor<string, []>("aw_257_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_257_cast_fp16 = einsum(equation = aw_257_equation_0, values = (var_2200_cast_fp16_8, var_2186_cast_fp16_8))[name = tensor<string, []>("aw_257_cast_fp16")]; |
| tensor<string, []> aw_259_equation_0 = const()[name = tensor<string, []>("aw_259_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_259_cast_fp16 = einsum(equation = aw_259_equation_0, values = (var_2200_cast_fp16_9, var_2186_cast_fp16_9))[name = tensor<string, []>("aw_259_cast_fp16")]; |
| tensor<string, []> aw_261_equation_0 = const()[name = tensor<string, []>("aw_261_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_261_cast_fp16 = einsum(equation = aw_261_equation_0, values = (var_2200_cast_fp16_10, var_2186_cast_fp16_10))[name = tensor<string, []>("aw_261_cast_fp16")]; |
| tensor<string, []> aw_263_equation_0 = const()[name = tensor<string, []>("aw_263_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_263_cast_fp16 = einsum(equation = aw_263_equation_0, values = (var_2200_cast_fp16_11, var_2186_cast_fp16_11))[name = tensor<string, []>("aw_263_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2250_cast_fp16 = softmax(axis = var_2134, x = aw_241_cast_fp16)[name = tensor<string, []>("op_2250_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2251_cast_fp16 = softmax(axis = var_2134, x = aw_243_cast_fp16)[name = tensor<string, []>("op_2251_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2252_cast_fp16 = softmax(axis = var_2134, x = aw_245_cast_fp16)[name = tensor<string, []>("op_2252_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2253_cast_fp16 = softmax(axis = var_2134, x = aw_247_cast_fp16)[name = tensor<string, []>("op_2253_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2254_cast_fp16 = softmax(axis = var_2134, x = aw_249_cast_fp16)[name = tensor<string, []>("op_2254_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2255_cast_fp16 = softmax(axis = var_2134, x = aw_251_cast_fp16)[name = tensor<string, []>("op_2255_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2256_cast_fp16 = softmax(axis = var_2134, x = aw_253_cast_fp16)[name = tensor<string, []>("op_2256_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2257_cast_fp16 = softmax(axis = var_2134, x = aw_255_cast_fp16)[name = tensor<string, []>("op_2257_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2258_cast_fp16 = softmax(axis = var_2134, x = aw_257_cast_fp16)[name = tensor<string, []>("op_2258_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2259_cast_fp16 = softmax(axis = var_2134, x = aw_259_cast_fp16)[name = tensor<string, []>("op_2259_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2260_cast_fp16 = softmax(axis = var_2134, x = aw_261_cast_fp16)[name = tensor<string, []>("op_2260_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2261_cast_fp16 = softmax(axis = var_2134, x = aw_263_cast_fp16)[name = tensor<string, []>("op_2261_cast_fp16")]; |
| tensor<string, []> var_2263_equation_0 = const()[name = tensor<string, []>("op_2263_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2263_cast_fp16 = einsum(equation = var_2263_equation_0, values = (var_2213_cast_fp16_0, var_2250_cast_fp16))[name = tensor<string, []>("op_2263_cast_fp16")]; |
| tensor<string, []> var_2265_equation_0 = const()[name = tensor<string, []>("op_2265_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2265_cast_fp16 = einsum(equation = var_2265_equation_0, values = (var_2213_cast_fp16_1, var_2251_cast_fp16))[name = tensor<string, []>("op_2265_cast_fp16")]; |
| tensor<string, []> var_2267_equation_0 = const()[name = tensor<string, []>("op_2267_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2267_cast_fp16 = einsum(equation = var_2267_equation_0, values = (var_2213_cast_fp16_2, var_2252_cast_fp16))[name = tensor<string, []>("op_2267_cast_fp16")]; |
| tensor<string, []> var_2269_equation_0 = const()[name = tensor<string, []>("op_2269_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2269_cast_fp16 = einsum(equation = var_2269_equation_0, values = (var_2213_cast_fp16_3, var_2253_cast_fp16))[name = tensor<string, []>("op_2269_cast_fp16")]; |
| tensor<string, []> var_2271_equation_0 = const()[name = tensor<string, []>("op_2271_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2271_cast_fp16 = einsum(equation = var_2271_equation_0, values = (var_2213_cast_fp16_4, var_2254_cast_fp16))[name = tensor<string, []>("op_2271_cast_fp16")]; |
| tensor<string, []> var_2273_equation_0 = const()[name = tensor<string, []>("op_2273_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2273_cast_fp16 = einsum(equation = var_2273_equation_0, values = (var_2213_cast_fp16_5, var_2255_cast_fp16))[name = tensor<string, []>("op_2273_cast_fp16")]; |
| tensor<string, []> var_2275_equation_0 = const()[name = tensor<string, []>("op_2275_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2275_cast_fp16 = einsum(equation = var_2275_equation_0, values = (var_2213_cast_fp16_6, var_2256_cast_fp16))[name = tensor<string, []>("op_2275_cast_fp16")]; |
| tensor<string, []> var_2277_equation_0 = const()[name = tensor<string, []>("op_2277_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2277_cast_fp16 = einsum(equation = var_2277_equation_0, values = (var_2213_cast_fp16_7, var_2257_cast_fp16))[name = tensor<string, []>("op_2277_cast_fp16")]; |
| tensor<string, []> var_2279_equation_0 = const()[name = tensor<string, []>("op_2279_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2279_cast_fp16 = einsum(equation = var_2279_equation_0, values = (var_2213_cast_fp16_8, var_2258_cast_fp16))[name = tensor<string, []>("op_2279_cast_fp16")]; |
| tensor<string, []> var_2281_equation_0 = const()[name = tensor<string, []>("op_2281_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2281_cast_fp16 = einsum(equation = var_2281_equation_0, values = (var_2213_cast_fp16_9, var_2259_cast_fp16))[name = tensor<string, []>("op_2281_cast_fp16")]; |
| tensor<string, []> var_2283_equation_0 = const()[name = tensor<string, []>("op_2283_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2283_cast_fp16 = einsum(equation = var_2283_equation_0, values = (var_2213_cast_fp16_10, var_2260_cast_fp16))[name = tensor<string, []>("op_2283_cast_fp16")]; |
| tensor<string, []> var_2285_equation_0 = const()[name = tensor<string, []>("op_2285_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2285_cast_fp16 = einsum(equation = var_2285_equation_0, values = (var_2213_cast_fp16_11, var_2261_cast_fp16))[name = tensor<string, []>("op_2285_cast_fp16")]; |
| tensor<bool, []> input_105_interleave_0 = const()[name = tensor<string, []>("input_105_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_105_cast_fp16 = concat(axis = var_2134, interleave = input_105_interleave_0, values = (var_2263_cast_fp16, var_2265_cast_fp16, var_2267_cast_fp16, var_2269_cast_fp16, var_2271_cast_fp16, var_2273_cast_fp16, var_2275_cast_fp16, var_2277_cast_fp16, var_2279_cast_fp16, var_2281_cast_fp16, var_2283_cast_fp16, var_2285_cast_fp16))[name = tensor<string, []>("input_105_cast_fp16")]; |
| tensor<string, []> var_2294_pad_type_0 = const()[name = tensor<string, []>("op_2294_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_2294_strides_0 = const()[name = tensor<string, []>("op_2294_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_2294_pad_0 = const()[name = tensor<string, []>("op_2294_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_2294_dilations_0 = const()[name = tensor<string, []>("op_2294_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_2294_groups_0 = const()[name = tensor<string, []>("op_2294_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_10_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151512256)))]; |
| tensor<fp16, [768]> blocks_10_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152691968)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2294_cast_fp16 = conv(bias = blocks_10_attn_out_bias_to_fp16, dilations = var_2294_dilations_0, groups = var_2294_groups_0, pad = var_2294_pad_0, pad_type = var_2294_pad_type_0, strides = var_2294_strides_0, weight = blocks_10_attn_out_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("op_2294_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = var_2294_cast_fp16)[name = tensor<string, []>("inputs_43_cast_fp16")]; |
| tensor<int32, [1]> input_107_axes_0 = const()[name = tensor<string, []>("input_107_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_107_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_107_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152693568)))]; |
| tensor<fp16, [768]> input_107_beta_0_to_fp16 = const()[name = tensor<string, []>("input_107_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152695168)))]; |
| tensor<fp16, []> var_2304_to_fp16 = const()[name = tensor<string, []>("op_2304_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_107_cast_fp16 = layer_norm(axes = input_107_axes_0, beta = input_107_beta_0_to_fp16, epsilon = var_2304_to_fp16, gamma = input_107_gamma_0_to_fp16, x = inputs_43_cast_fp16)[name = tensor<string, []>("input_107_cast_fp16")]; |
| tensor<string, []> input_109_pad_type_0 = const()[name = tensor<string, []>("input_109_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_109_strides_0 = const()[name = tensor<string, []>("input_109_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_109_pad_0 = const()[name = tensor<string, []>("input_109_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_109_dilations_0 = const()[name = tensor<string, []>("input_109_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_109_groups_0 = const()[name = tensor<string, []>("input_109_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_10_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152696768)))]; |
| tensor<fp16, [3072]> blocks_10_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157415424)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_109_cast_fp16 = conv(bias = blocks_10_mlp_0_bias_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = blocks_10_mlp_0_weight_to_fp16, x = input_107_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")]; |
| tensor<string, []> input_111_mode_0 = const()[name = tensor<string, []>("input_111_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_111_cast_fp16 = gelu(mode = input_111_mode_0, x = input_109_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")]; |
| tensor<string, []> var_2330_pad_type_0 = const()[name = tensor<string, []>("op_2330_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_2330_strides_0 = const()[name = tensor<string, []>("op_2330_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_2330_pad_0 = const()[name = tensor<string, []>("op_2330_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_2330_dilations_0 = const()[name = tensor<string, []>("op_2330_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_2330_groups_0 = const()[name = tensor<string, []>("op_2330_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_10_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157421632)))]; |
| tensor<fp16, [768]> blocks_10_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162140288)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2330_cast_fp16 = conv(bias = blocks_10_mlp_2_bias_to_fp16, dilations = var_2330_dilations_0, groups = var_2330_groups_0, pad = var_2330_pad_0, pad_type = var_2330_pad_type_0, strides = var_2330_strides_0, weight = blocks_10_mlp_2_weight_to_fp16, x = input_111_cast_fp16)[name = tensor<string, []>("op_2330_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = var_2330_cast_fp16)[name = tensor<string, []>("inputs_45_cast_fp16")]; |
| tensor<int32, []> var_2339 = const()[name = tensor<string, []>("op_2339"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> input_113_axes_0 = const()[name = tensor<string, []>("input_113_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_113_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_113_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162141888)))]; |
| tensor<fp16, [768]> input_113_beta_0_to_fp16 = const()[name = tensor<string, []>("input_113_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162143488)))]; |
| tensor<fp16, []> var_2355_to_fp16 = const()[name = tensor<string, []>("op_2355_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_113_cast_fp16 = layer_norm(axes = input_113_axes_0, beta = input_113_beta_0_to_fp16, epsilon = var_2355_to_fp16, gamma = input_113_gamma_0_to_fp16, x = inputs_45_cast_fp16)[name = tensor<string, []>("input_113_cast_fp16")]; |
| tensor<string, []> q_pad_type_0 = const()[name = tensor<string, []>("q_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> q_strides_0 = const()[name = tensor<string, []>("q_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> q_pad_0 = const()[name = tensor<string, []>("q_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> q_dilations_0 = const()[name = tensor<string, []>("q_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> q_groups_0 = const()[name = tensor<string, []>("q_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> var_2390_weight_0_to_fp16 = const()[name = tensor<string, []>("op_2390_weight_0_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162145088)))]; |
| tensor<fp16, [768]> var_2390_bias_0_to_fp16 = const()[name = tensor<string, []>("op_2390_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163324800)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2390_cast_fp16 = conv(bias = var_2390_bias_0_to_fp16, dilations = q_dilations_0, groups = q_groups_0, pad = q_pad_0, pad_type = q_pad_type_0, strides = q_strides_0, weight = var_2390_weight_0_to_fp16, x = input_113_cast_fp16)[name = tensor<string, []>("op_2390_cast_fp16")]; |
| tensor<string, []> k_pad_type_0 = const()[name = tensor<string, []>("k_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> k_strides_0 = const()[name = tensor<string, []>("k_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> k_pad_0 = const()[name = tensor<string, []>("k_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> k_dilations_0 = const()[name = tensor<string, []>("k_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> k_groups_0 = const()[name = tensor<string, []>("k_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_11_attn_key_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_key_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163326400)))]; |
| tensor<fp16, [1, 768, 1, 1500]> k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = blocks_11_attn_key_weight_to_fp16, x = input_113_cast_fp16)[name = tensor<string, []>("k_cast_fp16")]; |
| tensor<string, []> var_2388_pad_type_0 = const()[name = tensor<string, []>("op_2388_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_2388_strides_0 = const()[name = tensor<string, []>("op_2388_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_2388_pad_0 = const()[name = tensor<string, []>("op_2388_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_2388_dilations_0 = const()[name = tensor<string, []>("op_2388_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_2388_groups_0 = const()[name = tensor<string, []>("op_2388_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_11_attn_value_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_value_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164506112)))]; |
| tensor<fp16, [768]> blocks_11_attn_value_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_value_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165685824)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2388_cast_fp16 = conv(bias = blocks_11_attn_value_bias_to_fp16, dilations = var_2388_dilations_0, groups = var_2388_groups_0, pad = var_2388_pad_0, pad_type = var_2388_pad_type_0, strides = var_2388_strides_0, weight = blocks_11_attn_value_weight_to_fp16, x = input_113_cast_fp16)[name = tensor<string, []>("op_2388_cast_fp16")]; |
| tensor<int32, [12]> tile_33 = const()[name = tensor<string, []>("tile_33"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_2391_axis_0 = const()[name = tensor<string, []>("op_2391_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_2391_cast_fp16_11 = split(axis = var_2391_axis_0, split_sizes = tile_33, x = var_2390_cast_fp16)[name = tensor<string, []>("op_2391_cast_fp16")]; |
| tensor<int32, [4]> var_2404_perm_0 = const()[name = tensor<string, []>("op_2404_perm_0"), val = tensor<int32, [4]>([0, 3, 2, 1])]; |
| tensor<int32, [12]> tile_34 = const()[name = tensor<string, []>("tile_34"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_2405_axis_0 = const()[name = tensor<string, []>("op_2405_axis_0"), val = tensor<int32, []>(3)]; |
| tensor<fp16, [1, 1500, 1, 768]> var_2404_cast_fp16 = transpose(perm = var_2404_perm_0, x = k_cast_fp16)[name = tensor<string, []>("transpose_1")]; |
| tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_0, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_1, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_2, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_3, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_4, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_5, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_6, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_7, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_8, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_9, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_10, tensor<fp16, [1, 1500, 1, 64]> var_2405_cast_fp16_11 = split(axis = var_2405_axis_0, split_sizes = tile_34, x = var_2404_cast_fp16)[name = tensor<string, []>("op_2405_cast_fp16")]; |
| tensor<int32, [12]> tile_35 = const()[name = tensor<string, []>("tile_35"), val = tensor<int32, [12]>([64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64])]; |
| tensor<int32, []> var_2418_axis_0 = const()[name = tensor<string, []>("op_2418_axis_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_0, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_1, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_2, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_3, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_4, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_5, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_6, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_7, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_8, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_9, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_10, tensor<fp16, [1, 64, 1, 1500]> var_2418_cast_fp16_11 = split(axis = var_2418_axis_0, split_sizes = tile_35, x = var_2388_cast_fp16)[name = tensor<string, []>("op_2418_cast_fp16")]; |
| tensor<string, []> aw_265_equation_0 = const()[name = tensor<string, []>("aw_265_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_265_cast_fp16 = einsum(equation = aw_265_equation_0, values = (var_2405_cast_fp16_0, var_2391_cast_fp16_0))[name = tensor<string, []>("aw_265_cast_fp16")]; |
| tensor<string, []> aw_267_equation_0 = const()[name = tensor<string, []>("aw_267_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_267_cast_fp16 = einsum(equation = aw_267_equation_0, values = (var_2405_cast_fp16_1, var_2391_cast_fp16_1))[name = tensor<string, []>("aw_267_cast_fp16")]; |
| tensor<string, []> aw_269_equation_0 = const()[name = tensor<string, []>("aw_269_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_269_cast_fp16 = einsum(equation = aw_269_equation_0, values = (var_2405_cast_fp16_2, var_2391_cast_fp16_2))[name = tensor<string, []>("aw_269_cast_fp16")]; |
| tensor<string, []> aw_271_equation_0 = const()[name = tensor<string, []>("aw_271_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_271_cast_fp16 = einsum(equation = aw_271_equation_0, values = (var_2405_cast_fp16_3, var_2391_cast_fp16_3))[name = tensor<string, []>("aw_271_cast_fp16")]; |
| tensor<string, []> aw_273_equation_0 = const()[name = tensor<string, []>("aw_273_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_273_cast_fp16 = einsum(equation = aw_273_equation_0, values = (var_2405_cast_fp16_4, var_2391_cast_fp16_4))[name = tensor<string, []>("aw_273_cast_fp16")]; |
| tensor<string, []> aw_275_equation_0 = const()[name = tensor<string, []>("aw_275_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_275_cast_fp16 = einsum(equation = aw_275_equation_0, values = (var_2405_cast_fp16_5, var_2391_cast_fp16_5))[name = tensor<string, []>("aw_275_cast_fp16")]; |
| tensor<string, []> aw_277_equation_0 = const()[name = tensor<string, []>("aw_277_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_277_cast_fp16 = einsum(equation = aw_277_equation_0, values = (var_2405_cast_fp16_6, var_2391_cast_fp16_6))[name = tensor<string, []>("aw_277_cast_fp16")]; |
| tensor<string, []> aw_279_equation_0 = const()[name = tensor<string, []>("aw_279_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_279_cast_fp16 = einsum(equation = aw_279_equation_0, values = (var_2405_cast_fp16_7, var_2391_cast_fp16_7))[name = tensor<string, []>("aw_279_cast_fp16")]; |
| tensor<string, []> aw_281_equation_0 = const()[name = tensor<string, []>("aw_281_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_281_cast_fp16 = einsum(equation = aw_281_equation_0, values = (var_2405_cast_fp16_8, var_2391_cast_fp16_8))[name = tensor<string, []>("aw_281_cast_fp16")]; |
| tensor<string, []> aw_283_equation_0 = const()[name = tensor<string, []>("aw_283_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_283_cast_fp16 = einsum(equation = aw_283_equation_0, values = (var_2405_cast_fp16_9, var_2391_cast_fp16_9))[name = tensor<string, []>("aw_283_cast_fp16")]; |
| tensor<string, []> aw_285_equation_0 = const()[name = tensor<string, []>("aw_285_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_285_cast_fp16 = einsum(equation = aw_285_equation_0, values = (var_2405_cast_fp16_10, var_2391_cast_fp16_10))[name = tensor<string, []>("aw_285_cast_fp16")]; |
| tensor<string, []> aw_equation_0 = const()[name = tensor<string, []>("aw_equation_0"), val = tensor<string, []>("bkhc,bchq->bkhq")]; |
| tensor<fp16, [1, 1500, 1, 1500]> aw_cast_fp16 = einsum(equation = aw_equation_0, values = (var_2405_cast_fp16_11, var_2391_cast_fp16_11))[name = tensor<string, []>("aw_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2455_cast_fp16 = softmax(axis = var_2339, x = aw_265_cast_fp16)[name = tensor<string, []>("op_2455_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2456_cast_fp16 = softmax(axis = var_2339, x = aw_267_cast_fp16)[name = tensor<string, []>("op_2456_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2457_cast_fp16 = softmax(axis = var_2339, x = aw_269_cast_fp16)[name = tensor<string, []>("op_2457_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2458_cast_fp16 = softmax(axis = var_2339, x = aw_271_cast_fp16)[name = tensor<string, []>("op_2458_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2459_cast_fp16 = softmax(axis = var_2339, x = aw_273_cast_fp16)[name = tensor<string, []>("op_2459_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2460_cast_fp16 = softmax(axis = var_2339, x = aw_275_cast_fp16)[name = tensor<string, []>("op_2460_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2461_cast_fp16 = softmax(axis = var_2339, x = aw_277_cast_fp16)[name = tensor<string, []>("op_2461_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2462_cast_fp16 = softmax(axis = var_2339, x = aw_279_cast_fp16)[name = tensor<string, []>("op_2462_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2463_cast_fp16 = softmax(axis = var_2339, x = aw_281_cast_fp16)[name = tensor<string, []>("op_2463_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2464_cast_fp16 = softmax(axis = var_2339, x = aw_283_cast_fp16)[name = tensor<string, []>("op_2464_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2465_cast_fp16 = softmax(axis = var_2339, x = aw_285_cast_fp16)[name = tensor<string, []>("op_2465_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1, 1500]> var_2466_cast_fp16 = softmax(axis = var_2339, x = aw_cast_fp16)[name = tensor<string, []>("op_2466_cast_fp16")]; |
| tensor<string, []> var_2468_equation_0 = const()[name = tensor<string, []>("op_2468_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2468_cast_fp16 = einsum(equation = var_2468_equation_0, values = (var_2418_cast_fp16_0, var_2455_cast_fp16))[name = tensor<string, []>("op_2468_cast_fp16")]; |
| tensor<string, []> var_2470_equation_0 = const()[name = tensor<string, []>("op_2470_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2470_cast_fp16 = einsum(equation = var_2470_equation_0, values = (var_2418_cast_fp16_1, var_2456_cast_fp16))[name = tensor<string, []>("op_2470_cast_fp16")]; |
| tensor<string, []> var_2472_equation_0 = const()[name = tensor<string, []>("op_2472_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2472_cast_fp16 = einsum(equation = var_2472_equation_0, values = (var_2418_cast_fp16_2, var_2457_cast_fp16))[name = tensor<string, []>("op_2472_cast_fp16")]; |
| tensor<string, []> var_2474_equation_0 = const()[name = tensor<string, []>("op_2474_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2474_cast_fp16 = einsum(equation = var_2474_equation_0, values = (var_2418_cast_fp16_3, var_2458_cast_fp16))[name = tensor<string, []>("op_2474_cast_fp16")]; |
| tensor<string, []> var_2476_equation_0 = const()[name = tensor<string, []>("op_2476_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2476_cast_fp16 = einsum(equation = var_2476_equation_0, values = (var_2418_cast_fp16_4, var_2459_cast_fp16))[name = tensor<string, []>("op_2476_cast_fp16")]; |
| tensor<string, []> var_2478_equation_0 = const()[name = tensor<string, []>("op_2478_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2478_cast_fp16 = einsum(equation = var_2478_equation_0, values = (var_2418_cast_fp16_5, var_2460_cast_fp16))[name = tensor<string, []>("op_2478_cast_fp16")]; |
| tensor<string, []> var_2480_equation_0 = const()[name = tensor<string, []>("op_2480_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2480_cast_fp16 = einsum(equation = var_2480_equation_0, values = (var_2418_cast_fp16_6, var_2461_cast_fp16))[name = tensor<string, []>("op_2480_cast_fp16")]; |
| tensor<string, []> var_2482_equation_0 = const()[name = tensor<string, []>("op_2482_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2482_cast_fp16 = einsum(equation = var_2482_equation_0, values = (var_2418_cast_fp16_7, var_2462_cast_fp16))[name = tensor<string, []>("op_2482_cast_fp16")]; |
| tensor<string, []> var_2484_equation_0 = const()[name = tensor<string, []>("op_2484_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2484_cast_fp16 = einsum(equation = var_2484_equation_0, values = (var_2418_cast_fp16_8, var_2463_cast_fp16))[name = tensor<string, []>("op_2484_cast_fp16")]; |
| tensor<string, []> var_2486_equation_0 = const()[name = tensor<string, []>("op_2486_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2486_cast_fp16 = einsum(equation = var_2486_equation_0, values = (var_2418_cast_fp16_9, var_2464_cast_fp16))[name = tensor<string, []>("op_2486_cast_fp16")]; |
| tensor<string, []> var_2488_equation_0 = const()[name = tensor<string, []>("op_2488_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2488_cast_fp16 = einsum(equation = var_2488_equation_0, values = (var_2418_cast_fp16_10, var_2465_cast_fp16))[name = tensor<string, []>("op_2488_cast_fp16")]; |
| tensor<string, []> var_2490_equation_0 = const()[name = tensor<string, []>("op_2490_equation_0"), val = tensor<string, []>("bchk,bkhq->bchq")]; |
| tensor<fp16, [1, 64, 1, 1500]> var_2490_cast_fp16 = einsum(equation = var_2490_equation_0, values = (var_2418_cast_fp16_11, var_2466_cast_fp16))[name = tensor<string, []>("op_2490_cast_fp16")]; |
| tensor<bool, []> input_115_interleave_0 = const()[name = tensor<string, []>("input_115_interleave_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_115_cast_fp16 = concat(axis = var_2339, interleave = input_115_interleave_0, values = (var_2468_cast_fp16, var_2470_cast_fp16, var_2472_cast_fp16, var_2474_cast_fp16, var_2476_cast_fp16, var_2478_cast_fp16, var_2480_cast_fp16, var_2482_cast_fp16, var_2484_cast_fp16, var_2486_cast_fp16, var_2488_cast_fp16, var_2490_cast_fp16))[name = tensor<string, []>("input_115_cast_fp16")]; |
| tensor<string, []> var_2499_pad_type_0 = const()[name = tensor<string, []>("op_2499_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_2499_strides_0 = const()[name = tensor<string, []>("op_2499_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_2499_pad_0 = const()[name = tensor<string, []>("op_2499_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_2499_dilations_0 = const()[name = tensor<string, []>("op_2499_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_2499_groups_0 = const()[name = tensor<string, []>("op_2499_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 768, 1, 1]> blocks_11_attn_out_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_out_weight_to_fp16"), val = tensor<fp16, [768, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165687424)))]; |
| tensor<fp16, [768]> blocks_11_attn_out_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_out_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166867136)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2499_cast_fp16 = conv(bias = blocks_11_attn_out_bias_to_fp16, dilations = var_2499_dilations_0, groups = var_2499_groups_0, pad = var_2499_pad_0, pad_type = var_2499_pad_type_0, strides = var_2499_strides_0, weight = blocks_11_attn_out_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("op_2499_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = var_2499_cast_fp16)[name = tensor<string, []>("inputs_47_cast_fp16")]; |
| tensor<int32, [1]> input_117_axes_0 = const()[name = tensor<string, []>("input_117_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> input_117_gamma_0_to_fp16 = const()[name = tensor<string, []>("input_117_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166868736)))]; |
| tensor<fp16, [768]> input_117_beta_0_to_fp16 = const()[name = tensor<string, []>("input_117_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166870336)))]; |
| tensor<fp16, []> var_2509_to_fp16 = const()[name = tensor<string, []>("op_2509_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> input_117_cast_fp16 = layer_norm(axes = input_117_axes_0, beta = input_117_beta_0_to_fp16, epsilon = var_2509_to_fp16, gamma = input_117_gamma_0_to_fp16, x = inputs_47_cast_fp16)[name = tensor<string, []>("input_117_cast_fp16")]; |
| tensor<string, []> input_119_pad_type_0 = const()[name = tensor<string, []>("input_119_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> input_119_strides_0 = const()[name = tensor<string, []>("input_119_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> input_119_pad_0 = const()[name = tensor<string, []>("input_119_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> input_119_dilations_0 = const()[name = tensor<string, []>("input_119_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> input_119_groups_0 = const()[name = tensor<string, []>("input_119_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [3072, 768, 1, 1]> blocks_11_mlp_0_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_0_weight_to_fp16"), val = tensor<fp16, [3072, 768, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166871936)))]; |
| tensor<fp16, [3072]> blocks_11_mlp_0_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_0_bias_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171590592)))]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_119_cast_fp16 = conv(bias = blocks_11_mlp_0_bias_to_fp16, dilations = input_119_dilations_0, groups = input_119_groups_0, pad = input_119_pad_0, pad_type = input_119_pad_type_0, strides = input_119_strides_0, weight = blocks_11_mlp_0_weight_to_fp16, x = input_117_cast_fp16)[name = tensor<string, []>("input_119_cast_fp16")]; |
| tensor<string, []> input_mode_0 = const()[name = tensor<string, []>("input_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 3072, 1, 1500]> input_cast_fp16 = gelu(mode = input_mode_0, x = input_119_cast_fp16)[name = tensor<string, []>("input_cast_fp16")]; |
| tensor<string, []> var_2535_pad_type_0 = const()[name = tensor<string, []>("op_2535_pad_type_0"), val = tensor<string, []>("valid")]; |
| tensor<int32, [2]> var_2535_strides_0 = const()[name = tensor<string, []>("op_2535_strides_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [4]> var_2535_pad_0 = const()[name = tensor<string, []>("op_2535_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])]; |
| tensor<int32, [2]> var_2535_dilations_0 = const()[name = tensor<string, []>("op_2535_dilations_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, []> var_2535_groups_0 = const()[name = tensor<string, []>("op_2535_groups_0"), val = tensor<int32, []>(1)]; |
| tensor<fp16, [768, 3072, 1, 1]> blocks_11_mlp_2_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_2_weight_to_fp16"), val = tensor<fp16, [768, 3072, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171596800)))]; |
| tensor<fp16, [768]> blocks_11_mlp_2_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_2_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176315456)))]; |
| tensor<fp16, [1, 768, 1, 1500]> var_2535_cast_fp16 = conv(bias = blocks_11_mlp_2_bias_to_fp16, dilations = var_2535_dilations_0, groups = var_2535_groups_0, pad = var_2535_pad_0, pad_type = var_2535_pad_type_0, strides = var_2535_strides_0, weight = blocks_11_mlp_2_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("op_2535_cast_fp16")]; |
| tensor<fp16, [1, 768, 1, 1500]> inputs_cast_fp16 = add(x = inputs_47_cast_fp16, y = var_2535_cast_fp16)[name = tensor<string, []>("inputs_cast_fp16")]; |
| tensor<int32, [1]> x_axes_0 = const()[name = tensor<string, []>("x_axes_0"), val = tensor<int32, [1]>([1])]; |
| tensor<fp16, [768]> x_gamma_0_to_fp16 = const()[name = tensor<string, []>("x_gamma_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176317056)))]; |
| tensor<fp16, [768]> x_beta_0_to_fp16 = const()[name = tensor<string, []>("x_beta_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176318656)))]; |
| tensor<fp16, []> var_2549_to_fp16 = const()[name = tensor<string, []>("op_2549_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 768, 1, 1500]> x_cast_fp16 = layer_norm(axes = x_axes_0, beta = x_beta_0_to_fp16, epsilon = var_2549_to_fp16, gamma = x_gamma_0_to_fp16, x = inputs_cast_fp16)[name = tensor<string, []>("x_cast_fp16")]; |
| tensor<int32, [1]> var_2560_axes_0 = const()[name = tensor<string, []>("op_2560_axes_0"), val = tensor<int32, [1]>([2])]; |
| tensor<fp16, [1, 768, 1500]> var_2560_cast_fp16 = squeeze(axes = var_2560_axes_0, x = x_cast_fp16)[name = tensor<string, []>("op_2560_cast_fp16")]; |
| tensor<int32, [3]> var_2563_perm_0 = const()[name = tensor<string, []>("op_2563_perm_0"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<string, []> var_2563_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_2563_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp16, [1, 1500, 768]> var_2563_cast_fp16 = transpose(perm = var_2563_perm_0, x = var_2560_cast_fp16)[name = tensor<string, []>("transpose_0")]; |
| tensor<fp32, [1, 1500, 768]> output = cast(dtype = var_2563_cast_fp16_to_fp32_dtype_0, x = var_2563_cast_fp16)[name = tensor<string, []>("cast_51")]; |
| } -> (output); |
| } |
